{ "description": "Selected published ATOM AgentX records. Full serving configurations with mono enabled; no matched mono-off arm. Values are preserved from the InferenceX API.", "sources": { "MiniMax-M3": "https://inferencex.semianalysis.com/api/v1/benchmarks?model=MiniMax-M3", "DeepSeek-V4.1-Flash": "https://inferencex.semianalysis.com/api/v1/benchmarks?model=DeepSeek-V4.1-Flash" }, "retrieved_utc": "2026-10-19", "records": [ { "id": "444377", "hardware": "mi355x", "framework": "atom", "model": "minimaxm3", "precision": "fp4", "spec_method": "mtp", "disagg": false, "is_multinode": false, "prefill_tp": 4, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 4, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 4, "num_decode_gpu": 4, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 1, "image": "rocm/atom-dev:nightly_202610070333", "recipe_fingerprint": "45f691dbcb55d011a3ccc5a581d62b13030b64f000665e76776e111d864d68f7", "metrics": { "pp": 1, "p75_itl": 0.00208, "p75_qps": 0, "p90_itl": 0.00213, "p90_qps": 0, "p95_itl": 0.00231, "p95_qps": 1, "std_itl": 0.00022, "std_qps": 0.28275, "dcp_size": 1, "mean_itl": 0.002, "mean_qps": 0.08493, "p75_e2el": 6.35658, "p75_tpot": 0.00207, "p75_ttft": 0.96709, "p90_e2el": 11.3406, "p90_tpot": 0.00213, "p90_ttft": 1.1283, "p95_e2el": 13.45938, "p95_tpot": 0.00231, "p95_ttft": 1.18648, "pcp_size": 1, "std_e2el": 5.11384, "std_tpot": 0.00022, "std_ttft": 0.36326, "mean_e2el": 4.68393, "mean_tpot": 0.002, "mean_ttft": 0.73447, "median_itl": 0.00197, "median_qps": 0, "p75_intvty": 480.76923076923083, "p90_intvty": 469.4835680751174, "p95_intvty": 432.9004329004329, "std_intvty": 42.96503, "mean_intvty": 500, "median_e2el": 2.51522, "median_tpot": 0.00197, "median_ttft": 0.73047, "offload_mode": "off", "tput_per_gpu": 6580.16709, "kv_offloading": "none", "median_intvty": 507.61421319796955, "input_tput_tps": 26150.70967, "total_tput_tps": 26320.66837, "output_tput_tps": 169.9587, "duration_seconds": 3595.1146, "p75_input_tokens": 445575.5, "p90_input_tokens": 556134, "p95_input_tokens": 585127.75, "std_input_tokens": 179752.86802, "mean_input_tokens": 307237.90196, "input_tput_per_gpu": 6537.67742, "median_input_tokens": 340339, "output_tput_per_gpu": 42.48967, "total_prompt_tokens": 97593318, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00208, "p90_full_response_itl": 0.00213, "p95_full_response_itl": 0.00231, "std_full_response_itl": 0.00022, "gpu_kv_cache_usage_pct": 0.07866, "mean_full_response_itl": 0.002, "total_generation_tokens": 611032, "median_full_response_itl": 0.00197, "p75_full_response_intvty": 481.71762, "p75_output_tokens_actual": 2771, "p90_full_response_intvty": 468.3981, "p90_output_tokens_actual": 5082.5, "p95_full_response_intvty": 433.51395, "p95_output_tokens_actual": 6834.25, "std_full_response_intvty": 42.96503, "std_output_tokens_actual": 2553.86492, "total_requests_completed": 306, "mean_full_response_intvty": 500.30881, "mean_output_tokens_actual": 1996.80065, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.9794021348879643, "p75_output_tokens_expected": 3047.75, "p90_output_tokens_expected": 5445.5, "p95_output_tokens_expected": 6908.25, "std_output_tokens_expected": 2795.2732, "theoretical_cache_hit_rate": 0.98315, "mean_output_tokens_expected": 2159.62745, "median_full_response_intvty": 507.20586, "median_output_tokens_actual": 862, "median_output_tokens_expected": 1085, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-08", "workflow_run_id": "2609", "run_started_at": "2026-10-08 23:32:33+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/37598205362/attempts/1", "curve_date": "2026-10-08", "curve_workflow_run_id": "2609", "curve_run_started_at": "2026-10-08 23:32:33+00" }, { "id": "444373", "hardware": "mi355x", "framework": "atom", "model": "minimaxm3", "precision": "fp4", "spec_method": "mtp", "disagg": false, "is_multinode": false, "prefill_tp": 4, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 4, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 4, "num_decode_gpu": 4, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 2, "image": "rocm/atom-dev:nightly_202610070333", "recipe_fingerprint": "45f691dbcb55d011a3ccc5a581d62b13030b64f000665e76776e111d864d68f7", "metrics": { "pp": 1, "p75_itl": 0.0021, "p75_qps": 0, "p90_itl": 0.00245, "p90_qps": 1, "p95_itl": 0.00269, "p95_qps": 1, "std_itl": 0.00078, "std_qps": 0.3638, "dcp_size": 1, "mean_itl": 0.00206, "mean_qps": 0.1343, "p75_e2el": 4.43696, "p75_tpot": 0.0021, "p75_ttft": 0.66843, "p90_e2el": 10.03833, "p90_tpot": 0.00245, "p90_ttft": 0.90123, "p95_e2el": 12.66341, "p95_tpot": 0.00268, "p95_ttft": 1.02581, "pcp_size": 1, "std_e2el": 4.83501, "std_tpot": 0.00078, "std_ttft": 0.34939, "mean_e2el": 3.48767, "mean_tpot": 0.00206, "mean_ttft": 0.46316, "median_itl": 0.00195, "median_qps": 0, "p75_intvty": 476.1904761904762, "p90_intvty": 408.16326530612247, "p95_intvty": 371.74721189591077, "std_intvty": 67.90387, "mean_intvty": 485.43689320388347, "median_e2el": 1.48274, "median_tpot": 0.00195, "median_ttft": 0.35234, "offload_mode": "off", "tput_per_gpu": 6981.0702, "kv_offloading": "none", "median_intvty": 512.8205128205128, "input_tput_tps": 27721.1256, "total_tput_tps": 27924.28078, "output_tput_tps": 203.15518, "duration_seconds": 3605.66737, "p75_input_tokens": 362391, "p90_input_tokens": 481631.4, "p95_input_tokens": 535678.4, "std_input_tokens": 171496.74979, "mean_input_tokens": 206088.98557, "input_tput_per_gpu": 6930.2814, "median_input_tokens": 147132, "output_tput_per_gpu": 50.78879, "total_prompt_tokens": 104314340, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.0021, "p90_full_response_itl": 0.00245, "p95_full_response_itl": 0.00269, "std_full_response_itl": 0.00078, "gpu_kv_cache_usage_pct": 0.09613, "mean_full_response_itl": 0.00206, "total_generation_tokens": 732532, "median_full_response_itl": 0.00195, "p75_full_response_intvty": 475.2472, "p75_output_tokens_actual": 1978, "p90_full_response_intvty": 407.50325, "p90_output_tokens_actual": 4658.4, "p95_full_response_intvty": 371.61608, "p95_output_tokens_actual": 5957.8, "std_full_response_intvty": 67.90387, "std_output_tokens_actual": 2329.4763, "total_requests_completed": 485, "mean_full_response_intvty": 484.37888, "mean_output_tokens_actual": 1510.3299, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.9730883788365051, "p75_output_tokens_expected": 3213, "p90_output_tokens_expected": 5359, "p95_output_tokens_expected": 6577.2, "std_output_tokens_expected": 2554.76516, "theoretical_cache_hit_rate": 0.97761, "mean_output_tokens_expected": 2119.35464, "median_full_response_intvty": 513.96777, "median_output_tokens_actual": 509, "median_output_tokens_expected": 1034, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-08", "workflow_run_id": "2609", "run_started_at": "2026-10-08 23:32:33+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/37598205362/attempts/1", "curve_date": "2026-10-08", "curve_workflow_run_id": "2609", "curve_run_started_at": "2026-10-08 23:32:33+00" }, { "id": "444387", "hardware": "mi355x", "framework": "atom", "model": "minimaxm3", "precision": "fp4", "spec_method": "mtp", "disagg": false, "is_multinode": false, "prefill_tp": 4, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 4, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 4, "num_decode_gpu": 4, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 4, "image": "rocm/atom-dev:nightly_202610070333", "recipe_fingerprint": "45f691dbcb55d011a3ccc5a581d62b13030b64f000665e76776e111d864d68f7", "metrics": { "pp": 1, "p75_itl": 0.00227, "p75_qps": 0, "p90_itl": 0.00267, "p90_qps": 1, "p95_itl": 0.00304, "p95_qps": 1, "std_itl": 0.00103, "std_qps": 0.4403, "dcp_size": 1, "mean_itl": 0.00219, "mean_qps": 0.19438, "p75_e2el": 3.70474, "p75_tpot": 0.00227, "p75_ttft": 0.56974, "p90_e2el": 8.44316, "p90_tpot": 0.00267, "p90_ttft": 0.96986, "p95_e2el": 11.87163, "p95_tpot": 0.00304, "p95_ttft": 1.08849, "pcp_size": 1, "std_e2el": 