{
  "publication": "llm-cpu-load",
  "study": {
    "schema": 1,
    "kind": "gpu_resource_study",
    "study_id": "20260926T100955Z-a48ec372",
    "ok": true,
    "failure_class": null,
    "record_sha256": "f2c758479956c700db6a4f33b1530a6731620a6601396484ba738eb366447b94",
    "hardware": {
      "gpu": "NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition",
      "gpu_reported_memory_mib": 97887.0,
      "gpu_power_limit_w": 300.0,
      "driver": "595.91.07",
      "cpu": [
        "AMD EPYC 4565P 16-Core Processor"
      ],
      "logical_cpus": 32,
      "physical_system_ram_gb": 96,
      "linux_visible_ram_bytes": 100271353856,
      "kernel": "6.12.0-211.56.1.el10_2.0.1.x86_64",
      "os": "Rocky Linux 10.2"
    },
    "tool": {
      "name": "llama.cpp",
      "version": "0.5.0-dev",
      "release": "b11146",
      "commit": "7fe450e19305b828c199d602c23a8337aaa1f03b"
    },
    "model": {
      "name": "Qwen3-8B",
      "repository": "Qwen/Qwen3-8B-GGUF",
      "revision": "7c41481f57cb95916b40956ab2f0b139b296d974",
      "license": "apache-2.0",
      "quantization": "Q4_K_M",
      "expected_total_bytes": 5027783488,
      "expected_hashes": [
        "d98cdcbd03e17ce47681435b5150e34c1417f50b5c0019dd560e4882c5745785"
      ]
    },
    "settings": {
      "cases": [
        {
          "concurrency": 1,
          "context": 2048
        },
        {
          "concurrency": 4,
          "context": 8192
        }
      ],
      "repetitions": 3,
      "generated_tokens": 256,
      "sample_interval_seconds": 2,
      "baseline_idle_seconds": 10,
      "ready_idle_seconds": 10,
      "threads": 16,
      "logical_cpus": 32,
      "clock_ticks_per_second": 100,
      "server_slots": 4,
      "context_per_slot": 8704,
      "total_context_tokens": 34816,
      "kv_cache": "f16",
      "flash_attention": "on",
      "gpu_layers": 999,
      "fit": "off",
      "batch": 2048,
      "ubatch": 512,
      "prompt_cache": false,
      "network_scope": "loopback-only",
      "cooldown_before_each_case_max_c": 40,
      "temperature_abort_c": 85,
      "minimum_thermal_margin_c": 7,
      "wall_time_limit_seconds": 1800,
      "cooldown_settle_seconds": 3
    },
    "definitions": {
      "process_percent_one_logical_cpu": "100% means one logical CPU fully occupied. Multithreaded process values can exceed 100%. Process user+system ticks / CLK_TCK / measured monotonic interval.",
      "process_percent_host_capacity": "Process one-logical-CPU percentage divided by 32 logical CPUs; 100% means the whole host CPU capacity.",
      "host_busy_percent": "Delta total CPU ticks excluding guest double-counting; busy excludes idle and iowait. Includes unrelated host activity and monitoring/client overhead.",
      "memory": "RSS/PSS are point-in-time process observations; host Cached and reclaimable memory can overlap file-backed process RSS/PSS. Never add them. PSS is proportional attribution, not minimum required RAM.",
      "power": "NVIDIA sampled GPU board watts only, not CPU/package/whole-server AC power or energy. Sample means and maxima can miss short peaks.",
      "phase_sampling": "Nominal 2-second sampling. Cross-phase CPU intervals are excluded. Fewer than three stable phase samples are explicitly marked sparse. Per-round CPU uses separate start/end /proc counters.",
      "case_cpu_mean": "Arithmetic mean of the three measured per-round CPU percentages/seconds. Phase point statistics are separate and exclude warmup/cooldown phases.",
      "cache": "Model SHA read precedes baseline/load, so OS cache is warm. Other cached files can remain from earlier work. No cache dropping, cold-start or storage-throughput claim.",
      "comparison": "Additional /proc/PSS telemetry adds overhead: timing here is not directly comparable with the previous throughput benchmark campaign."
