{
  "scope": "Single-seed heavy-chat comparison, Nodus resident shim versus native MPS",
  "date_pdt": "2026-10-29",
  "model": "Qwen/Qwen3-1.7B-FP8",
  "gpu": "NVIDIA A100-SXM4-40GB",
  "settings": {
    "tenants": 2,
    "engine_gib_each": 18,
    "vllm_target_gib_each": 16.2,
    "max_num_batched_tokens": 8192,
    "max_num_seqs": 64,
    "offered_window_s": 120,
    "actual_input_tokens": [
      1023,
      1024,
      1025,
      1026
    ],
    "actual_output_tokens": 256,
    "ttft_target_s": 1,
    "tpot_target_s": 0.1,
    "attainment_target": 0.99
  },
  "rates": {
    "12.0": {
      "native_mps_100": {
        "offered": 1440,
        "completed": 1440,
        "within_latency_targets_percent": 86.96,
        "ttft_p50_s": 0.0741,
        "ttft_p95_s": 0.08,
        "ttft_p99_s": 0.0812,
        "tpot_p99_s": 0.01551,
        "completed_output_tok_s": 2627.58,
        "slo_qualified_output_tok_s": 2627.58,
        "gpu": {
          "kernel_busy_mean_percent": 86.26,
          "memory_busy_mean_percent": 64.09,
          "sm_active_mean_percent": 78.78,
          "tensor_active_mean_percent": 23.57,
          "dram_active_mean_percent": 49.48
        }
      },
      "nodus_resident_100": {
        "offered": 1440,
        "completed": 1440,
        "within_latency_targets_percent": 100,
        "ttft_p50_s": 0.062,
        "ttft_p95_s": 0.067,
        "ttft_p99_s": 0.069,
        "tpot_p99_s": 0.01318,
        "completed_output_tok_s": 3021.69,
        "slo_qualified_output_tok_s": 3021.69,
        "gpu": {
          "kernel_busy_mean_percent": 99.2,
          "memory_busy_mean_percent": 73.8,
          "sm_active_mean_percent": 91,
          "tensor_active_mean_percent": 27.3,
          "dram_active_mean_percent": 57.2
        }
      }
    },
    "16.0": {
      "native_mps_100": {
        "offered": 1920,
        "completed": 1920,
        "within_latency_targets_percent": 31.7,
        "ttft_p50_s": 1.898,
        "ttft_p95_s": 5.814,
        "ttft_p99_s": 6.104,
        "tpot_p99_s": 0.03889,
        "completed_output_tok_s": 3325.05,
        "slo_qualified_output_tok_s": 1212.26,
        "gpu": {
          "kernel_busy_mean_percent": 86.26,
          "memory_busy_mean_percent": 69.04,
          "sm_active_mean_percent": 82.78,
          "tensor_active_mean_percent": 30.09,
          "dram_active_mean_percent": 53.74
        }
      },
      "nodus_resident_100": {
        "offered": 1920,
        "completed": 1920,
        "within_latency_targets_percent": 37.14,
        "ttft_p50_s": 1.556,
        "ttft_p95_s": 4.867,
        "ttft_p99_s": 5.091,
        "tpot_p99_s": 0.03304,
        "completed_output_tok_s": 3826.58,
        "slo_qualified_output_tok_s": 1421.02,
        "gpu": {
          "kernel_busy_mean_percent": 99.2,
          "memory_busy_mean_percent": 79.3,
          "sm_active_mean_percent": 95.2,
          "tensor_active_mean_percent": 34.5,
          "dram_active_mean_percent": 61.8
        }
      }
    }
  }
}