4.90668, "std_tpot": 0.00103, "std_ttft": 0.38775, "mean_e2el": 3.19962, "mean_tpot": 0.00218, "mean_ttft": 0.43042, "median_itl": 0.00196, "median_qps": 0, "p75_intvty": 440.5286343612335, "p90_intvty": 374.5318352059925, "p95_intvty": 328.9473684210526, "std_intvty": 84.95745, "mean_intvty": 456.62100456621005, "median_e2el": 1.34726, "median_tpot": 0.00196, "median_ttft": 0.28702, "offload_mode": "off", "tput_per_gpu": 9201.17103, "kv_offloading": "none", "median_intvty": 510.2040816326531, "input_tput_tps": 36550.78997, "total_tput_tps": 36804.68411, "output_tput_tps": 253.89414, "duration_seconds": 3601.60344, "p75_input_tokens": 249678.75, "p90_input_tokens": 530981.8, "p95_input_tokens": 603402.3, "std_input_tokens": 182927.05941, "mean_input_tokens": 188059.21571, "input_tput_per_gpu": 9137.69749, "median_input_tokens": 118433, "output_tput_per_gpu": 63.47353, "total_prompt_tokens": 142127996, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00227, "p90_full_response_itl": 0.00267, "p95_full_response_itl": 0.00304, "std_full_response_itl": 0.00103, "gpu_kv_cache_usage_pct": 0.1423, "mean_full_response_itl": 0.00219, "total_generation_tokens": 914470, "median_full_response_itl": 0.00196, "p75_full_response_intvty": 441.43526, "p75_output_tokens_actual": 1490, "p90_full_response_intvty": 374.98832, "p90_output_tokens_actual": 3767.1, "p95_full_response_intvty": 328.55673, "p95_output_tokens_actual": 5417.15, "std_full_response_intvty": 84.95745, "std_output_tokens_actual": 2255.20057, "total_requests_completed": 700, "mean_full_response_intvty": 457.58226, "mean_output_tokens_actual": 1306.32286, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.9709341711959408, "p75_output_tokens_expected": 2596, "p90_output_tokens_expected": 4848.3, "p95_output_tokens_expected": 6316, "std_output_tokens_expected": 2627.23836, "theoretical_cache_hit_rate": 0.97896, "mean_output_tokens_expected": 1798.82857, "median_full_response_intvty": 510.07203, "median_output_tokens_actual": 437, "median_output_tokens_expected": 722, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-08", "workflow_run_id": "2609", "run_started_at": "2026-10-08 23:32:33+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/37598205362/attempts/1", "curve_date": "2026-10-08", "curve_workflow_run_id": "2609", "curve_run_started_at": "2026-10-08 23:32:33+00" }, { "id": "444638", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 4, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 4, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 4, "num_decode_gpu": 4, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 1, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "2b31da7c71cae62527fdb06f6f9ab6b58053cae6625b192fcc6a11b5d6943927", "metrics": { "pp": 1, "p75_itl": 0.00149, "p75_qps": 0, "p90_itl": 0.00156, "p90_qps": 0, "p95_itl": 0.00173, "p95_qps": 1, "std_itl": 0.00012, "std_qps": 0.29351, "dcp_size": 1, "mean_itl": 0.00149, "mean_qps": 0.09179, "p75_e2el": 4.84577, "p75_tpot": 0.00149, "p75_ttft": 1.01753, "p90_e2el": 8.23691, "p90_tpot": 0.00156, "p90_ttft": 1.19648, "p95_e2el": 10.28456, "p95_tpot": 0.00173, "p95_ttft": 1.26592, "pcp_size": 1, "std_e2el": 3.71502, "std_tpot": 0.00012, "std_ttft": 0.38115, "mean_e2el": 3.57941, "mean_tpot": 0.00149, "mean_ttft": 0.6892, "median_itl": 0.00146, "median_qps": 0, "p75_intvty": 671.1409395973154, "p90_intvty": 641.025641025641, "p95_intvty": 578.0346820809249, "std_intvty": 41.01459, "mean_intvty": 671.1409395973154, "median_e2el": 2.09957, "median_tpot": 0.00146, "median_ttft": 0.67829, "offload_mode": "off", "tput_per_gpu": 7587.67226, "kv_offloading": "none", "median_intvty": 684.9315068493152, "input_tput_tps": 30168.69943, "total_tput_tps": 30350.68903, "output_tput_tps": 181.98961, "duration_seconds": 3598.38684, "p75_input_tokens": 490486.5, "p90_input_tokens": 596955, "p95_input_tokens": 641161, "std_input_tokens": 193516.21812, "mean_input_tokens": 327971.75529, "input_tput_per_gpu": 