    },
    "phases": [
      {
        "phase": "baseline_idle_warm_cache",
        "started_at": "2026-09-26T10:09:55.814666+00:00",
        "finished_at": "2026-09-26T10:10:05.815053+00:00",
        "duration_seconds": 10.000369853019947,
        "sample_count": 5,
        "cpu_interval_count": 4,
        "sparse_under_three_samples": false,
        "metrics": {
          "gpu_board_power_w": {
            "n": 5,
            "minimum": 14.3,
            "maximum": 15.15,
            "mean": 14.55
          },
          "gpu_memory_used_mib": {
            "n": 5,
            "minimum": 2.0,
            "maximum": 2.0,
            "mean": 2.0
          },
          "gpu_temperature_c": {
            "n": 5,
            "minimum": 30.0,
            "maximum": 30.0,
            "mean": 30.0
          },
          "gpu_thermal_margin_c": {
            "n": 5,
            "minimum": 62.0,
            "maximum": 62.0,
            "mean": 62.0
          },
          "gpu_utilization_percent": {
            "n": 5,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "host_mem_total_bytes": {
            "n": 5,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 5,
            "minimum": 20271435776,
            "maximum": 20277530624,
            "mean": 20274212044.8
          },
          "host_mem_available_bytes": {
            "n": 5,
            "minimum": 96396267520,
            "maximum": 96402272256,
            "mean": 96399013478.4
          },
          "host_cached_bytes": {
            "n": 5,
            "minimum": 74722340864,
            "maximum": 74722435072,
            "mean": 74722395750.4
          },
          "host_buffers_bytes": {
            "n": 5,
            "minimum": 190226432,
            "maximum": 190238720,
            "mean": 190233804.8
          },
          "host_sreclaimable_bytes": {
            "n": 5,
            "minimum": 2149826560,
            "maximum": 2149830656,
            "mean": 2149829017.6
          },
          "host_shmem_bytes": {
            "n": 5,
            "minimum": 21094400,
            "maximum": 21094400,
            "mean": 21094400
          },
          "host_swap_cached_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_cpu_busy_percent": {
            "n": 4,
            "minimum": 0.03125,
            "maximum": 0.062490235900640526,
            "mean": 0.0507696574193992
          },
          "host_swap_used_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 5,
            "minimum": 22.775641991756856,
            "maximum": 27.451472997199744,
            "mean": 26.30348799866624
          }
        }
      },
      {
        "phase": "model_loading",
        "started_at": "2026-09-26T10:10:05.815102+00:00",
        "finished_at": "2026-09-26T10:10:07.108872+00:00",
        "duration_seconds": 1.2938315520004835,
        "sample_count": 1,
        "cpu_interval_count": 0,
        "sparse_under_three_samples": true,
        "metrics": {
          "gpu_board_power_w": {
            "n": 1,
            "minimum": 14.58,
            "maximum": 14.58,
            "mean": 14.58
          },
          "gpu_memory_used_mib": {
            "n": 1,
            "minimum": 2.0,
            "maximum": 2.0,
            "mean": 2.0
          },
          "gpu_temperature_c": {
            "n": 1,
            "minimum": 30.0,
            "maximum": 30.0,
            "mean": 30.0
          },
          "gpu_thermal_margin_c": {
            "n": 1,
            "minimum": 62.0,
            "maximum": 62.0,
            "mean": 62.0
          },
          "gpu_utilization_percent": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_rss_bytes": {
            "n": 1,
            "minimum": 1949696,
            "maximum": 1949696,
            "mean": 1949696
          },
          "process_pss_bytes": {
            "n": 1,
            "minimum": 1345536,
            "maximum": 1345536,
            "mean": 1345536
          },
          "process_pss_anon_bytes": {
            "n": 1,
            "minimum": 184320,
            "maximum": 184320,
            "mean": 184320
          },
          "process_pss_file_bytes": {
            "n": 1,
            "minimum": 1161216,
            "maximum": 1161216,
            "mean": 1161216
          },
          "process_pss_shmem_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 1,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 1,
            "minimum": 20271996928,
            "maximum": 20271996928,
            "mean": 20271996928
          },
          "host_mem_available_bytes": {
            "n": 1,
            "minimum": 96396857344,
            "maximum": 96396857344,
            "mean": 96396857344
          },
          "host_cached_bytes": {
            "n": 1,
            "minimum": 74722435072,
            "maximum": 74722435072,
            "mean": 74722435072
          },
          "host_buffers_bytes": {
            "n": 1,
            "minimum": 190246912,
            "maximum": 190246912,
            "mean": 190246912
          },
          "host_sreclaimable_bytes": {
            "n": 1,
            "minimum": 2149830656,
