{
 "file": "gpu-ql3090-24g-nemo-q8-0-control-3090-q8_0-ladder.json",
 "model_tag": "mistral-nemo:12b-instruct-2407-q8_0",
 "arm": "control-3090",
 "box_class": "gpu-ql3090-24g",
 "quantization": "Q8_0",
 "store_gib": 12.13,
 "largest_whole": 98304,
 "topped_out": false,
 "stopped_at": {
  "num_ctx": 131072,
  "why": "the first rung that did not fit whole on the card",
  "verdict": "does not fit: only part of the model is on the card (92.1% VRAM / 7.9% RAM)",
  "memory_used_mib_per_card": {
   "<withheld: GPU UUID of board A>": 22790.0
  }
 },
 "errors": [],
 "ups_before": 12.0,
 "ups_after": 26.0,
 "windows": [
  {
   "num_ctx": 4096,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 12503.0
   },
   "runs": 3,
   "decode_median": 60.545,
   "ttft_ms_median": 548.22,
   "board_w_median": 259.9,
   "j_per_1k_median": 4293.87,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 54.0,
   "spread": 0.0005,
   "decode_range": 0.033,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.545
  },
  {
   "num_ctx": 8192,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 470.0
   },
   "runs": 3,
   "decode_median": 60.54,
   "ttft_ms_median": 477.88,
   "board_w_median": 273.71,
   "j_per_1k_median": 4521.13,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 60.0,
   "spread": 0.0014,
   "decode_range": 0.084,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.54
  },
  {
   "num_ctx": 16384,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 696.0
   },
   "runs": 3,
   "decode_median": 60.434,
   "ttft_ms_median": 509.83,
   "board_w_median": 273.48,
   "j_per_1k_median": 4527.2,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0014,
   "decode_range": 0.082,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.434
  },
  {
   "num_ctx": 32768,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1294.0
   },
   "runs": 3,
   "decode_median": 60.41,
   "ttft_ms_median": 488.39,
   "board_w_median": 274.21,
   "j_per_1k_median": 4539.13,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0006,
   "decode_range": 0.035,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.41
  },
  {
   "num_ctx": 49152,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1440.0
   },
   "runs": 3,
   "decode_median": 60.423,
   "ttft_ms_median": 506.09,
   "board_w_median": 273.89,
   "j_per_1k_median": 4532.36,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0008,
   "decode_range": 0.051,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.423
  },
  {
   "num_ctx": 65536,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1440.0
   },
   "runs": 3,
   "decode_median": 60.31,
   "ttft_ms_median": 439.68,
   "board_w_median": 275.48,
   "j_per_1k_median": 4563.76,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 65.0,
   "spread": 0.0015,
   "decode_range": 0.093,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.31
  },
  {
   "num_ctx": 98304,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 2880.0
   },
   "runs": 3,
   "decode_median": 60.394,
   "ttft_ms_median": 475.14,
   "board_w_median": 273.73,
   "j_per_1k_median": 4529.11,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 63.0,
   "spread": 0.0012,
   "decode_range": 0.073,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 60.394
  },
  {
   "num_ctx": 131072,
   "whole": false,
   "verdict": "does not fit: only part of the model is on the card (92.1% VRAM / 7.9% RAM)",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 2066.0
   },
   "runs": 0,
   "decode_median": null,
   "ttft_ms_median": null,
   "board_w_median": null,
   "j_per_1k_median": null,
   "sw_power_cap_fraction_median": null,
   "temp_max_c": null,
   "spread": null,
   "decode_range": null,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [
    "not whole: not tested"
   ],
   "passes": false,
   "headline_decode": null
  }
 ],
 "resident_named_card_growth_mib": 12503.0,
 "resident": true,
 "passing_windows": [
  4096,
  8192,
  16384,
  32768,
  49152,
  65536,
  98304
 ]
}