Result statement
NVIDIA ran DeepSeek-R1 on NVIDIA GB300 NVL72 and achieved a verified 546,980.1 tokens/second System TPS at 15,410.8ms TTFT P90 in the Standardized division of the Available category .
[1] MLPerf Endpoints v1.0 DeepSeek-R1; retrieved from endpoints.mlcommons.org on October 18, 2026. Result —, run 14: concurrency = 28,080 queries, TTFT P90 = 15,410.8ms, System TPS = 546,980.1 tokens/sec and interactivity = 19.5 tokens/sec per user. Result verified by MLCommons Association. The MLPerf name and logo are registered and unregistered trademarks of MLCommons Association in the United States and other countries. All rights reserved. Unauthorized use strictly prohibited. See www.mlcommons.org for more information.
NVIDIA
27168b87-e0f9-4488-ae93-88a45c1c81c9
Submission Status PUBLISHED
2026-06-25
—
Division Standardized
Deployment Client on Premises
Model Name DeepSeek-R1
Division Standardized
Model Precision —
Model Link —
Calibration Link —
Model Notes —
Dataset Name dsr1-perf
Dataset Type performance
Average Input Tokens —
Dataset Link —
System TPS vs. Tokens/s per User
System TPS vs. Concurrency
Time to First Token P90 vs. Concurrency
Tokens/s per User vs. Concurrency
NVIDIA GB300 NVL72
Available
System Category Datacenter · GB300 NVL72 rack, disaggregated TRT-LLM, NVFP4 weights
—
1
18
Max Supported Concurrency 28,080
Node Type ID 0
Node Count 18
Framework TensorRT-LLM
TensorRT-LLM
TensorRT-LLM
—
Link to Container —
Operating System unspecified
Filesystem —
Other Software —
Software Notes —
Processor
NVIDIA Grace (ARM Neoverse V2)
1
72
VCPUs Per Processor —
Accelerator
NVIDIA GB300
4
—
288 GB HBM3e
NVLink 5 + NVSwitch (NVL72)
NVLink-C2C (Grace-Blackwell)
Host / Storage
unspecified
Memory Configuration —
Storage Capacity unspecified
Storage Type unspecified
Networking
—
NVLink + InfiniBand
Network Topology NVL72 single-rack NVLink domain
Other
Other Hardware —
Cooling Liquid-cooled
Hardware Notes —
| Field Name | Run 1 | Run 2 | Run 3 | Run 4 | Run 5 | Run 6 | Run 7 | Run 8 | Run 9 | Run 10 | Run 11 | Run 12 | Run 13 | Run 14 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Run Date | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 | 06/25/2026 |
| Concurrency | 1 | 2 | 4 | 48 | 64 | 128 | 256 | 512 | 1,024 | 2,048 | 4,096 | 8,192 | 16,384 | 28,080 |
| System Tokens/Second | 338.8 | 640.9 | 1,225.1 | 10,556.5 | 14,161.0 | 27,256.6 | 45,293.7 | 81,627.8 | 130,376.4 | 198,881.9 | 306,707.1 | 395,958.1 | 469,668.6 | 546,980.1 |
| Tokens/Second per User | 338.8 | 320.5 | 306.3 | 219.9 | 221.3 | 212.9 | 176.9 | 159.4 | 127.3 | 97.1 | 74.9 | 48.3 | 28.7 | 19.5 |
| TTFT P90 (ms) | 1123.1 | 251.8 | 259.1 | 432.8 | 423.9 | 492.7 | 2051.5 | 491.5 | 541.0 | 989.2 | 2851.9 | 5612.9 | 15454.0 | 15410.8 |
| Utilization | 0.1% | 0.1% | 0.2% | 1.9% | 2.6% | 5.0% | 8.3% | 14.9% | 23.8% | 36.4% | 56.1% | 72.4% | 85.9% | 100.0% |
| Configuration Summary | dsr1-gb300x72-pareto-nvl72-c1 · submission | dsr1-gb300x72-pareto-nvl72-c2 · submission | dsr1-gb300x72-pareto-nvl72-c4 · submission | dsr1-gb300x72-pareto-nvl72-c48 · submission | dsr1-gb300x72-pareto-nvl72-c64 · submission | dsr1-gb300x72-pareto-nvl72-c128 · submission | dsr1-gb300x72-pareto-nvl72-c256 · submission | dsr1-gb300x72-pareto-nvl72-c512 · submission | dsr1-gb300x72-pareto-nvl72-c1024 · submission | dsr1-gb300x72-pareto-nvl72-c2048 · submission | dsr1-gb300x72-pareto-nvl72-c4096 · submission | dsr1-gb300x72-pareto-nvl72-c8192 · submission | dsr1-gb300x72-pareto-nvl72-c16384 · submission | dsr1-gb300x72-pareto-nvl72-c28080-numa · submission |