            "maximum": 2149830656,
            "mean": 2149830656
          },
          "host_shmem_bytes": {
            "n": 1,
            "minimum": 21094400,
            "maximum": 21094400,
            "mean": 21094400
          },
          "host_swap_cached_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_swap_used_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 1,
            "minimum": 25.631655007600784,
            "maximum": 25.631655007600784,
            "mean": 25.631655007600784
          }
        }
      },
      {
        "phase": "model_ready_idle",
        "started_at": "2026-09-26T10:10:07.108885+00:00",
        "finished_at": "2026-09-26T10:10:17.115216+00:00",
        "duration_seconds": 10.006343746004859,
        "sample_count": 5,
        "cpu_interval_count": 4,
        "sparse_under_three_samples": false,
        "metrics": {
          "gpu_board_power_w": {
            "n": 5,
            "minimum": 66.6,
            "maximum": 74.3,
            "mean": 68.55
          },
          "gpu_memory_used_mib": {
            "n": 5,
            "minimum": 10094.0,
            "maximum": 10094.0,
            "mean": 10094.0
          },
          "gpu_temperature_c": {
            "n": 5,
            "minimum": 31.0,
            "maximum": 32.0,
            "mean": 31.6
          },
          "gpu_thermal_margin_c": {
            "n": 5,
            "minimum": 60.0,
            "maximum": 61.0,
            "mean": 60.4
          },
          "gpu_utilization_percent": {
            "n": 5,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_rss_bytes": {
            "n": 5,
            "minimum": 789393408,
            "maximum": 789393408,
            "mean": 789393408
          },
          "process_pss_bytes": {
            "n": 5,
            "minimum": 786450432,
            "maximum": 786450432,
            "mean": 786450432
          },
          "process_pss_anon_bytes": {
            "n": 5,
            "minimum": 239915008,
            "maximum": 239915008,
            "mean": 239915008
          },
          "process_pss_file_bytes": {
            "n": 5,
            "minimum": 507910144,
            "maximum": 507910144,
            "mean": 507910144
          },
          "process_pss_shmem_bytes": {
            "n": 5,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 5,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 5,
            "minimum": 19955904512,
            "maximum": 19986944000,
            "mean": 19971932979.2
          },
          "host_mem_available_bytes": {
            "n": 5,
            "minimum": 96004739072,
            "maximum": 96035745792,
            "mean": 96020754432
          },
          "host_cached_bytes": {
            "n": 5,
            "minimum": 74761068544,
            "maximum": 74761080832,
            "mean": 74761077555.2
          },
          "host_buffers_bytes": {
            "n": 5,
            "minimum": 190246912,
            "maximum": 190255104,
            "mean": 190251827.2
          },
          "host_sreclaimable_bytes": {
            "n": 5,
            "minimum": 2149847040,
            "maximum": 2149847040,
            "mean": 2149847040
          },
          "host_shmem_bytes": {
            "n": 5,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_cpu_busy_percent": {
            "n": 4,
            "minimum": 0.031245117950320263,
            "maximum": 0.0625,
            "mean": 0.046867677688062354
          },
          "process_cpu_percent_one_logical_cpu": {
            "n": 4,
            "minimum": 0.0,
            "maximum": 0.4998851481340106,
            "mean": 0.12497128703350265
          },
          "process_cpu_percent_host_capacity": {
            "n": 4,
            "minimum": 0.0,
            "maximum": 0.01562141087918783,
            "mean": 0.0039053527197969577
          },
          "host_swap_used_bytes": {
            "n": 5,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 5,
            "minimum": 19.527422002283856,
            "maximum": 21.64345100754872,
            "mean": 20.43187140370719
          }
        }
      },
      {
        "phase": "2048t-1c-cooldown",
        "started_at": "2026-09-26T10:10:17.115240+00:00",
        "finished_at": "2026-09-26T10:10:20.146065+00:00",
        "duration_seconds": 3.0308504899730906,
        "sample_count": 2,
        "cpu_interval_count": 1,
        "sparse_under_three_samples": true,
        "metrics": {
          "gpu_board_power_w": {
            "n": 2,
            "minimum": 67.14,
            "maximum": 67.39,
            "mean": 67.265
          },
          "gpu_memory_used_mib": {
            "n": 2,
            "minimum": 10094.0,
            "maximum": 10094.0,
            "mean": 10094.0
          },
          "gpu_temperature_c": {
            "n": 2,
            "minimum": 32.0,
            "maximum": 33.0,
            "mean": 32.5
          },
          "gpu_thermal_margin_c": {
            "n": 2,
            "minimum": 59.0,
            "maximum": 60.0,
            "mean": 59.5
          },
          "gpu_utilization_percent": {
