{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":54.84,"tok_s_decode":58.09,"vram_used_gb":8.014,"power_w":150.082,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":51.19,"tok_s_decode":56.96,"vram_used_gb":8.068,"power_w":146.854,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":44.18,"tok_s_decode":54.19,"vram_used_gb":8.291,"power_w":145.876,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":27.73,"tok_s_decode":42.24,"vram_used_gb":9.359,"power_w":121.551,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":54.81,"tok_s_decode":57.31,"vram_used_gb":8.67,"power_w":188.319,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":52.08,"tok_s_decode":56.54,"vram_used_gb":8.764,"power_w":183.555,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":44.74,"tok_s_decode":53.7,"vram_used_gb":9.15,"power_w":188.536,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-12b-ggml-q4km","params_b":12,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":28.07,"tok_s_decode":42.04,"vram_used_gb":11.053,"power_w":159.081,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-06","notes":"Seeded from llm-lab card-bench gemma4-12b-ggml-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":13064.82,"tok_s_decode":120.63,"vram_used_gb":21.262,"power_w":140.535,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":12605.53,"tok_s_decode":118.49,"vram_used_gb":21.934,"power_w":146.558,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":9334.83,"tok_s_decode":110.79,"vram_used_gb":22.355,"power_w":149.598,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":4470.77,"tok_s_decode":103.3,"vram_used_gb":22.357,"power_w":149.786,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":11943.84,"tok_s_decode":182.58,"vram_used_gb":43.984,"power_w":169.035,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":12379.46,"tok_s_decode":174.03,"vram_used_gb":44.34,"power_w":206.528,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":10850.93,"tok_s_decode":154.39,"vram_used_gb":44.598,"power_w":246.071,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-aeon-nvfp4","params_b":26,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":6685.85,"tok_s_decode":149.28,"vram_used_gb":44.531,"power_w":280.518,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-aeon-nvfp4-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":94.87,"tok_s_decode":121.76,"vram_used_gb":16.613,"power_w":130.124,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":87.0,"tok_s_decode":118.36,"vram_used_gb":16.682,"power_w":126.528,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":66.11,"tok_s_decode":99.04,"vram_used_gb":17.039,"power_w":127.894,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":34.35,"tok_s_decode":62.13,"vram_used_gb":18.504,"power_w":127.859,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":96.44,"tok_s_decode":124.78,"vram_used_gb":17.166,"power_w":173.613,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":88.8,"tok_s_decode":121.34,"vram_used_gb":17.273,"power_w":172.334,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":68.78,"tok_s_decode":102.48,"vram_used_gb":17.893,"power_w":171.012,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-26b-a4b-q4-k-m","params_b":26,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":34.48,"tok_s_decode":62.36,"vram_used_gb":20.389,"power_w":161.146,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-10","notes":"Seeded from llm-lab card-bench gemma4-26b-a4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-31b-q4-k-m","params_b":31,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":25.9,"tok_s_decode":27.33,"vram_used_gb":18.811,"power_w":149.912,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-31b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-31b-q4-k-m","params_b":31,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":26.3,"tok_s_decode":28.24,"vram_used_gb":19.635,"power_w":206.227,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-31b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-31b-q4-k-m","params_b":31,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":25.14,"tok_s_decode":27.36,"vram_used_gb":19.928,"power_w":205.479,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-31b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-31b-q8-0","params_b":31,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":16.37,"tok_s_decode":17.16,"vram_used_gb":32.963,"power_w":194.326,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-31b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-31b-q8-0","params_b":31,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":15.97,"tok_s_decode":16.83,"vram_used_gb":33.256,"power_w":195.838,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-31b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-e4b-q4-k-m","params_b":4,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":94.57,"tok_s_decode":118.57,"vram_used_gb":3.652,"power_w":126.544,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-e4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"gemma4-e4b-q4-k-m","params_b":4,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":86.32,"tok_s_decode":113.53,"vram_used_gb":3.707,"power_w":131.442,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-e4b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-e4b-q4-k-m","params_b":4,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":92.83,"tok_s_decode":115.71,"vram_used_gb":4.297,"power_w":171.542,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-e4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"gemma4-e4b-q4-k-m","params_b":4,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":85.34,"tok_s_decode":111.64,"vram_used_gb":4.396,"power_w":172.237,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench gemma4-e4b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"granite41-30b-q4-k-m","params_b":30,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":30.13,"tok_s_decode":30.75,"vram_used_gb":17.682,"power_w":148.817,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-30b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"granite41-30b-q4-k-m","params_b":30,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":27.95,"tok_s_decode":28.63,"vram_used_gb":18.484,"power_w":148.818,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-30b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"granite41-30b-q4-k-m","params_b":30,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":30.67,"tok_s_decode":31.71,"vram_used_gb":18.576,"power_w":203.93,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-30b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"granite41-30b-q4-k-m","params_b":30,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":28.24,"tok_s_decode":29.63,"vram_used_gb":19.42,"power_w":202.924,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-30b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"granite41-8b-q4-k-m","params_b":8,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":82.54,"tok_s_decode":92.88,"vram_used_gb":5.773,"power_w":148.572,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-8b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"granite41-8b-q4-k-m","params_b":8,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":75.7,"tok_s_decode":80.86,"vram_used_gb":6.277,"power_w":139.174,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-8b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"granite41-8b-q4-k-m","params_b":8,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":84.33,"tok_s_decode":95.06,"vram_used_gb":6.379,"power_w":193.811,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-8b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"granite41-8b-q4-k-m","params_b":8,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":77.67,"tok_s_decode":84.18,"vram_used_gb":6.924,"power_w":196.692,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench granite41-8b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"nex-n2-mini-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":186.99,"tok_s_decode":148.18,"vram_used_gb":20.953,"power_w":172.378,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench nex-n2-mini-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"nex-n2-mini-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":181.67,"tok_s_decode":143.54,"vram_used_gb":21.062,"power_w":176.56,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench nex-n2-mini-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":4732.34,"tok_s_decode":33.61,"vram_used_gb":23.295,"power_w":149.63,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-11","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":4796.8,"tok_s_decode":33.56,"vram_used_gb":21.512,"power_w":149.919,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-11","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":3838.19,"tok_s_decode":32.39,"vram_used_gb":21.518,"power_w":150.993,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-11","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":5323.27,"tok_s_decode":58.96,"vram_used_gb":45.285,"power_w":277.982,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-12","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":5513.98,"tok_s_decode":58.82,"vram_used_gb":45.285,"power_w":275.046,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-12","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":4848.17,"tok_s_decode":57.78,"vram_used_gb":45.293,"power_w":270.915,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-12","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-aeon-text-nvfp4-mtp-xs","params_b":27,"quant":"nvfp4","backend":"vllm","backend_version":"0.22.1-f1900e87","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":3104.49,"tok_s_decode":53.69,"vram_used_gb":45.285,"power_w":285.68,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-12","notes":"Seeded from llm-lab card-bench