7542.17486, "median_input_tokens": 353674, "output_tput_per_gpu": 45.4974, "total_prompt_tokens": 112130135, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00149, "p90_full_response_itl": 0.00156, "p95_full_response_itl": 0.00173, "std_full_response_itl": 0.00012, "gpu_kv_cache_usage_pct": 0.01153, "mean_full_response_itl": 0.00149, "total_generation_tokens": 654880, "median_full_response_itl": 0.00146, "p75_full_response_intvty": 672.13479, "p75_output_tokens_actual": 2676.5, "p90_full_response_intvty": 640.17354, "p90_output_tokens_actual": 5071, "p95_full_response_intvty": 577.54131, "p95_output_tokens_actual": 6501, "std_full_response_intvty": 41.01459, "std_output_tokens_actual": 2490.16668, "total_requests_completed": 331, "mean_full_response_intvty": 669.7581, "mean_output_tokens_actual": 1978.45619, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.9816, "p75_output_tokens_expected": 2911, "p90_output_tokens_expected": 5417, "p95_output_tokens_expected": 6903, "std_output_tokens_expected": 2724.56559, "theoretical_cache_hit_rate": 0.98493, "mean_output_tokens_expected": 2155.3565, "median_full_response_intvty": 682.76004, "median_output_tokens_actual": 904, "median_output_tokens_expected": 1184, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444634", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 4, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 4, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 4, "num_decode_gpu": 4, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 8, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "2b31da7c71cae62527fdb06f6f9ab6b58053cae6625b192fcc6a11b5d6943927", "metrics": { "pp": 1, "p75_itl": 0.0019, "p75_qps": 1, "p90_itl": 0.00227, "p90_qps": 1, "p95_itl": 0.00247, "p95_qps": 2, "std_itl": 0.00058, "std_qps": 0.63528, "dcp_size": 1, "mean_itl": 0.00179, "mean_qps": 0.398, "p75_e2el": 2.6333, "p75_tpot": 0.0019, "p75_ttft": 0.49806, "p90_e2el": 5.72406, "p90_tpot": 0.00227, "p90_ttft": 0.86473, "p95_e2el": 8.45937, "p95_tpot": 0.00247, "p95_ttft": 1.08447, "pcp_size": 1, "std_e2el": 3.47373, "std_tpot": 0.00058, "std_ttft": 0.28625, "mean_e2el": 2.43776, "mean_tpot": 0.00179, "mean_ttft": 0.4371, "median_itl": 0.00169, "median_qps": 0, "p75_intvty": 526.3157894736842, "p90_intvty": 440.5286343612335, "p95_intvty": 404.8582995951417, "std_intvty": 100.98025, "mean_intvty": 558.659217877095, "median_e2el": 1.19393, "median_tpot": 0.00169, "median_ttft": 0.32714, "offload_mode": "off", "tput_per_gpu": 16424.34511, "kv_offloading": "none", "median_intvty": 591.7159763313609, "input_tput_tps": 65239.07747, "total_tput_tps": 65697.38042, "output_tput_tps": 458.30295, "duration_seconds": 3598.17888, "p75_input_tokens": 187368, "p90_input_tokens": 395189, "p95_input_tokens": 517820.4, "std_input_tokens": 144445.10733, "mean_input_tokens": 163811.49407, "input_tput_per_gpu": 16309.76937, "median_input_tokens": 108942, "output_tput_per_gpu": 114.57574, "total_prompt_tokens": 252898075, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.0019, "p90_full_response_itl": 0.00227, "p95_full_response_itl": 0.00247, "std_full_response_itl": 0.00058, "gpu_kv_cache_usage_pct": 0.02615, "mean_full_response_itl": 0.00179, "total_generation_tokens": 1649143, "median_full_response_itl": 0.00169, "p75_full_response_intvty": 526.08926, "p75_output_tokens_actual": 1227, "p90_full_response_intvty": 440.70411, "p90_output_tokens_actual": 2924.2, "p95_full_response_intvty": 405.17763, "p95_output_tokens_actual": 4534.2, "std_full_response_intvty": 100.98025, "std_output_tokens_actual": 1871.7133, "total_requests_completed": 1433, "mean_full_response_intvty": 559.50563, "mean_output_tokens_actual": 1150.77181, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.96958, "p75_output_tokens_expected": 2613.5, "p90_output_tokens_expected": 4708.2, "p95_output_tokens_expected": 6727, "std_output_tokens_expected": 2482.06696, "theoretical_cache_hit_rate": 0.979, "mean_output_tokens_expected": 