            "n": 2,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_rss_bytes": {
            "n": 2,
            "minimum": 789442560,
            "maximum": 789442560,
            "mean": 789442560
          },
          "process_pss_bytes": {
            "n": 2,
            "minimum": 786499584,
            "maximum": 786499584,
            "mean": 786499584
          },
          "process_pss_anon_bytes": {
            "n": 2,
            "minimum": 239964160,
            "maximum": 239964160,
            "mean": 239964160
          },
          "process_pss_file_bytes": {
            "n": 2,
            "minimum": 507910144,
            "maximum": 507910144,
            "mean": 507910144
          },
          "process_pss_shmem_bytes": {
            "n": 2,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 2,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 2,
            "minimum": 19946815488,
            "maximum": 19955933184,
            "mean": 19951374336
          },
          "host_mem_available_bytes": {
            "n": 2,
            "minimum": 95995650048,
            "maximum": 96004767744,
            "mean": 96000208896
          },
          "host_cached_bytes": {
            "n": 2,
            "minimum": 74761089024,
            "maximum": 74761089024,
            "mean": 74761089024
          },
          "host_buffers_bytes": {
            "n": 2,
            "minimum": 190267392,
            "maximum": 190267392,
            "mean": 190267392
          },
          "host_sreclaimable_bytes": {
            "n": 2,
            "minimum": 2149847040,
            "maximum": 2149847040,
            "mean": 2149847040
          },
          "host_shmem_bytes": {
            "n": 2,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_cpu_busy_percent": {
            "n": 1,
            "minimum": 0.062490235900640526,
            "maximum": 0.062490235900640526,
            "mean": 0.062490235900640526
          },
          "process_cpu_percent_one_logical_cpu": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_cpu_percent_host_capacity": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "host_swap_used_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 2,
            "minimum": 21.04338599019684,
            "maximum": 21.41049801139161,
            "mean": 21.226942000794224
          }
        }
      },
      {
        "phase": "2048t-1c-warmup",
        "started_at": "2026-09-26T10:10:20.146083+00:00",
        "finished_at": "2026-09-26T10:10:21.630469+00:00",
        "duration_seconds": 1.484408024000004,
        "sample_count": 0,
        "cpu_interval_count": 0,
        "sparse_under_three_samples": true,
        "metrics": {}
      },
      {
        "phase": "2048t-1c-measured",
        "started_at": "2026-09-26T10:10:21.630476+00:00",
        "finished_at": "2026-09-26T10:10:26.096319+00:00",
        "duration_seconds": 4.465849998028716,
        "sample_count": 3,
        "cpu_interval_count": 0,
        "sparse_under_three_samples": false,
        "metrics": {
          "gpu_board_power_w": {
            "n": 3,
            "minimum": 299.93,
            "maximum": 300.24,
            "mean": 300.0366666666667
          },
          "gpu_memory_used_mib": {
            "n": 3,
            "minimum": 10112.0,
            "maximum": 10112.0,
            "mean": 10112.0
          },
          "gpu_temperature_c": {
            "n": 3,
            "minimum": 40.0,
            "maximum": 42.0,
            "mean": 40.666666666666664
          },
          "gpu_thermal_margin_c": {
            "n": 3,
            "minimum": 50.0,
            "maximum": 52.0,
            "mean": 51.0
          },
          "gpu_utilization_percent": {
            "n": 3,
            "minimum": 97.0,
            "maximum": 100.0,
            "mean": 98.0
          },
          "process_rss_bytes": {
            "n": 3,
            "minimum": 820756480,
            "maximum": 823386112,
            "mean": 822386688
          },
          "process_pss_bytes": {
            "n": 3,
            "minimum": 817813504,
            "maximum": 820443136,
            "mean": 819443712
          },
          "process_pss_anon_bytes": {
            "n": 3,
            "minimum": 266559488,
            "maximum": 269189120,
            "mean": 268189696
          },
          "process_pss_file_bytes": {
            "n": 3,
            "minimum": 512628736,
            "maximum": 512628736,
            "mean": 512628736
          },
          "process_pss_shmem_bytes": {
            "n": 3,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 3,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 3,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 3,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 3,
            "minimum": 19934502912,
            "maximum": 19942019072,
            "mean": 19937696426.666668
          },
          "host_mem_available_bytes": {
            "n": 3,
            "minimum": 95983390720,
            "maximum": 95990890496,
            "mean": 95986623829.33333
          },
          "host_cached_bytes": {
