Back previous/next report: k j

NVIDIA NVIDIA Vera Rubin NVL72 (4-GPU SUT)

Submitted by NVIDIAon 2026-10-17 Published on 2026-10-19

Result statement

NVIDIA ran Qwen3.6 35B-A3B on NVIDIA Vera Rubin NVL72 (4-GPU SUT) and achieved a provisional 64,242.7 tokens/second System TPS at 329.1ms TTFT P90 in the Standardized division of the Available category on October 19, 2026. Pending MLCommons Association verification, results are subject to change pending peer review. .

[1] MLPerf Endpoints v1.0 Qwen3.6 35B-A3B, published October 19, 2026; retrieved from results.mlcommons.org on November 18, 2026. Result —, run 11: concurrency = 1,024 queries, TTFT P90 = 329.1ms, System TPS = 64,242.7 tokens/sec, interactivity = 62.7 tokens/sec per user and total provisioned system power = 12.61 kW. 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
03381a91-8b69-4aa2-bf75-0b6dd7730ad8
Submission Status REVIEW_PENDING
2026-10-17
2026-10-19
Version v1.0
Division Standardized
Deployment Client on Premises
Model
Model Name Qwen3.6 35B-A3B
Division Standardized
Model Precision —
Model Link —
Calibration Link —
Model Notes —
Datasets
Dataset Name agentic_coding
Dataset Type performance
Average Input Tokens —
Dataset Link —

System TPS vs. Tokens/s per User

20k40k60kSystem TPS50100150200Tokens/s per User

System TPS vs. Concurrency

20k40k60kSystem TPS02004006008001,000# of queries

Time to First Token P90 vs. Concurrency

100200300TTFT P90 (ms)02004006008001,000# of queries

Tokens/s per User vs. Concurrency

50100150200Tokens/s per User02004006008001,000# of queries
SUT Summary
NVIDIA
NVIDIA Vera Rubin NVL72 (4-GPU SUT)
Available
System Category — · —
4 accelerators
1
1
Max Supported Concurrency 1,024
Node Type ID 0
Node Count 1
Software
Framework vLLM:d48687fdeedd
vLLM:d48687fdeedd
vLLM 0.1.dev1+gd48687fde.d20261008.cu134
620
Link to Container qwen3.6-vllm-dynamo@sha256:f56c2ac713a2228372afcf295dc102c29d07609462a2a9e734033a5f4223f7e8
Operating System unspecified
Filesystem Shared Lustre for persistent files; per-allocation RAID for model and compilation scratch
Other Software Dynamo 1.4.2; torch 2.15.0.dev20260908+cu134; transformers 5.19.0
Software Notes vLLM (YibaiMeng/vllm), branch mlperf-end-multiturn-v1.0-qwen-v2 (d48687fd), behind Dynamo 1.4.2
Hardware

Processor

NVIDIA Vera CPU
2
176
VCPUs Per Processor 352

Accelerator

—
0
—
—
—
—

Host / Storage

unspecified
Memory Configuration unspecified
Storage Capacity unspecified
Storage Type unspecified

