Back previous/next report: k j

Google GB300-NVL72_GB300-288GB_aarch64x72_TRTLLM

Submitted by Google LLCon 2026-10-16 Published on 2026-11-04

Result statement

Google LLC ran DeepSeek-R1 on GB300-NVL72_GB300-288GB_aarch64x72_TRTLLM and achieved a verified 395,546.7 tokens/second System TPS at 3,429.2ms TTFT P90 in the Standardized division of the Available category on November 4, 2026 .

[1] MLPerf Endpoints v1.0 DeepSeek-R1, published November 4, 2026; retrieved from results.mlcommons.org on November 18, 2026. Result —, run 10: concurrency = 8,192 queries, TTFT P90 = 3,429.2ms, System TPS = 395,546.7 tokens/sec, interactivity = 48.3 tokens/sec per user and total provisioned system power = 151.82 kW (MLC Estimated Power). 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
Google LLC
a4f7df58-6d33-4ebd-8ca1-164bc8e7355d
Submission Status PUBLISHED
2026-10-16
2026-11-04
Version v1.0
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 TPS50100150200Tokens/s per User

System TPS vs. Concurrency

0100k200k300k400kSystem TPS02,0004,0006,0008,000# of queries

Time to First Token P90 vs. Concurrency

1.0k2.0k3.0kTTFT P90 (ms)02,0004,0006,0008,000# of queries

Tokens/s per User vs. Concurrency

50100150200Tokens/s per User02,0004,0006,0008,000# of queries
SUT Summary
NVIDIA
GB300-NVL72_GB300-288GB_aarch64x72_TRTLLM
Available
System Category Datacenter · —
18 nodes x 4 NVIDIA GB300 (72 GPUs, one NVL72 domain)
1
18
Max Supported Concurrency 8,192
Node Type ID 0
Node Count 18
Software
Framework TensorRT-LLM 1.3.0rc14 (disaggregated prefill/decode)
TensorRT-LLM 1.3.0rc14 (disaggregated prefill/decode)
TensorRT-LLM
NVIDIA 580.173.02 (GKE 'default' COS driver; CUDA 13.0 driver API)
Link to Container nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc14
Operating System Container-Optimized OS, kernel 6.12.94+, GKE 1.35.8-gke.1036000
Filesystem ext4 on node-local NVMe SSD (/mnt/stateful_partition/kube-ephemeral-ssd)
Other Software MLCommons endpoints client ghcr.io/mlcommons/endpoints:e9f427c; containerd 2.2.7
Software Notes —
Hardware

Processor

NVIDIA Grace (Arm Neoverse V2)
0
—
VCPUs Per Processor 144

Accelerator

—
0
—
—
—
—

Host / Storage

~943 GiB per node (kubectl capacity 988634624Ki)
Memory Configuration 960 GB LPDDR5X (2x NVIDIA Grace CPU integrated memory)
Storage Capacity ~12.5 TB local SSD per node (kubectl ephemeral-storage)
Storage Type —

Networking

4x NVIDIA ConnectX-7 + 1x Google Cloud Titanium NIC (a4x-maxgpu-4g-metal)
Google Cloud Titanium / RoCE + NVLink 5 (NVL72)
Network Topology —