            "n": 3,
            "minimum": 74761138176,
            "maximum": 74761265152,
            "mean": 74761200981.33333
          },
          "host_buffers_bytes": {
            "n": 3,
            "minimum": 190267392,
            "maximum": 190279680,
            "mean": 190275584
          },
          "host_sreclaimable_bytes": {
            "n": 3,
            "minimum": 2149847040,
            "maximum": 2149847040,
            "mean": 2149847040
          },
          "host_shmem_bytes": {
            "n": 3,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 3,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_swap_used_bytes": {
            "n": 3,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 3,
            "minimum": 22.787594993133098,
            "maximum": 23.263879003934562,
            "mean": 23.040284004916128
          }
        }
      },
      {
        "phase": "8192t-4c-cooldown",
        "started_at": "2026-09-26T10:10:26.096322+00:00",
        "finished_at": "2026-09-26T10:10:29.124753+00:00",
        "duration_seconds": 3.0284301059728023,
        "sample_count": 1,
        "cpu_interval_count": 0,
        "sparse_under_three_samples": true,
        "metrics": {
          "gpu_board_power_w": {
            "n": 1,
            "minimum": 78.67,
            "maximum": 78.67,
            "mean": 78.67
          },
          "gpu_memory_used_mib": {
            "n": 1,
            "minimum": 10112.0,
            "maximum": 10112.0,
            "mean": 10112.0
          },
          "gpu_temperature_c": {
            "n": 1,
            "minimum": 38.0,
            "maximum": 38.0,
            "mean": 38.0
          },
          "gpu_thermal_margin_c": {
            "n": 1,
            "minimum": 54.0,
            "maximum": 54.0,
            "mean": 54.0
          },
          "gpu_utilization_percent": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_rss_bytes": {
            "n": 1,
            "minimum": 823455744,
            "maximum": 823455744,
            "mean": 823455744
          },
          "process_pss_bytes": {
            "n": 1,
            "minimum": 820512768,
            "maximum": 820512768,
            "mean": 820512768
          },
          "process_pss_anon_bytes": {
            "n": 1,
            "minimum": 269258752,
            "maximum": 269258752,
            "mean": 269258752
          },
          "process_pss_file_bytes": {
            "n": 1,
            "minimum": 512628736,
            "maximum": 512628736,
            "mean": 512628736
          },
          "process_pss_shmem_bytes": {
            "n": 1,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 1,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 1,
            "minimum": 19932971008,
            "maximum": 19932971008,
            "mean": 19932971008
          },
          "host_mem_available_bytes": {
            "n": 1,
            "minimum": 95982018560,
            "maximum": 95982018560,
            "mean": 95982018560
          },
          "host_cached_bytes": {
            "n": 1,
            "minimum": 74761281536,
            "maximum": 74761281536,
            "mean": 74761281536
          },
          "host_buffers_bytes": {
            "n": 1,
            "minimum": 190287872,
            "maximum": 190287872,
            "mean": 190287872
          },
          "host_sreclaimable_bytes": {
            "n": 1,
            "minimum": 2149847040,
            "maximum": 2149847040,
            "mean": 2149847040
          },
          "host_shmem_bytes": {
            "n": 1,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_swap_used_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 1,
            "minimum": 21.650292997946963,
            "maximum": 21.650292997946963,
            "mean": 21.650292997946963
          }
        }
      },
      {
        "phase": "8192t-4c-warmup",
        "started_at": "2026-09-26T10:10:29.124771+00:00",
        "finished_at": "2026-09-26T10:10:31.454506+00:00",
        "duration_seconds": 2.3297550620045513,
        "sample_count": 1,
        "cpu_interval_count": 0,
        "sparse_under_three_samples": true,
        "metrics": {
          "gpu_board_power_w": {
            "n": 1,
            "minimum": 159.81,
            "maximum": 159.81,
            "mean": 159.81
          },
          "gpu_memory_used_mib": {
            "n": 1,
            "minimum": 10112.0,
            "maximum": 10112.0,
            "mean": 10112.0
          },
          "gpu_temperature_c": {
            "n": 1,
            "minimum": 44.0,
            "maximum": 44.0,
            "mean": 44.0
          },
          "gpu_thermal_margin_c": {
            "n": 1,
            "minimum": 48.0,
            "maximum": 48.0,
            "mean": 48.0
          },
          "gpu_utilization_percent": {
            "n": 1,
            "minimum": 100.0,
            "maximum": 100.0,
            "mean": 100.0
          },
          "process_rss_bytes": {
            "n": 1,
            "minimum": 824164352,
            "maximum": 824164352,
            "mean": 824164352