qwen36-27b-aeon-text-nvfp4-mtp-xs-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-dense-stock-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":62.01,"tok_s_decode":31.1,"vram_used_gb":16.096,"power_w":149.937,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwen36-27b-dense-stock-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-dense-stock-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":65.26,"tok_s_decode":32.78,"vram_used_gb":16.609,"power_w":205.533,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwen36-27b-dense-stock-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-dense-stock-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":65.98,"tok_s_decode":31.86,"vram_used_gb":16.855,"power_w":206.801,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwen36-27b-dense-stock-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-pi-tune-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":67.08,"tok_s_decode":32.5,"vram_used_gb":16.879,"power_w":205.577,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-27b-pi-tune-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":66.92,"tok_s_decode":31.22,"vram_used_gb":16.141,"power_w":149.438,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":66.02,"tok_s_decode":30.51,"vram_used_gb":16.346,"power_w":149.68,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":55.45,"tok_s_decode":28.29,"vram_used_gb":17.25,"power_w":149.166,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":39.28,"tok_s_decode":21.12,"vram_used_gb":20.906,"power_w":148.321,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":64.84,"tok_s_decode":32.46,"vram_used_gb":16.861,"power_w":206.597,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":63.35,"tok_s_decode":31.81,"vram_used_gb":17.107,"power_w":207.185,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":57.12,"tok_s_decode":29.5,"vram_used_gb":18.26,"power_w":208.383,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q4-k-m","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":40.45,"tok_s_decode":22.57,"vram_used_gb":22.947,"power_w":209.611,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":64.02,"tok_s_decode":27.53,"vram_used_gb":18.5,"power_w":150.187,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":60.35,"tok_s_decode":27.18,"vram_used_gb":18.705,"power_w":149.927,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":53.78,"tok_s_decode":25.18,"vram_used_gb":19.609,"power_w":149.663,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":38.8,"tok_s_decode":19.32,"vram_used_gb":23.266,"power_w":148.805,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":62.54,"tok_s_decode":28.63,"vram_used_gb":19.221,"power_w":206.442,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":61.5,"tok_s_decode":28.11,"vram_used_gb":19.467,"power_w":206.911,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":54.54,"tok_s_decode":26.31,"vram_used_gb":20.619,"power_w":208.565,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q5-k-m","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":39.65,"tok_s_decode":20.61,"vram_used_gb":25.307,"power_w":209.822,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":60.69,"tok_s_decode":24.29,"vram_used_gb":21.15,"power_w":149.933,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":59.91,"tok_s_decode":23.84,"vram_used_gb":21.355,"power_w":150.005,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":61.42,"tok_s_decode":25.22,"vram_used_gb":21.869,"power_w":208.977,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":60.29,"tok_s_decode":24.82,"vram_used_gb":22.115,"power_w":209.72,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":54.38,"tok_s_decode":23.41,"vram_used_gb":23.268,"power_w":210.662,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q6-k","params_b":27,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":39.75,"tok_s_decode":18.71,"vram_used_gb":27.955,"power_w":211.029,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-20","notes":"Seeded from llm-lab card-bench qwen36-27b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q8-0","params_b":27,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":57.44,"tok_s_decode":20.56,"vram_used_gb":27.238,"power_w":195.785,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q8-0","params_b":27,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":56.33,"tok_s_decode":20.3,"vram_used_gb":27.484,"power_w":197.105,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-27b-q8-0","params_b":27,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":51.36,"tok_s_decode":19.36,"vram_used_gb":28.637,"power_w":198.847,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-21","notes":"Seeded from llm-lab card-bench qwen36-27b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":190.68,"tok_s_decode":131.93,"vram_used_gb":20.932,"power_w":129.957,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":183.94,"tok_s_decode":128.38,"vram_used_gb":21.0,"power_w":127.788,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":16384,"concurrency":1,"tok_s_prefill":175.01,"tok_s_decode":122.31,"vram_used_gb":21.105,"power_w":131.545,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":20000,"concurrency":1,"tok_s_prefill":170.63,"tok_s_decode":120.78,"vram_used_gb":21.154,"power_w":131.666,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":158.7,"tok_s_decode":113.1,"vram_used_gb":21.318,"power_w":136.912,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":191.06,"tok_s_decode":135.2,"vram_used_gb":21.408,"power_w":166.278,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":183.92,"tok_s_decode":130.08,"vram_used_gb":21.518,"power_w":165.614,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":16384,"concurrency":1,"tok_s_prefill":177.3,"tok_s_decode":124.76,"vram_used_gb":21.689,"power_w":167.464,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":159.38,"tok_s_decode":116.25,"vram_used_gb":22.041,"power_