Networking

unspecified
NVLink + InfiniBand
Network Topology —

Other

Other Hardware —
Cooling Liquid-cooled
Hardware Notes —
Run Data
Field NameRun 1Run 2Run 3Run 4Run 5Run 6Run 7Run 8Run 9Run 10Run 11
Run Date10/16/202610/16/202610/16/202610/16/202610/17/202610/17/202610/17/202610/17/202610/17/202610/17/202610/17/2026
Concurrency3236641281441922563203845121,024
System Tokens/Second6,278.76,975.411,803.121,687.624,012.430,115.436,753.642,840.947,245.855,038.564,242.7
Tokens/Second per User196.2193.8184.4169.4166.8156.9143.6133.9123.0107.562.7
TTFT P90 (ms)47.546.850.754.368.877.392.8106.3117.1145.5329.1
Utilization9.8%10.9%18.4%33.8%37.4%46.9%57.2%66.7%73.5%85.7%100.0%
Configuration Summaryqwen-agentic-benchmark · online · c32qwen-agentic-benchmark · online · c36qwen-agentic-benchmark · online · c64qwen-agentic-benchmark · online · c128qwen-agentic-benchmark · online · c144qwen-agentic-benchmark · online · c192qwen-agentic-benchmark · online · c256qwen-agentic-benchmark · online · c320qwen-agentic-benchmark · online · c384qwen-agentic-benchmark · online · c512qwen-agentic-benchmark · online · c1024
Config Typeonlineonlineonlineonlineonlineonlineonlineonlineonlineonlineonline
Config Region———————————
Target QPS———————————
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score———————————
Benchmark Versionunknownunknownunknownunknownunknownunknownunknownunknownunknownunknownunknown
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/a668253d-17ca-4063-9dc0-a531eafdc3ec/archive.tar.gzgs://mlperf-endpoints-blob/runs/284033f9-aa4a-4ef1-bc72-26e58399b692/archive.tar.gzgs://mlperf-endpoints-blob/runs/63dcf477-1081-4309-b496-2d87f7180580/archive.tar.gzgs://mlperf-endpoints-blob/runs/5daa793c-107f-4c60-9ee4-7d4d08444f7b/archive.tar.gzgs://mlperf-endpoints-blob/runs/ebbc8092-a4a9-41f3-a002-266cad21508d/archive.tar.gzgs://mlperf-endpoints-blob/runs/19dd7a48-485b-4fcd-bef7-349dec07396c/archive.tar.gzgs://mlperf-endpoints-blob/runs/eaa3a07e-8784-441b-9a36-16609cfa5fe0/archive.tar.gzgs://mlperf-endpoints-blob/runs/a2d295e7-df30-4315-8624-b50e6c2edea2/archive.tar.gzgs://mlperf-endpoints-blob/runs/a398f3af-f15a-431b-98a2-b9067f1390e5/archive.tar.gzgs://mlperf-endpoints-blob/runs/1a26a662-7409-4094-88d0-572346c9eb81/archive.tar.gzgs://mlperf-endpoints-blob/runs/c9cf7be2-6377-4180-a86f-7563c6db9821/archive.tar.gz
QPS16.793918.764531.440058.029863.909680.806898.7019114.3206126.1757147.5071171.9494
Total Output Tokens30,328,53030,134,64645,321,83682,596,70082,782,50981,605,44196,042,024118,460,181124,884,638145,212,571172,998,588
Average Output Tokens373.87371.73375.42373.73375.72372.68372.37374.74374.44373.12373.61
Run Duration (s)4,830.394,320.133,839.823,808.483,447.492,709.762,613.132,765.122,643.302,638.382,692.89
Total Queries81,12181,065120,724221,005220,328218,967257,921316,110333,520389,180463,041
Total Queries Issued81,12181,065120,724221,005220,328218,967257,921316,110333,520389,180463,041
Total Queries Failed00000000000
Time to First Token (TTFT) (ms)
Minimum14.113.614.115.415.216.718.320.020.121.234.4
Average33.933.437.340.050.856.767.476.583.9103.2227.3
P5031.330.734.837.248.754.364.673.480.399.0201.0
P9047.546.850.754.368.877.392.8106.3117.1145.5329.1
P9553.953.456.861.877.387.1104.3119.3131.2162.5379.3
P9977.376.981.089.8108.1122.8143.8162.7178.5217.7571.7
P999146.2149.0168.1208.6248.2315.1379.3377.3440.5502.26294.0
Maximum366.4371.6387.0522.0548.8591.0693.3810.0919.51144.310026.6
Time Per Output Token (TPOT) (ms)
Minimum-0.000.100.000.350.010.000.040.040.690.992.31
Average0.860.911.171.651.742.062.603.113.734.8010.9
P500.800.841.101.531.621.922.432.893.524.5110.1
P901.181.241.622.242.352.803.524.224.876.2415.2
P951.311.371.792.442.563.053.844.595.256.7316.9
P991.571.642.102.832.983.544.445.285.977.6620.4
P9992.062.122.733.773.974.635.656.667.699.5050.9
Maximum215.6349.0374.5632.3533.1582.8896.41515.32236.43191.83804.5
Query Latency (ms)
Minimum32.228.829.329.042.146.163.376.774.161.4157.2
Average343.2355.4463.2631.9675.4794.6999.51198.01430.61824.84154.4
P50128.8132.0166.4219.0240.9282.2351.1415.6486.7619.21389.8
P90837.0884.21151.21587.31680.31992.52507.73000.43549.94544.310446.1
P951317.31367.71809.42503.12653.43132.23948.54768.85683.77336.116759.1
P992847.12929.63879.35440.65757.46731.28490.310275.412636.315971.236812.6
P9995881.76099.78093.711184.011644.613696.317531.521153.025718.232795.976708.3
Maximum9602.010107.612411.118581.621327.423708.329378.634035.739109.950350.5160259.0