          },
          "process_pss_bytes": {
            "n": 1,
            "minimum": 821221376,
            "maximum": 821221376,
            "mean": 821221376
          },
          "process_pss_anon_bytes": {
            "n": 1,
            "minimum": 269967360,
            "maximum": 269967360,
            "mean": 269967360
          },
          "process_pss_file_bytes": {
            "n": 1,
            "minimum": 512628736,
            "maximum": 512628736,
            "mean": 512628736
          },
          "process_pss_shmem_bytes": {
            "n": 1,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 1,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 1,
            "minimum": 19930443776,
            "maximum": 19930443776,
            "mean": 19930443776
          },
          "host_mem_available_bytes": {
            "n": 1,
            "minimum": 95979495424,
            "maximum": 95979495424,
            "mean": 95979495424
          },
          "host_cached_bytes": {
            "n": 1,
            "minimum": 74761289728,
            "maximum": 74761289728,
            "mean": 74761289728
          },
          "host_buffers_bytes": {
            "n": 1,
            "minimum": 190287872,
            "maximum": 190287872,
            "mean": 190287872
          },
          "host_sreclaimable_bytes": {
            "n": 1,
            "minimum": 2149847040,
            "maximum": 2149847040,
            "mean": 2149847040
          },
          "host_shmem_bytes": {
            "n": 1,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_swap_used_bytes": {
            "n": 1,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 1,
            "minimum": 21.38663298683241,
            "maximum": 21.38663298683241,
            "mean": 21.38663298683241
          }
        }
      },
      {
        "phase": "8192t-4c-measured",
        "started_at": "2026-09-26T10:10:31.454518+00:00",
        "finished_at": "2026-09-26T10:10:50.913900+00:00",
        "duration_seconds": 19.45939229099895,
        "sample_count": 10,
        "cpu_interval_count": 7,
        "sparse_under_three_samples": false,
        "metrics": {
          "gpu_board_power_w": {
            "n": 10,
            "minimum": 299.77,
            "maximum": 300.33,
            "mean": 299.969
          },
          "gpu_memory_used_mib": {
            "n": 10,
            "minimum": 10112.0,
            "maximum": 10114.0,
            "mean": 10113.6
          },
          "gpu_temperature_c": {
            "n": 10,
            "minimum": 46.0,
            "maximum": 56.0,
            "mean": 50.8
          },
          "gpu_thermal_margin_c": {
            "n": 10,
            "minimum": 36.0,
            "maximum": 46.0,
            "mean": 41.2
          },
          "gpu_utilization_percent": {
            "n": 10,
            "minimum": 96.0,
            "maximum": 100.0,
            "mean": 98.2
          },
          "process_rss_bytes": {
            "n": 10,
            "minimum": 827002880,
            "maximum": 855646208,
            "mean": 844576358.4
          },
          "process_pss_bytes": {
            "n": 10,
            "minimum": 824059904,
            "maximum": 852703232,
            "mean": 841633382.4
          },
          "process_pss_anon_bytes": {
            "n": 10,
            "minimum": 272805888,
            "maximum": 299745280,
            "mean": 289527398.4
          },
          "process_pss_file_bytes": {
            "n": 10,
            "minimum": 512628736,
            "maximum": 514332672,
            "mean": 513480704
          },
          "process_pss_shmem_bytes": {
            "n": 10,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 10,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 10,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 10,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 10,
            "minimum": 19890122752,
            "maximum": 19923288064,
            "mean": 19905699840
          },
          "host_mem_available_bytes": {
            "n": 10,
            "minimum": 95939960832,
            "maximum": 95972360192,
            "mean": 95955153715.2
          },
          "host_cached_bytes": {
            "n": 10,
            "minimum": 74761375744,
            "maximum": 74762108928,
            "mean": 74761701376
          },
          "host_buffers_bytes": {
            "n": 10,
            "minimum": 190287872,
            "maximum": 190316544,
            "mean": 190304256
          },
          "host_sreclaimable_bytes": {
            "n": 10,
            "minimum": 2149847040,
            "maximum": 2149855232,
            "mean": 2149851955.2
          },
          "host_shmem_bytes": {
            "n": 10,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 10,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_cpu_busy_percent": {
            "n": 7,
            "minimum": 3.15625,
            "maximum": 3.219253008282544,
            "mean": 3.1812306019686556
          },
          "process_cpu_percent_one_logical_cpu": {
            "n": 7,
            "minimum": 99.49638216266679,