w":170.65,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":98.39,"tok_s_decode":77.03,"vram_used_gb":24.162,"power_w":181.737,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-15","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q4-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q5-k-m","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":182.97,"tok_s_decode":128.49,"vram_used_gb":25.434,"power_w":165.47,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q5-k-m","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":178.37,"tok_s_decode":124.51,"vram_used_gb":25.543,"power_w":181.605,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q5-k-m","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":154.71,"tok_s_decode":111.12,"vram_used_gb":26.066,"power_w":180.086,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q5-k-m","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":98.21,"tok_s_decode":75.1,"vram_used_gb":28.188,"power_w":183.598,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q5-k-m-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q6-k","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":177.95,"tok_s_decode":125.62,"vram_used_gb":28.09,"power_w":166.017,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q6-k","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":171.19,"tok_s_decode":121.63,"vram_used_gb":28.199,"power_w":181.394,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q6-k","params_b":35,"quant":"other","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":95.98,"tok_s_decode":74.12,"vram_used_gb":30.844,"power_w":183.706,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q6-k-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q8-0","params_b":35,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":178.06,"tok_s_decode":119.41,"vram_used_gb":35.166,"power_w":159.313,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q8-0","params_b":35,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":173.8,"tok_s_decode":115.91,"vram_used_gb":35.275,"power_w":175.325,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q8-0","params_b":35,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":151.94,"tok_s_decode":105.0,"vram_used_gb":35.799,"power_w":180.171,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwen36-35b-a3b-q8-0","params_b":35,"quant":"gguf-q8_0","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":95.49,"tok_s_decode":71.84,"vram_used_gb":37.92,"power_w":179.894,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-17","notes":"Seeded from llm-lab card-bench qwen36-35b-a3b-q8-0-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"qwopus36-27b-q4km","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":61.47,"tok_s_decode":31.1,"vram_used_gb":16.34,"power_w":149.381,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwopus36-27b-q4km-1gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwopus36-27b-q4km","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":64.89,"tok_s_decode":32.79,"vram_used_gb":16.855,"power_w":206.083,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwopus36-27b-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"qwopus36-27b-q4km","params_b":27,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":65.15,"tok_s_decode":31.86,"vram_used_gb":17.102,"power_w":205.447,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-18","notes":"Seeded from llm-lab card-bench qwopus36-27b-q4km-2gpu; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant": "desktop", "vram_gb": 24, "memory_type": "gddr7", "bus_width_bit": 192, "bandwidth_gbps": 672, "tdp_w": 140, "pcie_lanes": 16, "gpu_count": 2, "model": "qwen36-35b-a3b-nvfp4-verifier", "params_b": 35, "quant": "nvfp4", "backend": "vllm", "backend_version": "0.23.0-a24a03aa", "driver_version": "610.43.02", "os": "Ubuntu 24.04.4 LTS", "context_len": 8192, "concurrency": 1, "tok_s_prefill": 14178.05, "tok_s_decode": 198.99, "vram_used_gb": 44.912, "power_w": 232.333, "thinking": "off", "source": "owner-measured", "submitter": "Andrew", "date": "2026-06-22", "notes": "vLLM concurrency bench 2026-06-22: stock NVFP4 ModelOpt verifier of Qwen3.6-35B-A3B served alone (no DFlash draft) on the 0.23.0-cu129 container, dual-GPU tensor-parallel + expert-parallel. Single-stream decode, 3-run median, client-stream timing. Same-session llama.cpp Q4_K_M baseline 130.69 tok/s decode at this context. Production stack stopped for clean measurement."}
{"variant": "desktop", "vram_gb": 24, "memory_type": "gddr7", "bus_width_bit": 192, "bandwidth_gbps": 672, "tdp_w": 140, "pcie_lanes": 16, "gpu_count": 2, "model": "qwen36-35b-a3b-nvfp4-verifier", "params_b": 35, "quant": "nvfp4", "backend": "vllm", "backend_version": "0.23.0-a24a03aa", "driver_version": "610.43.02", "os": "Ubuntu 24.04.4 LTS", "context_len": 4096, "concurrency": 4, "tok_s_prefill": null, "tok_s_decode": 391.79, "vram_used_gb": 45.12, "power_w": null, "thinking": "off", "source": "owner-measured", "submitter": "Andrew", "date": "2026-06-22", "notes": "vLLM concurrency bench 2026-06-22 (CC-PROMPT-vllm-concurrency-bench). 4-way concurrent SUSTAINED decode: tok_s_decode is the aggregate over 4 streams at 256 tok/stream forced via ignore_eos (equal-length vs llama's natural 256); EP = tensor-parallel-size 2 + expert-parallel, 0.23.0-cu129 container. NOT natural-chat throughput; single measured volley (one-shot), not a 3-run median. tok_s_prefill and power_w are null - not separately measured at concurrency 4; vram_used_gb is the served dual-card footprint at 0.90 util. Same-session llama.cpp Q4_K_M 4-way aggregate 253.44 tok/s. Re-measure script: bench/vllm_concurrent_remeasure.sh. Production stopped for clean measurement."}
