Menu
Back previous/next report: k j

NVIDIA NVIDIA GB200 NVL72

Submitted by NVIDIAon 2026-06-25

Result statement

NVIDIA ran DeepSeek-R1 on NVIDIA GB200 NVL72 and achieved a verified 408,585.4 tokens/second System TPS at 7,793.8ms TTFT P90 in the Standardized division of the Available category .

[1] MLPerf Endpoints v1.0 DeepSeek-R1; retrieved from endpoints.mlcommons.org on October 22, 2026. Result —, run 11: concurrency = 16,384 queries, TTFT P90 = 7,793.8ms, System TPS = 408,585.4 tokens/sec and interactivity = 24.9 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.

Submission Info
NVIDIA
503d593f-d5a3-4a7d-a08e-17228ef09f2f
Submission Status PUBLISHED
2026-06-25
—
Division Standardized
Deployment Client on Premises
Model
Model Name DeepSeek-R1
Division Standardized
Model Precision —
Model Link —
Calibration Link —
Model Notes —
Datasets
Dataset Name dsr1-perf
Dataset Type performance
Average Input Tokens —
Dataset Link —

System TPS vs. Tokens/s per User

0100k200k300k400kSystem TPS0100200300Tokens/s per User

System TPS vs. Concurrency

0100k200k300k400kSystem TPS05,00010,00015,000# of queries

Time to First Token P90 vs. Concurrency

02.0k4.0k6.0k8.0kTTFT P90 (ms)05,00010,00015,000# of queries

Tokens/s per User vs. Concurrency

0100200300Tokens/s per User05,00010,00015,000# of queries
SUT Summary
NVIDIA
NVIDIA GB200 NVL72
Available
System Category Datacenter · GB200 NVL72 rack, disaggregated TRT-LLM, NVFP4 weights
—
1
18
Max Supported Concurrency 16,384
Node Type ID 0
Node Count 18
Software
Framework TensorRT-LLM
TensorRT-LLM
TensorRT-LLM
—
Link to Container —
Operating System unspecified
Filesystem —
Other Software —
Software Notes —
Hardware