            "maximum": 100.99623506331014,
            "mean": 100.28192711850411
          },
          "process_cpu_percent_host_capacity": {
            "n": 7,
            "minimum": 3.109261942583337,
            "maximum": 3.156132345728442,
            "mean": 3.1338102224532536
          },
          "host_swap_used_bytes": {
            "n": 10,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 10,
            "minimum": 21.51426099590026,
            "maximum": 26.885983010288328,
            "mean": 23.685753301833756
          }
        }
      },
      {
        "phase": "final_ready_idle",
        "started_at": "2026-09-26T10:10:50.913912+00:00",
        "finished_at": "2026-09-26T10:10:54.914384+00:00",
        "duration_seconds": 4.0004805520002265,
        "sample_count": 2,
        "cpu_interval_count": 1,
        "sparse_under_three_samples": true,
        "metrics": {
          "gpu_board_power_w": {
            "n": 2,
            "minimum": 77.62,
            "maximum": 111.05,
            "mean": 94.33500000000001
          },
          "gpu_memory_used_mib": {
            "n": 2,
            "minimum": 10114.0,
            "maximum": 10114.0,
            "mean": 10114.0
          },
          "gpu_temperature_c": {
            "n": 2,
            "minimum": 51.0,
            "maximum": 52.0,
            "mean": 51.5
          },
          "gpu_thermal_margin_c": {
            "n": 2,
            "minimum": 40.0,
            "maximum": 41.0,
            "mean": 40.5
          },
          "gpu_utilization_percent": {
            "n": 2,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_rss_bytes": {
            "n": 2,
            "minimum": 855674880,
            "maximum": 855674880,
            "mean": 855674880
          },
          "process_pss_bytes": {
            "n": 2,
            "minimum": 852731904,
            "maximum": 852731904,
            "mean": 852731904
          },
          "process_pss_anon_bytes": {
            "n": 2,
            "minimum": 299773952,
            "maximum": 299773952,
            "mean": 299773952
          },
          "process_pss_file_bytes": {
            "n": 2,
            "minimum": 514332672,
            "maximum": 514332672,
            "mean": 514332672
          },
          "process_pss_shmem_bytes": {
            "n": 2,
            "minimum": 38625280,
            "maximum": 38625280,
            "mean": 38625280
          },
          "process_swap_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "process_swap_pss_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_mem_total_bytes": {
            "n": 2,
            "minimum": 100271353856,
            "maximum": 100271353856,
            "mean": 100271353856
          },
          "host_mem_free_bytes": {
            "n": 2,
            "minimum": 19892224000,
            "maximum": 19893010432,
            "mean": 19892617216
          },
          "host_mem_available_bytes": {
            "n": 2,
            "minimum": 95942295552,
            "maximum": 95943110656,
            "mean": 95942703104
          },
          "host_cached_bytes": {
            "n": 2,
            "minimum": 74762354688,
            "maximum": 74762358784,
            "mean": 74762356736
          },
          "host_buffers_bytes": {
            "n": 2,
            "minimum": 190316544,
            "maximum": 190324736,
            "mean": 190320640
          },
          "host_sreclaimable_bytes": {
            "n": 2,
            "minimum": 2149859328,
            "maximum": 2149859328,
            "mean": 2149859328
          },
          "host_shmem_bytes": {
            "n": 2,
            "minimum": 59719680,
            "maximum": 59719680,
            "mean": 59719680
          },
          "host_swap_cached_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "host_cpu_busy_percent": {
            "n": 1,
            "minimum": 0.062490235900640526,
            "maximum": 0.062490235900640526,
            "mean": 0.062490235900640526
          },
          "process_cpu_percent_one_logical_cpu": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "process_cpu_percent_host_capacity": {
            "n": 1,
            "minimum": 0.0,
            "maximum": 0.0,
            "mean": 0.0
          },
          "host_swap_used_bytes": {
            "n": 2,
            "minimum": 0,
            "maximum": 0,
            "mean": 0
          },
          "sampling_cost_ms": {
            "n": 2,
            "minimum": 21.43468300346285,
            "maximum": 22.267869993811473,
            "mean": 21.851276498637162
          }
        }
      }
    ],
    "cases": [
      {
        "concurrency": 1,
        "context": 2048,
        "completed_rounds": 3,
        "rounds": [
          {
            "round": 0,
            "started_at": "2026-09-26T10:10:21.630642+00:00",
            "finished_at": "2026-09-26T10:10:23.119131+00:00",
            "duration_seconds": 1.488484821980819,
            "interval_seconds": 1.488484821980819,
            "host_busy_percent": 3.1709365812683745,
            "process_cpu_seconds": 1.48,