1934.95026, "median_full_response_intvty": 593.20865, "median_output_tokens_actual": 454, "median_output_tokens_expected": 1012, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444632", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 2, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 2, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 2, "num_decode_gpu": 2, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 1, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "5485fd3db19cab9cf070859d2fc446b6c77002efa741f35d6d0c27f27956588e", "metrics": { "pp": 1, "p75_itl": 0.00179, "p75_qps": 0, "p90_itl": 0.00198, "p90_qps": 0, "p95_itl": 0.00221, "p95_qps": 1, "std_itl": 0.00022, "std_qps": 0.28904, "dcp_size": 1, "mean_itl": 0.00182, "mean_qps": 0.08727, "p75_e2el": 5.62421, "p75_tpot": 0.00179, "p75_ttft": 1.01492, "p90_e2el": 9.75853, "p90_tpot": 0.00197, "p90_ttft": 1.20289, "p95_e2el": 12.54239, "p95_tpot": 0.00221, "p95_ttft": 1.28125, "pcp_size": 1, "std_e2el": 4.50351, "std_tpot": 0.00022, "std_ttft": 0.37685, "mean_e2el": 4.26732, "mean_tpot": 0.00182, "mean_ttft": 0.77543, "median_itl": 0.00176, "median_qps": 0, "p75_intvty": 558.659217877095, "p90_intvty": 505.0505050505051, "p95_intvty": 452.4886877828054, "std_intvty": 46.61585, "mean_intvty": 549.4505494505495, "median_e2el": 2.38932, "median_tpot": 0.00176, "median_ttft": 0.78981, "offload_mode": "off", "tput_per_gpu": 13915.15606, "kv_offloading": "none", "median_intvty": 568.1818181818181, "input_tput_tps": 27657.37484, "total_tput_tps": 27830.31212, "output_tput_tps": 172.93728, "duration_seconds": 3601.79723, "p75_input_tokens": 467739.5, "p90_input_tokens": 569336.8, "p95_input_tokens": 605864, "std_input_tokens": 185320.65863, "mean_input_tokens": 316242.08254, "input_tput_per_gpu": 13828.68742, "median_input_tokens": 343294, "output_tput_per_gpu": 86.46864, "total_prompt_tokens": 103187762, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00179, "p90_full_response_itl": 0.00198, "p95_full_response_itl": 0.00221, "std_full_response_itl": 0.00022, "gpu_kv_cache_usage_pct": 0.01876, "mean_full_response_itl": 0.00182, "total_generation_tokens": 622896, "median_full_response_itl": 0.00176, "p75_full_response_intvty": 557.92733, "p75_output_tokens_actual": 2715, "p90_full_response_intvty": 505.30145, "p90_output_tokens_actual": 5068.2, "p95_full_response_intvty": 453.09192, "p95_output_tokens_actual": 6710.5, "std_full_response_intvty": 46.61585, "std_output_tokens_actual": 2524.37025, "total_requests_completed": 315, "mean_full_response_intvty": 549.30417, "mean_output_tokens_actual": 1977.4127, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.980355635583995, "p75_output_tokens_expected": 2964, "p90_output_tokens_expected": 5379.4, "p95_output_tokens_expected": 6905.1, "std_output_tokens_expected": 2762.23702, "theoretical_cache_hit_rate": 0.98393, "mean_output_tokens_expected": 2150.13968, "median_full_response_intvty": 567.58284, "median_output_tokens_actual": 867, "median_output_tokens_expected": 1155, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444641", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 2, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 2, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 2, "num_decode_gpu": 2, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 2, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "5485fd3db19cab9cf070859d2fc446b6c77002efa741f35d6d0c27f27956588e", "metrics": { "pp": 1, "p75_itl": 0.00194, "p75_qps": 0, "p90_itl": 0.00225, "p90_qps": 1, "p95_itl": 0.00244, "p95_qps": 1, "std_itl": 0.00078, "std_qps": 0.3654, "dcp_size": 1, "mean_itl": 0.00194, "mean_qps": 0.13513, "p75_e2el": 4.20416, "p75_tpot": 0.00194, "p75_ttft": 0.80597, "p90_e2el": 9.0584, "p90_tpot": 0.00225, "p90_ttft": 1.0165, "p95_e2el": 11.83423, "p95_tpot": 0.00244, "p95_ttft": 1.1548, "pcp_size": 1, "std_e2el": 4.31935, "std_tpot": 0.00078, "std_ttft": 0.48902, "mean_e2el": 