Processor

NVIDIA Grace (ARM Neoverse V2)
1
72
VCPUs Per Processor —

Accelerator

NVIDIA GB200
4
—
192 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 —
Run Data
Field NameRun 1Run 2Run 3Run 4Run 5Run 6Run 7Run 8Run 9Run 10Run 11
Run Date06/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/2026
Concurrency1632481282565121,0242,0484,0968,19216,384
System Tokens/Second4,433.17,109.710,640.427,342.846,563.478,501.1125,726.9190,731.6303,719.7373,313.6408,585.4
Tokens/Second per User277.1222.2221.7213.6181.9153.3122.893.174.245.624.9
TTFT P90 (ms)291.4208.2248.2306.01610.41452.01787.62403.54493.73340.67793.8
Utilization1.1%1.7%2.6%6.7%11.4%19.2%30.8%46.7%74.3%91.4%100.0%
Configuration Summarydsr1-gb200x72-pareto-nvl72-c16 · submissiondsr1-gb200x72-pareto-nvl72-c32 · submissiondsr1-gb200x72-pareto-nvl72-c48 · submissiondsr1-gb200x72-pareto-nvl72-c128 · submissiondsr1-gb200x72-pareto-nvl72-c256 · submissiondsr1-gb200x72-pareto-nvl72-c512 · submissiondsr1-gb200x72-pareto-nvl72-c1024 · submissiondsr1-gb200x72-pareto-nvl72-c2048 · submissiondsr1-gb200x72-pareto-nvl72-c4096 · submissiondsr1-gb200x72-pareto-nvl72-c8192 · onlinedsr1-gb200x72-pareto-nvl72-c16384 · online
Config Typesubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissiononlineonline
Config Region———————————
Target QPS———————————
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score———————————
Benchmark Versiond093d67067de9be0420864cfc43e4687511c00be8eunknownc00be8ec00be8e87a9d1cb71f8e9
Result Version0.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.0
Archive URIgs://mlperf-endpoints-blob/runs/8528d1f1-602d-452e-ad7a-ef8b8727e53b/archive.tar.gzgs://mlperf-endpoints-blob/runs/d2b792bf-81ae-46a8-a363-2b7808d107cd/archive.tar.gzgs://mlperf-endpoints-blob/runs/bcc431a5-240e-4271-977e-371bd6ce8f20/archive.tar.gzgs://mlperf-endpoints-blob/runs/0caf637b-c6e9-416b-bd6a-32483c3135c8/archive.tar.gzgs://mlperf-endpoints-blob/runs/561a6d4c-0813-4b7f-8ff7-82a493e79177/archive.tar.gzgs://mlperf-endpoints-blob/runs/905f9bc4-fdcb-42be-b090-3159a494f52e/archive.tar.gzgs://mlperf-endpoints-blob/runs/fc7bde94-556e-4612-a206-0f4d54c83be1/archive.tar.gzgs://mlperf-endpoints-blob/runs/db93541b-51d2-4f78-a9aa-50d49d8b54f7/archive.tar.gzgs://mlperf-endpoints-blob/runs/241c9c6c-4683-498c-9221-652588d95652/archive.tar.gzgs://mlperf-endpoints-blob/runs/2542562e-f00c-457d-8044-5f48a5721148/archive.tar.gzgs://mlperf-endpoints-blob/runs/a59e2194-6710-4f70-8fd8-74a11488229d/archive.tar.gz
QPS1.21791.87752.82337.251512.317620.769733.243050.924383.7246120.4745108.6147
Total Output Tokens15,972,38828,779,87242,964,150105,954,385185,610,169313,328,255500,365,864755,446,1201,105,693,0621,170,066,8071,704,841,539
Average Output Tokens3,640.023,786.833,768.793,770.623,780.253,779.593,782.053,745.393,627.63,196.723,761.79
Run Duration (s)3,602.994,047.954,037.833,875.043,986.183,991.393,979.793,960.783,640.503,134.274,172.55
Total Queries4,3887,60011,40028,10049,10082,900132,300201,700304,800377,600453,200
Total Queries Issued4,3887,60011,40028,10049,10082,900132,300201,700304,800377,600453,200
Total Queries Failed00000000000
Time to First Token (TTFT) (ms)
Minimum159.8139.9137.5141.3139.2135.2148.2159.3166.2191.5164.6
Average241.4172.4187.1232.3632.1583.9716.1909.91382.11620.13508.9
P50216.4153.7161.7200.0294.7272.7383.3376.3381.1494.5452.5
P90291.4208.2248.2306.01610.41452.01787.62403.54493.73340.67793.8
P95349.8245.5276.0323.42250.41881.52229.92881.05416.75601.112684.8
P99419.5285.5319.2373.73396.43039.03395.64013.67535.021605.448314.2
P9994224.92197.12224.93056.34817.83643.45857.18475.016459.533212.062517.8
Maximum9308.82204.72243.13060.76563.34418.76263.99664.218064.335321.664880.0
Time Per Output Token (TPOT) (ms)
Minimum2.441.871.321.711.762.612.482.523.043.8511.9
Average3.644.524.554.685.376.407.9810.513.019.935.8
P503.604.474.514.625.326.347.9310.412.919.936.8
P904.264.995.035.165.947.108.8811.714.422.337.3
P954.455.095.135.276.067.249.0612.014.822.737.4
P994.825.295.305.456.277.519.4012.415.423.637.7
P9995.025.445.525.686.527.829.7812.916.024.737.9
Maximum5.085.775.735.936.988.3510.514.518.0208.038.4
Query Latency (ms)
Minimum906.31064.0997.8890.4983.61041.21422.91796.92075.33234.54535.5
Average13024.916825.116891.917404.020412.024167.630148.739181.247293.965170.9138263.4
P507515.99724.09621.99963.111967.414073.517536.722919.728599.642994.979191.0
P9031124.541347.241400.643024.149435.359266.673984.796538.1119391.7180287.8342153.8
P9546506.159273.059844.960953.970118.383906.2104894.6137745.1169572.0180893.6482314.5
P9965783.383037.584546.887269.599598.1118569.2147865.9180698.8181042.3181270.6729728.1
P99974680.689515.892010.695836.0108046.2129310.4161702.3181199.9181345.8202043.3748546.2
Maximum80212.394733.699769.4118757.1123393.5156836.0181140.1188480.6198294.5215971.3758154.0