| Config Type | submission | submission | submission | submission | submission | submission | submission | submission | submission | submission | submission | submission | submission | submission |
| Config Region | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Target QPS | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Result State | complete | complete | complete | complete | complete | complete | complete | complete | complete | complete | complete | complete | complete | complete |
| Measured Accuracy Score | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Benchmark Version | 05ce400 | 05ce400 | 05ce400 | ced1dcd | ed0c53d | 9019316 | cd477d4 | 429c5b8 | 9cbfefe | af6bf98 | cd477d4 | 0470dab | 7265704 | e072ac2 |
| Result Version | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 | 0.1.0 |
| Archive URI | gs://mlperf-endpoints-blob/runs/49655a2c-3d3d-45f4-916e-54c439080875/archive.tar.gz | gs://mlperf-endpoints-blob/runs/00747f95-5433-4718-99a6-17c91e6000bb/archive.tar.gz | gs://mlperf-endpoints-blob/runs/cbe204a9-7436-4ff0-85b0-a7b37b4b785c/archive.tar.gz | gs://mlperf-endpoints-blob/runs/6ae0b5b8-33d9-4640-af10-5adfa595b66f/archive.tar.gz | gs://mlperf-endpoints-blob/runs/4f76a1f0-e4a7-4fc1-ab15-f767674462b2/archive.tar.gz | gs://mlperf-endpoints-blob/runs/1d17bdfe-f13c-44c3-beda-3041f3cfd2ae/archive.tar.gz | gs://mlperf-endpoints-blob/runs/a913ac1d-7f38-453a-ba89-e0cbd1ab5250/archive.tar.gz | gs://mlperf-endpoints-blob/runs/aa5cc47c-c121-41c2-99d8-ccd4bbd607b9/archive.tar.gz | gs://mlperf-endpoints-blob/runs/a9c5e6a2-71f1-4d0d-8302-f1cc94a43ab4/archive.tar.gz | gs://mlperf-endpoints-blob/runs/9784c932-acdd-46de-9096-fbb6a1e4a723/archive.tar.gz | gs://mlperf-endpoints-blob/runs/523ba87c-7040-4147-bb76-360ae7979a56/archive.tar.gz | gs://mlperf-endpoints-blob/runs/ec5cf744-36e8-4e9d-8356-618204ec2339/archive.tar.gz | gs://mlperf-endpoints-blob/runs/164c6f8d-96d1-4aab-b7e1-5fbcf7d1ff40/archive.tar.gz | gs://mlperf-endpoints-blob/runs/68478c69-8d3c-4be8-82d9-8f79a3673c63/archive.tar.gz |
| QPS | 0.0887 | 0.1568 | 0.3305 | 2.8298 | 3.7759 | 7.2743 | 11.9376 | 21.7036 | 34.7665 | 52.9578 | 81.3117 | 105.4162 | 125.2644 | 145.1861 |
| Total Output Tokens | 954,745 | 2,043,724 | 3,706,833 | 24,553,924 | 38,253,793 | 82,208,882 | 83,244,526 | 280,557,875 | 394,925,643 | 725,078,131 | 1,075,847,720 | 1,318,555,925 | 1,349,790,344 | 2,479,730,589 |
| Average Output Tokens | 3,818.98 | 4,087.45 | 3,706.83 | 3,730.47 | 3,750.37 | 3,746.99 | 3,794.19 | 3,761.03 | 3,750.05 | 3,755.48 | 3,771.99 | 3,756.17 | 3,749.42 | 3,767.44 |
| Run Duration (s) | 2,817.61 | 3,188.72 | 3,025.62 | 2,325.96 | 2,701.36 | 3,016.11 | 1,837.88 | 3,437.04 | 3,029.12 | 3,645.77 | 3,507.74 | 3,330.04 | 2,873.92 | 4,533.49 |
| Total Queries | 250 | 500 | 1,000 | 6,582 | 10,200 | 21,940 | 21,940 | 74,596 | 105,312 | 193,072 | 285,220 | 351,040 | 360,000 | 658,200 |
| Total Queries Issued | 250 | 500 | 1,000 | 6,582 | 10,200 | 21,940 | 21,940 | 74,596 | 105,312 | 193,072 | 285,220 | 351,040 | 360,000 | 658,200 |