3.30792, "mean_tpot": 0.00194, "mean_ttft": 0.57476, "median_itl": 0.00177, "median_qps": 0, "p75_intvty": 515.4639175257731, "p90_intvty": 444.44444444444446, "p95_intvty": 409.8360655737705, "std_intvty": 70.51194, "mean_intvty": 515.4639175257731, "median_e2el": 1.53504, "median_tpot": 0.00177, "median_ttft": 0.42126, "offload_mode": "off", "tput_per_gpu": 14370.98804, "kv_offloading": "none", "median_intvty": 564.9717514124294, "input_tput_tps": 28538.01887, "total_tput_tps": 28741.97607, "output_tput_tps": 203.9572, "duration_seconds": 3605.47705, "p75_input_tokens": 365801.25, "p90_input_tokens": 494647.9, "p95_input_tokens": 548167.65, "std_input_tokens": 175786.82866, "mean_input_tokens": 210846.66393, "input_tput_per_gpu": 14269.00944, "median_input_tokens": 149848.5, "output_tput_per_gpu": 101.9786, "total_prompt_tokens": 107245452, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00194, "p90_full_response_itl": 0.00225, "p95_full_response_itl": 0.00244, "std_full_response_itl": 0.00078, "gpu_kv_cache_usage_pct": 0.02297, "mean_full_response_itl": 0.00194, "total_generation_tokens": 735385, "median_full_response_itl": 0.00177, "p75_full_response_intvty": 514.26415, "p75_output_tokens_actual": 1959.25, "p90_full_response_intvty": 443.59826, "p90_output_tokens_actual": 4655.7, "p95_full_response_intvty": 410.30416, "p95_output_tokens_actual": 5950.15, "std_full_response_intvty": 70.51194, "std_output_tokens_actual": 2317.75533, "total_requests_completed": 488, "mean_full_response_intvty": 514.51259, "mean_output_tokens_actual": 1506.89139, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.97376, "p75_output_tokens_expected": 3213, "p90_output_tokens_expected": 5359, "p95_output_tokens_expected": 6539.1, "std_output_tokens_expected": 2550.92078, "theoretical_cache_hit_rate": 0.97818, "mean_output_tokens_expected": 2127.77869, "median_full_response_intvty": 564.20062, "median_output_tokens_actual": 520, "median_output_tokens_expected": 1085, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444633", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 2, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 2, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 2, "num_decode_gpu": 2, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 4, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "5485fd3db19cab9cf070859d2fc446b6c77002efa741f35d6d0c27f27956588e", "metrics": { "pp": 1, "p75_itl": 0.00213, "p75_qps": 0, "p90_itl": 0.00233, "p90_qps": 1, "p95_itl": 0.00284, "p95_qps": 1, "std_itl": 0.00112, "std_qps": 0.42807, "dcp_size": 1, "mean_itl": 0.00205, "mean_qps": 0.19617, "p75_e2el": 3.68529, "p75_tpot": 0.00213, "p75_ttft": 0.64489, "p90_e2el": 7.69484, "p90_tpot": 0.00233, "p90_ttft": 1.15877, "p95_e2el": 11.31937, "p95_tpot": 0.00284, "p95_ttft": 1.24745, "pcp_size": 1, "std_e2el": 4.27674, "std_tpot": 0.00111, "std_ttft": 0.39194, "mean_e2el": 3.01436, "mean_tpot": 0.00205, "mean_ttft": 0.51756, "median_itl": 0.00179, "median_qps": 0, "p75_intvty": 469.4835680751174, "p90_intvty": 429.18454935622316, "p95_intvty": 352.11267605633805, "std_intvty": 82.22068, "mean_intvty": 487.80487804878044, "median_e2el": 1.38169, "median_tpot": 0.00179, "median_ttft": 0.3881, "offload_mode": "off", "tput_per_gpu": 18603.74721, "kv_offloading": "none", "median_intvty": 558.659217877095, "input_tput_tps": 36949.52892, "total_tput_tps": 37207.49442, "output_tput_tps": 257.9655, "duration_seconds": 3604.59448, "p75_input_tokens": 251891, "p90_input_tokens": 529102.8, "p95_input_tokens": 601710.6, "std_input_tokens": 181884.83095, "mean_input_tokens": 188384.82037, "input_tput_per_gpu": 18474.76446, "median_input_tokens": 119465, "output_tput_per_gpu": 128.98275, "total_prompt_tokens": 143657055, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00213, "p90_full_response_itl": 0.00233, "p95_full_response_itl": 0.00284, "std_full_response_itl": 0.00112, "gpu_kv_cache_usage_pct": 0.03029, "mean_full_response_itl": 