            "process_percent_one_logical_cpu": 99.42996919716468,
            "process_percent_host_capacity": 3.1071865374113963
          },
          {
            "round": 1,
            "started_at": "2026-09-26T10:10:23.119416+00:00",
            "finished_at": "2026-09-26T10:10:24.607263+00:00",
            "duration_seconds": 1.4878047079837415,
            "interval_seconds": 1.4878047079837415,
            "host_busy_percent": 3.2129357412851745,
            "process_cpu_seconds": 1.49,
            "process_percent_one_logical_cpu": 100.14755243107366,
            "process_percent_host_capacity": 3.129611013471052
          },
          {
            "round": 2,
            "started_at": "2026-09-26T10:10:24.607459+00:00",
            "finished_at": "2026-09-26T10:10:26.096105+00:00",
            "duration_seconds": 1.488574326998787,
            "interval_seconds": 1.488574326998787,
            "host_busy_percent": 3.191266008817972,
            "process_cpu_seconds": 1.48,
            "process_percent_one_logical_cpu": 99.42399067058517,
            "process_percent_host_capacity": 3.1069997084557865
          }
        ],
        "cpu_mean": {
          "interval_seconds": 1.4882879523211159,
          "host_busy_percent": 3.1917127771238403,
          "process_cpu_seconds": 1.4833333333333334,
          "process_percent_one_logical_cpu": 99.66717076627451,
          "process_percent_host_capacity": 3.1145990864460784
        },
        "sampled_phase": "2048t-1c-measured"
      },
      {
        "concurrency": 4,
        "context": 8192,
        "completed_rounds": 3,
        "rounds": [
          {
            "round": 0,
            "started_at": "2026-09-26T10:10:31.454690+00:00",
            "finished_at": "2026-09-26T10:10:37.921320+00:00",
            "duration_seconds": 6.4666263250110205,
            "interval_seconds": 6.4666263250110205,
            "host_busy_percent": 3.1993040788710614,
            "process_cpu_seconds": 6.48,
            "process_percent_one_logical_cpu": 100.206810697214,
            "process_percent_host_capacity": 3.1314628342879374
          },
          {
            "round": 1,
            "started_at": "2026-09-26T10:10:37.921644+00:00",
            "finished_at": "2026-09-26T10:10:44.409101+00:00",
            "duration_seconds": 6.4873855040059425,
            "interval_seconds": 6.4873855040059425,
            "host_busy_percent": 3.1846213143187514,
            "process_cpu_seconds": 6.5,
            "process_percent_one_logical_cpu": 100.19444652990435,
            "process_percent_host_capacity": 3.131076454059511
          },
          {
            "round": 2,
            "started_at": "2026-09-26T10:10:44.409428+00:00",
            "finished_at": "2026-09-26T10:10:50.913598+00:00",
            "duration_seconds": 6.504104042018298,
            "interval_seconds": 6.504104042018298,
            "host_busy_percent": 3.1997693859901988,
            "process_cpu_seconds": 6.53,
            "process_percent_one_logical_cpu": 100.39814796649019,
            "process_percent_host_capacity": 3.1374421239528183
          }
        ],
        "cpu_mean": {
          "interval_seconds": 6.486038623678421,
          "host_busy_percent": 3.1945649263933373,
          "process_cpu_seconds": 6.503333333333334,
          "process_percent_one_logical_cpu": 100.26646839786952,
          "process_percent_host_capacity": 3.1333271374334224
        },
        "sampled_phase": "8192t-4c-measured"
      }
    ],
    "limitations": [
      "Only the actually measured 96 GB host configuration is represented.",
      "Short Qwen3-8B Q4_K_M local synthetic requests; no generic model-quality or production-capacity claim.",
      "PSS/RSS, host memory and GPU telemetry are sampled observations, not proof of exact peaks or minimum RAM.",
      "No CPU-only, CPU limiting, RAM limiting, OS A/B, driver A/B or whole-server power comparison was performed."
    ],
    "startup": {
      "warm_cache_ready_seconds": 1.2815848170139361,
      "health_poll_interval_seconds": 0.25,
      "health_http_status": 200
    }
  },
  "table": [
    {
      "context_tokens": 2048,
      "concurrent_requests": 1,
      "measured_rounds": 3,
      "interval_seconds": 1.4882879523211159,
      "host_busy_percent": 3.1917127771238403,
      "process_cpu_seconds": 1.4833333333333334,
      "process_percent_one_logical_cpu": 99.66717076627451,
      "process_percent_host_capacity": 3.1145990864460784,
      "process_percent_one_logical_cpu_minimum": 99.42399067058517,
      "process_percent_one_logical_cpu_maximum": 100.14755243107366
    },
    {
      "context_tokens": 8192,
      "concurrent_requests": 4,
      "measured_rounds": 3,
      "interval_seconds": 6.486038623678421,
      "host_busy_percent": 3.1945649263933373,
      "process_cpu_seconds": 6.503333333333334,
      "process_percent_one_logical_cpu": 100.26646839786952,
      "process_percent_host_capacity": 3.1333271374334224,
      "process_percent_one_logical_cpu_minimum": 100.19444652990435,
      "process_percent_one_logical_cpu_maximum": 100.39814796649019
    }
  ]
}
