Menu
Back previous/next report: k j

Google GB300-NVL72_GB300-288GB_aarch64x72_TRT

Submitted by Googleon 2026-06-26

Submission Info
Submitter Google
Submission ID 60a63420-d1d1-41d7-b807-61921f313e8f
Submission Status PUBLISHED
Submission Date 2026-06-26
Publication Date
Division Standardized
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 TPS100200300Tokens/s per User

System TPS vs. Concurrency

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

Time to First Token P95 vs. Concurrency

0246TTFT P95 (s)02,0004,0006,0008,000# of queries

Tokens/s per User vs. Concurrency

100200300Tokens/s per User02,0004,0006,0008,000# of queries
SUT Summary
NVIDIA
System Name GB300-NVL72_GB300-288GB_aarch64x72_TRT
System Availability Available
System Category Datacenter · GB300 NVL72 rack, disaggregated TRT-LLM, NVFP4 weights, fp8 KV cache; MTP-3 speculative decoding
System Size
Node Types 1
Total Node Count 18
Max Supported Concurrency 8,192
Node Type ID 0
Node Count 18
Software
Framework TensorRT-LLM
Serving Framework TensorRT-LLM
Inference Backend TensorRT-LLM
Driver
Link to Container
Operating System
Filesystem
Other Software
Software Notes
Hardware

Processor

Processor Model Name NVIDIA Grace (ARM Neoverse V2)
Processors per Node 1
Cores Per Processor 72
VCPUs Per Processor

Accelerator

Accelerator Model Name NVIDIA GB300
Accelerators per Node 4
Memory Type
Memory Capacity 288 GB HBM3e
Accelerator Interconnect NVLink 5 + NVSwitch (NVL72)
Host-Accelerator Interconnect NVLink-C2C (Grace-Blackwell)

Host / Storage

Host Memory Capacity unspecified
Memory Configuration
Storage Capacity 960 GB Instance Memory, Attached 12,000 GB LSSD
Storage Type

Networking

Network Interfaces
Network Details NVLink + RoCEv2
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 9
Run Date06/26/202606/26/202606/26/202606/26/202606/26/202606/26/202606/26/202606/26/202606/26/2026
Concurrency448642565121,0242,0484,0968,192
System Tokens/Second1,234.110,864.614,294.544,857.882,717.2132,474.9200,860.2325,843.1408,369.4
Tokens/Second per User308.5226.3223.4175.2161.6129.498.179.649.8
TTFT P95 (ms)192.6281.8292.02465.6309.2322.61428.52908.76380.1
Utilization0.3%2.7%3.5%11.0%20.3%32.4%49.2%79.8%100.0%
Configuration Summarydsr1-gb300x72-pareto-nvl72-c4 · submissiondsr1-gb300x72-pareto-nvl72-c48 · submissiondsr1-gb300x72-pareto-nvl72-c64 · submissiondsr1-gb300x72-pareto-nvl72-c256 · submissiondsr1-gb300x72-pareto-nvl72-c512 · submissiondsr1-gb300x72-pareto-nvl72-c1024 · submissiondsr1-gb300x72-pareto-nvl72-c2048 · submissiondsr1-gb300x72-pareto-nvl72-c4096 · submissiondsr1-gb300x72-pareto-nvl72-c8192 · submission
Config Typesubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmission
Config Region
Target QPS
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score
Benchmark Versionunknownunknownunknownunknownunknownunknownunknownunknownunknown
Result Version0.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.0
Archive URIgs://mlperf-endpoints-blob/runs/4c107b18-9410-4f80-90ce-88f9e75417d4/archive.tar.gzgs://mlperf-endpoints-blob/runs/0d2b7b67-4ea7-43ee-9e37-54ff64cc5941/archive.tar.gzgs://mlperf-endpoints-blob/runs/ee9d8ca8-ec42-4daa-b44b-8de26e089410/archive.tar.gzgs://mlperf-endpoints-blob/runs/03c2178d-6e91-4a2f-b69e-5fae60578750/archive.tar.gzgs://mlperf-endpoints-blob/runs/afd13761-b8ad-476d-9da5-ade649716249/archive.tar.gzgs://mlperf-endpoints-blob/runs/06cccf32-a58a-447a-b76e-be5ff31e3f87/archive.tar.gzgs://mlperf-endpoints-blob/runs/cecfc4d8-b47d-4190-9516-df3ee8f2591a/archive.tar.gzgs://mlperf-endpoints-blob/runs/a817ab19-0489-41a5-b6ab-22745a8fb213/archive.tar.gzgs://mlperf-endpoints-blob/runs/d17716a0-752a-4757-b484-4848c0038050/archive.tar.gz
QPS0.32962.90083.813911.925122.009135.197553.705588.7754108.7729
Total Output Tokens3,744,63424,652,15138,229,52182,529,856280,355,359396,368,631722,095,5901,046,877,1851,317,920,127
Average Output Tokens3,744.633,745.393,747.993,761.623,758.323,763.763,740.033,670.423,754.33
Run Duration (s)3,034.342,269.042,674.431,839.813,389.322,992.033,595.023,212.833,227.27
Total Queries1,0006,58210,20021,94074,596105,312193,072285,220351,040
Total Queries Issued1,0006,58210,20021,94074,596105,312193,072285,220351,040
Total Queries Failed000000000
Time to First Token (TTFT) P95 (ms)
Minimum139.8147.5149.5148.3150.3151.9162.7174.3189.5
Average206.6199.1214.8758.1328.0368.3497.2838.31547.7
P50160.8163.7164.1296.6233.9261.3298.9337.3345.1
P90176.8257.8268.61943.7297.3308.51054.42064.94171.9
P95192.6281.8292.02465.6309.2322.61428.52908.76380.1
P992136.2337.3511.73977.7327.71284.42345.67045.218417.6
P9992371.82055.44166.04306.116029.115314.86665.315484.227085.2
Maximum2436.12056.04167.04999.217199.215483.97667.516800.928647.5
Time Per Output Token (TPOT) (ms)
Minimum0.412.942.921.974.084.354.663.354.81
Average3.284.394.455.426.127.6210.012.219.1
P503.274.344.375.376.057.559.9412.119.0
P903.774.854.906.016.778.4611.213.521.5
P953.954.955.066.126.908.6211.413.821.9
P994.205.246.026.347.148.9211.814.322.7
P9994.365.817.476.627.449.2712.314.923.6
Maximum4.376.209.326.837.8814.913.216.325.7
Query Latency (ms)
Minimum790.0819.4790.31060.31062.71323.91741.01967.62410.9
Average12125.816222.816412.720639.622731.928290.636986.244431.771347.8
P507189.09490.79368.612191.513182.716371.921314.726406.041386.9
P9028816.640134.140504.149976.555804.969551.491463.5111636.5175435.5
P9542836.356322.057573.270677.179668.199637.5130027.8158720.9248289.1
P9959022.079539.880420.6100511.6112812.6140778.9180455.0180980.1354392.3
P99967050.187533.588910.9109949.5122736.7153037.5181135.6181285.6388915.5
Maximum75437.593245.3100102.6134660.7136603.6180878.3184593.9196969.6500784.6