0.00205, "total_generation_tokens": 929905, "median_full_response_itl": 0.00179, "p75_full_response_intvty": 470.22692, "p75_output_tokens_actual": 1514, "p90_full_response_intvty": 429.43083, "p90_output_tokens_actual": 3785.4, "p95_full_response_intvty": 352.14033, "p95_output_tokens_actual": 5399.6, "std_full_response_intvty": 82.22068, "std_output_tokens_actual": 2252.28124, "total_requests_completed": 707, "mean_full_response_intvty": 487.75171, "mean_output_tokens_actual": 1315.22065, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.97071, "p75_output_tokens_expected": 2552, "p90_output_tokens_expected": 4848, "p95_output_tokens_expected": 6238.3, "std_output_tokens_expected": 2618.18655, "theoretical_cache_hit_rate": 0.97914, "mean_output_tokens_expected": 1784.76096, "median_full_response_intvty": 557.5577, "median_output_tokens_actual": 446, "median_output_tokens_expected": 710, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444636", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 2, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 2, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 2, "num_decode_gpu": 2, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 6, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "5485fd3db19cab9cf070859d2fc446b6c77002efa741f35d6d0c27f27956588e", "metrics": { "pp": 1, "p75_itl": 0.00252, "p75_qps": 0, "p90_itl": 0.00309, "p90_qps": 1, "p95_itl": 0.00347, "p95_qps": 1, "std_itl": 0.0015, "std_qps": 0.57242, "dcp_size": 1, "mean_itl": 0.00234, "mean_qps": 0.29502, "p75_e2el": 3.58735, "p75_tpot": 0.00252, "p75_ttft": 0.58864, "p90_e2el": 7.11371, "p90_tpot": 0.00309, "p90_ttft": 0.93212, "p95_e2el": 9.81491, "p95_tpot": 0.00347, "p95_ttft": 1.14106, "pcp_size": 1, "std_e2el": 3.6869, "std_tpot": 0.0015, "std_ttft": 0.32306, "mean_e2el": 2.99263, "mean_tpot": 0.00233, "mean_ttft": 0.4587, "median_itl": 0.00209, "median_qps": 0, "p75_intvty": 396.8253968253968, "p90_intvty": 323.62459546925567, "p95_intvty": 288.1844380403458, "std_intvty": 105.16646, "mean_intvty": 427.3504273504273, "median_e2el": 1.60983, "median_tpot": 0.00209, "median_ttft": 0.32251, "offload_mode": "off", "tput_per_gpu": 24203.64883, "kv_offloading": "none", "median_intvty": 478.4688995215311, "input_tput_tps": 48057.059, "total_tput_tps": 48407.29766, "output_tput_tps": 350.23866, "duration_seconds": 3596.0536, "p75_input_tokens": 233037, "p90_input_tokens": 401333, "p95_input_tokens": 538439, "std_input_tokens": 156347.22179, "mean_input_tokens": 162880.0754, "input_tput_per_gpu": 24028.5295, "median_input_tokens": 97088, "output_tput_per_gpu": 175.11933, "total_prompt_tokens": 189863910, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00252, "p90_full_response_itl": 0.00309, "p95_full_response_itl": 0.00347, "std_full_response_itl": 0.0015, "gpu_kv_cache_usage_pct": 0.0395, "mean_full_response_itl": 0.00234, "total_generation_tokens": 1259542, "median_full_response_itl": 0.00209, "p75_full_response_intvty": 396.93735, "p75_output_tokens_actual": 1437, "p90_full_response_intvty": 323.82205, "p90_output_tokens_actual": 3049, "p95_full_response_intvty": 287.81835, "p95_output_tokens_actual": 4388, "std_full_response_intvty": 105.16646, "std_output_tokens_actual": 1739.8037, "total_requests_completed": 1061, "mean_full_response_intvty": 428.19728, "mean_output_tokens_actual": 1187.06598, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.96853, "p75_output_tokens_expected": 1923, "p90_output_tokens_expected": 4293.4, "p95_output_tokens_expected": 6491, "std_output_tokens_expected": 2342.65279, "theoretical_cache_hit_rate": 0.97802, "mean_output_tokens_expected": 1640.75281, "median_full_response_intvty": 479.33013, "median_output_tokens_actual": 550, "median_output_tokens_expected": 785, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" }, { "id": "444639", "hardware": "mi355x", "framework": "atom", "model": "dsv41flash", "precision": "fp4", "spec_method": "draft_model", "disagg": false, "is_multinode": false, "prefill_tp": 