Other

Other Hardware —
Cooling Liquid-cooled
Hardware Notes —
Run Data
Field NameRun 1Run 2Run 3Run 4Run 5Run 6Run 7Run 8Run 9Run 10
Run Date10/16/202610/16/202610/16/202610/16/202610/16/202610/16/202610/16/202610/16/202610/16/202610/16/2026
Concurrency3248641282565121,0242,0484,0968,192
System Tokens/Second6,794.510,334.813,605.527,256.045,933.182,685.7131,440.4198,873.0308,595.0395,546.7
Tokens/Second per User212.3215.3212.6212.9179.4161.5128.497.175.348.3
TTFT P90 (ms)360.4324.0355.2358.21613.9355.3366.81021.82046.53429.2
Utilization1.7%2.6%3.4%6.9%11.6%20.9%33.2%50.3%78.0%100.0%
Configuration Summarydeepseek-r1-c32 · submissiondeepseek-r1-c48 · submissiondeepseek-r1-c64 · submissiondeepseek-r1-c128 · submissiondeepseek-r1-c256 · submissiondeepseek-r1-c512 · submissiondeepseek-r1-c1024 · submissiondeepseek-r1-c2048 · submissiondeepseek-r1-c4096 · submissiondeepseek-r1-c8192 · submission
Config Typesubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmission
Config Region——————————
Target QPS——————————
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score——————————
Benchmark Versionunknownunknownunknownunknownunknownunknownunknownunknownunknownunknown
Result Version0.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.0
Archive URIgs://mlperf-endpoints-blob/runs/99b6ffe0-a127-4bcf-ad86-91f19efa8f0b/archive.tar.gzgs://mlperf-endpoints-blob/runs/6dfdcc49-1a05-4adf-baea-a2e21f62bc66/archive.tar.gzgs://mlperf-endpoints-blob/runs/c3527453-8f0d-4ea2-8562-e1bfdc7fa983/archive.tar.gzgs://mlperf-endpoints-blob/runs/84692721-437d-49d0-a56d-6f5fda66876c/archive.tar.gzgs://mlperf-endpoints-blob/runs/9e872a5d-f552-4cf0-bfd7-5a0d512e44fe/archive.tar.gzgs://mlperf-endpoints-blob/runs/f67dfbbb-de0c-4d96-8f21-68d80197261a/archive.tar.gzgs://mlperf-endpoints-blob/runs/a809fb6a-f9b3-4aa4-9ca7-837556545e05/archive.tar.gzgs://mlperf-endpoints-blob/runs/12ff0074-6884-4601-bf05-d77451c786db/archive.tar.gzgs://mlperf-endpoints-blob/runs/592d4bd6-c2a4-4e93-8058-32b8b595fb6f/archive.tar.gzgs://mlperf-endpoints-blob/runs/b1546dfc-4f56-4d71-8b26-3417b513dc0f/archive.tar.gz
QPS1.78232.69953.52397.128912.046821.576234.402351.890280.6346103.2341
Total Output Tokens33,455,67067,195,98767,766,477167,768,13283,654,953285,872,100402,364,416739,962,4801,091,559,3571,345,028,357
Average Output Tokens3,812.183,828.393,860.93,823.343,812.93,832.273,820.693,832.573,827.083,831.57
Run Duration (s)4,923.946,501.934,980.836,155.271,821.233,457.333,061.193,720.783,537.193,400.43
Total Queries8,77617,55217,55243,88021,94074,596105,312193,072285,220351,040
Total Queries Issued8,77617,55217,55243,88021,94074,596105,312193,072285,220351,040
Total Queries Failed0000000000
Time to First Token (TTFT) (ms)
Minimum159.2154.0155.1149.0154.7154.5159.4170.0187.6198.4
Average262.0232.9248.2251.2630.1284.1315.4508.7921.81867.1
P50245.9207.6226.7232.7298.9275.9286.6326.0377.2421.7
P90360.4324.0355.2358.21613.9355.3366.81021.82046.53429.2
P95408.0390.4410.5393.72137.4376.4443.31372.02840.66783.7
P99474.3461.0474.6450.92935.0442.3544.72698.29077.626000.1
P9991431.21457.31500.11609.33709.92678.44324.17379.619412.644812.5
Maximum1595.21526.91576.51675.74331.12739.94454.88283.222484.645376.2
Time Per Output Token (TPOT) (ms)
Minimum3.072.973.023.081.043.904.723.401.902.92
Average4.724.664.724.715.366.157.6810.112.920.1
P504.674.624.684.655.306.087.6110.112.819.1
P905.235.195.245.215.946.818.5411.314.421.7
P955.355.315.365.316.066.958.7111.514.722.3
P995.555.555.585.516.277.219.0211.915.225.5
P9995.845.835.875.806.587.569.4112.415.846.2
Maximum6.336.306.066.566.888.3810.713.717.766009.2
Query Latency (ms)
Minimum943.2927.31082.7893.7942.31188.51249.81747.31996.02908.5
Average17791.217588.717959.217767.520516.123235.328900.238371.348866.074096.6
P5010442.310270.210394.810437.812107.213544.616855.122404.428561.843387.7
P9043730.843293.943901.543797.249823.557143.071246.494309.0120546.1182669.8
P9561603.060583.663181.762008.170996.180864.2100591.9133632.2170185.8258122.7
P9986705.985608.988121.186741.399290.0114053.4141904.2188407.4239870.2360564.6
P99994401.993752.795104.993797.0107414.3123175.1153695.0204258.2260094.7399387.6
Maximum102175.6102749.1107059.5108357.8120524.8148225.0182501.1252894.2331181.4491556.2