| Total Queries Failed | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Time to First Token (TTFT) (ms) | ||||||||||||||
| Minimum | 143.7 | 140.7 | 201.4 | 164.1 | 165.0 | 168.8 | 142.8 | 179.5 | 157.9 | 215.6 | 174.1 | 180.1 | 217.2 | 176.2 |
| Average | 325.9 | 276.7 | 274.5 | 312.6 | 307.6 | 368.6 | 776.8 | 397.3 | 388.9 | 608.0 | 1109.7 | 1857.6 | 4698.1 | 6033.8 |
| P50 | 229.9 | 233.3 | 239.5 | 268.1 | 266.3 | 342.7 | 329.7 | 391.8 | 329.2 | 483.6 | 430.9 | 454.4 | 535.7 | 356.4 |
| P90 | 1123.1 | 251.8 | 259.1 | 432.8 | 423.9 | 492.7 | 2051.5 | 491.5 | 541.0 | 989.2 | 2851.9 | 5612.9 | 15454.0 | 15410.8 |
| P95 | 1193.0 | 1065.0 | 317.5 | 467.2 | 451.2 | 521.6 | 2679.3 | 520.1 | 572.1 | 1306.4 | 4550.5 | 7084.9 | 19877.0 | 36763.2 |
| P99 | 1312.3 | 1269.9 | 1284.1 | 534.2 | 522.0 | 579.4 | 3902.9 | 599.4 | 726.4 | 2272.3 | 8526.0 | 18516.0 | 46295.0 | 82744.4 |
| P999 | 1331.2 | 1340.4 | 1873.2 | 1536.6 | 1464.4 | 1959.1 | 5126.2 | 2912.5 | 4622.4 | 6870.2 | 15467.2 | 27366.3 | 55336.3 | 99115.9 |
| Maximum | 1332.3 | 1395.6 | 1875.5 | 1537.3 | 1465.2 | 1994.5 | 5791.3 | 2984.8 | 4636.5 | 7897.7 | 16710.5 | 28908.9 | 57636.7 | 101763.9 |
| Time Per Output Token (TPOT) (ms) | ||||||||||||||
| Minimum | 0.00 | 0.10 | 0.67 | 2.94 | 2.93 | 3.10 | 1.73 | 4.09 | 4.72 | 4.54 | 2.78 | 5.13 | 5.45 | 12.6 |
| Average | 2.86 | 3.09 | 3.30 | 4.49 | 4.49 | 4.63 | 5.35 | 6.21 | 7.74 | 10.1 | 12.9 | 19.2 | 31.8 | 45.5 |
| P50 | 2.90 | 3.08 | 3.26 | 4.42 | 4.43 | 4.57 | 5.30 | 6.14 | 7.65 | 10.0 | 12.8 | 19.1 | 32.1 | 47.3 |
| P90 | 3.22 | 3.54 | 3.80 | 5.00 | 4.99 | 5.13 | 5.92 | 6.87 | 8.56 | 11.3 | 14.4 | 21.5 | 36.5 | 47.9 |
| P95 | 3.29 | 3.62 | 4.02 | 5.26 | 5.16 | 5.28 | 6.03 | 7.04 | 8.76 | 11.5 | 14.7 | 22.0 | 37.3 | 48.1 |
| P99 | 3.39 | 3.80 | 4.25 | 5.90 | 5.69 | 5.83 | 6.24 | 7.50 | 9.23 | 11.9 | 15.2 | 22.8 | 38.8 | 48.5 |
| P999 | 3.41 | 3.87 | 4.39 | 6.61 | 6.13 | 6.44 | 6.45 | 8.44 | 10.7 | 12.4 | 15.8 | 24.0 | 40.6 | 49.1 |
| Maximum | 3.46 | 3.90 | 4.41 | 8.20 | 6.46 | 6.93 | 6.85 | 9.82 | 12.4 | 13.5 | 17.3 | 431.8 | 44.0 | 51.5 |
| Query Latency (ms) | ||||||||||||||
| Minimum | 1047.5 | 1073.3 | 875.7 | 1056.9 | 846.9 | 1028.1 | 1214.2 | 1173.7 | 1446.5 | 1795.7 | 2097.4 | 2405.4 | 3453.9 | 6473.4 |
| Average | 11270.3 | 12666.9 | 12075.1 | 16637.7 | 16696.3 | 17258.4 | 20562.9 | 23107.9 | 28604.7 | 37600.8 | 48620.9 | 72122.0 | 120823.5 | 177589.8 |
| P50 | 6101.3 | 7498.0 | 7388.8 | 9763.9 | 9667.0 | 9991.9 | 12080.7 | 13400.7 | 16651.7 | 21644.8 | 28269.4 | 41994.9 | 71387.1 | 103139.0 |
| P90 | 26705.5 | 34241.3 | 29495.5 | 40164.0 | 41694.6 | 42185.7 | 49759.9 | 56725.0 | 70369.8 | 92578.4 | 118723.6 | 176638.6 | 292831.2 | 435537.8 |
| P95 | 37603.6 | 45757.6 | 41506.1 | 58236.5 | 58608.2 | 60123.5 | 71469.2 | 80895.8 | 99463.5 | 131976.2 | 169503.8 | 250693.8 | 411163.5 | 613731.5 |
| P99 | 54126.9 | 58665.1 | 59404.1 | 81313.8 | 82159.6 | 84936.9 | 99759.1 | 114054.1 | 141703.1 | 186543.1 | 240107.9 | 356590.0 | 594623.3 | 937055.8 |
| P999 | 57437.5 | 62077.8 | 63359.1 | 93735.6 | 90663.4 | 93502.3 | 107668.1 | 124391.9 | 154678.5 | 203888.7 | 262579.8 | 391337.1 | 663067.2 | 955828.1 |
| Maximum | 57773.8 | 63195.9 | 63650.6 | 111624.6 | 104139.5 | 110061.7 | 122968.0 | 146297.9 | 173809.6 | 257970.0 | 346067.0 | 502808.5 | 833431.1 | 971749.4 |