2, "prefill_ep": 1, "prefill_dp_attention": false, "prefill_num_workers": 0, "decode_tp": 2, "decode_ep": 1, "decode_dp_attention": false, "decode_num_workers": 0, "num_prefill_gpu": 2, "num_decode_gpu": 2, "benchmark_type": "agentic_traces", "offload_mode": "off", "isl": null, "osl": null, "conc": 8, "image": "rocm/atom-dev:nightly_202610121516", "recipe_fingerprint": "5485fd3db19cab9cf070859d2fc446b6c77002efa741f35d6d0c27f27956588e", "metrics": { "pp": 1, "p75_itl": 0.00281, "p75_qps": 1, "p90_itl": 0.00317, "p90_qps": 1, "p95_itl": 0.00341, "p95_qps": 2, "std_itl": 0.00082, "std_qps": 0.63641, "dcp_size": 1, "mean_itl": 0.00239, "mean_qps": 0.3936, "p75_e2el": 3.3044, "p75_tpot": 0.00281, "p75_ttft": 0.47861, "p90_e2el": 7.14091, "p90_tpot": 0.00317, "p90_ttft": 0.78854, "p95_e2el": 11.02046, "p95_tpot": 0.00341, "p95_ttft": 0.9693, "pcp_size": 1, "std_e2el": 4.45058, "std_tpot": 0.00082, "std_ttft": 0.3235, "mean_e2el": 3.02775, "mean_tpot": 0.00239, "mean_ttft": 0.38609, "median_itl": 0.00221, "median_qps": 0, "p75_intvty": 355.87188612099646, "p90_intvty": 315.45741324921136, "p95_intvty": 293.2551319648094, "std_intvty": 101.78002, "mean_intvty": 418.41004184100416, "median_e2el": 1.423, "median_tpot": 0.00221, "median_ttft": 0.26934, "offload_mode": "off", "tput_per_gpu": 32004.57809, "kv_offloading": "none", "median_intvty": 452.4886877828054, "input_tput_tps": 63555.81335, "total_tput_tps": 64009.15618, "output_tput_tps": 453.34283, "duration_seconds": 3595.24161, "p75_input_tokens": 183642.5, "p90_input_tokens": 386636.5, "p95_input_tokens": 506072.5, "std_input_tokens": 141303.76513, "mean_input_tokens": 161369.00071, "input_tput_per_gpu": 31777.90667, "median_input_tokens": 108369, "output_tput_per_gpu": 226.67141, "total_prompt_tokens": 246645946, "allocated_cpu_dram_gb": 0, "p75_full_response_itl": 0.00281, "p90_full_response_itl": 0.00317, "p95_full_response_itl": 0.00341, "std_full_response_itl": 0.00082, "gpu_kv_cache_usage_pct": 0.04465, "mean_full_response_itl": 0.00239, "total_generation_tokens": 1629964, "median_full_response_itl": 0.00221, "p75_full_response_intvty": 356.38355, "p75_output_tokens_actual": 1222.5, "p90_full_response_intvty": 315.42842, "p90_output_tokens_actual": 2934.5, "p95_full_response_intvty": 293.09316, "p95_output_tokens_actual": 4571.25, "std_full_response_intvty": 101.78002, "std_output_tokens_actual": 1878.72562, "total_requests_completed": 1416, "mean_full_response_intvty": 418.10148, "mean_output_tokens_actual": 1151.04308, "server_cpu_cache_hit_rate": 0, "server_gpu_cache_hit_rate": 0.96892, "p75_output_tokens_expected": 2646.5, "p90_output_tokens_expected": 4796, "p95_output_tokens_expected": 6727, "std_output_tokens_expected": 2494.32516, "theoretical_cache_hit_rate": 0.97856, "mean_output_tokens_expected": 1943.10763, "median_full_response_intvty": 452.42844, "median_output_tokens_actual": 452.5, "median_output_tokens_expected": 1012, "server_external_cache_hit_rate": 0 }, "workers": null, "power_invalid_reasons": null, "power_audit": null, "date": "2026-10-13", "workflow_run_id": "2625", "run_started_at": "2026-10-13 05:24:09+00", "run_url": "https://github.com/SemiAnalysisAI/InferenceX/actions/runs/38276304505/attempts/2", "curve_date": "2026-10-13", "curve_workflow_run_id": "2625", "curve_run_started_at": "2026-10-13 05:24:09+00" } ], "acceptance_settings": { "MiniMax-M3-MXFP4": { "mode": "synthetic", "thinking_mode": "thinking_on", "draft_tokens": 3, "length": 2.78, "source": "https://github.com/SemiAnalysisAI/InferenceX/blob/4b94f096b74191b8af6b7c89e50e0b6bc69f7aba/inferencex-e2e/infx/golden_al_distribution/minimaxm3_eagle3_gqa.yaml" }, "DeepSeek-V4.1-Flash": { "mode": "synthetic", "thinking_mode": "thinking_on", "draft_tokens": 5, "length": 3.51, "source": "https://github.com/SemiAnalysisAI/InferenceX/blob/2597d3d1ce010ee10b82e08bd5b448fea054b5ac/inferencex-e2e/infx/golden_al_distribution/dsv41flash_dspark.yaml" } } }