{"variant": "desktop", "vram_gb": 24, "memory_type": "gddr7", "bus_width_bit": 192, "bandwidth_gbps": 672, "tdp_w": 140, "pcie_lanes": 16, "gpu_count": 2, "model": "qwen36-35b-a3b-q4-k-m", "params_b": 35, "quant": "gguf-q4_k_m", "backend": "llama.cpp", "backend_version": "94a220cd6745e6e3f8de62870b66fd5b9bc92700", "driver_version": "610.43.02", "os": "Ubuntu 24.04.4 LTS", "context_len": 4096, "concurrency": 4, "tok_s_prefill": null, "tok_s_decode": 253.44, "vram_used_gb": 22.368, "power_w": 178.307, "thinking": "off", "source": "owner-measured", "submitter": "Andrew", "date": "2026-06-22", "notes": "vLLM concurrency bench 2026-06-22: llama.cpp production baseline, 4-way concurrent SUSTAINED decode: tok_s_decode is the aggregate over 4 streams at 256 tok/stream (natural, hit the 256 cap), dual-GPU layer-split, q8_0 KV. Single measured volley (one-shot), not a 3-run median. tok_s_prefill is null - not separately measured at concurrency 4; power_w is the real 4-way concurrent measurement; vram_used_gb is the concurrent footprint. Same-session vLLM NVFP4 EP 4-way aggregate 391.79 tok/s. Production stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":193.76,"tok_s_decode":152.07,"vram_used_gb":20.27,"power_w":128.304,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-1gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":186.21,"tok_s_decode":145.67,"vram_used_gb":20.338,"power_w":132.346,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-1gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":160.79,"tok_s_decode":127.47,"vram_used_gb":20.656,"power_w":137.918,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-1gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":99.38,"tok_s_decode":79.63,"vram_used_gb":21.934,"power_w":139.528,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-1gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":195.39,"tok_s_decode":155.66,"vram_used_gb":20.748,"power_w":168.426,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-2gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":187.73,"tok_s_decode":149.5,"vram_used_gb":20.857,"power_w":176.072,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-2gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":163.43,"tok_s_decode":130.59,"vram_used_gb":21.381,"power_w":180.588,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-2gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"ornith-1.0-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":99.75,"tok_s_decode":83.03,"vram_used_gb":23.504,"power_w":188.749,"thinking":"on","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench ornith-1.0-35b-q4-k-m-2gpu; reasoning-on serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":191.12,"tok_s_decode":152.05,"vram_used_gb":20.27,"power_w":119.553,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-1gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":183.93,"tok_s_decode":145.94,"vram_used_gb":20.338,"power_w":129.156,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-1gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":157.88,"tok_s_decode":126.98,"vram_used_gb":20.656,"power_w":121.624,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-1gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":1,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":99.59,"tok_s_decode":80.85,"vram_used_gb":21.934,"power_w":134.196,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-1gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":2048,"concurrency":1,"tok_s_prefill":190.15,"tok_s_decode":156.47,"vram_used_gb":20.748,"power_w":158.476,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-2gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":8192,"concurrency":1,"tok_s_prefill":180.79,"tok_s_decode":149.69,"vram_used_gb":20.857,"power_w":174.615,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-2gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":32768,"concurrency":1,"tok_s_prefill":160.56,"tok_s_decode":131.36,"vram_used_gb":21.381,"power_w":160.318,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-2gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}
{"variant":"desktop","vram_gb":24,"memory_type":"gddr7","bus_width_bit":192,"bandwidth_gbps":672,"tdp_w":140,"pcie_lanes":16,"gpu_count":2,"model":"titus-cyber-35b-q4-k-m","params_b":35,"quant":"gguf-q4_k_m","backend":"llama.cpp","backend_version":"94a220cd6745e6e3f8de62870b66fd5b9bc92700","driver_version":"610.43.02","os":"Ubuntu 24.04.4 LTS","context_len":131072,"concurrency":1,"tok_s_prefill":97.74,"tok_s_decode":83.88,"vram_used_gb":23.502,"power_w":166.126,"thinking":"off","source":"owner-measured","submitter":"Andrew","date":"2026-06-26","notes":"Seeded from llm-lab card-bench titus-cyber-35b-q4-k-m-2gpu; reasoning-off serve; one row per context; median of kept runs; production stack stopped for clean measurement."}