Menu
Back previous/next report: k j

NVIDIA NVIDIA GB300 NVL72

Submitted by NVIDIAon 2026-06-25

Submission Info
Submitter NVIDIA
Submission ID 1907bca4-63a0-4b60-ba97-ef47b657d195
Submission Status PUBLISHED
Submission Date 2026-06-25
Publication Date
Division Standardized
Model
Model Name GPT-OSS 120B
Division Standardized
Model Precision
Model Link
Calibration Link
Model Notes
Datasets
Dataset Name perf
Dataset Type performance
Average Input Tokens
Dataset Link

System TPS vs. Tokens/s per User

0200k400k600k800k1.0MSystem TPS100200300400Tokens/s per User

System TPS vs. Concurrency

0200k400k600k800k1.0MSystem TPS05,00010,00015,00020,000# of queries

Time to First Token P95 vs. Concurrency

0246TTFT P95 (s)05,00010,00015,00020,000# of queries

Tokens/s per User vs. Concurrency

100200300400Tokens/s per User05,00010,00015,00020,000# of queries
SUT Summary
NVIDIA
System Name NVIDIA GB300 NVL72
System Availability Available
System Category Datacenter · GB300 NVL72 rack, disaggregated TRT-LLM, MXFP4 weights
System Size
Node Types 1
Total Node Count 18
Max Supported Concurrency 22,528
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 unspecified
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 unspecified
Storage Type unspecified

Networking

Network Interfaces
Network Details 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 11Run 12Run 13Run 14
Run Date06/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/202606/25/2026
Concurrency81735701402725441,0242,0483,8407,16812,28812,80022,528
System Tokens/Second3,226.66,684.712,934.223,078.141,092.771,693.1125,323.6192,455.6303,515.8441,207.9628,875.2778,099.7797,819.5954,230.2
Tokens/Second per User403.3393.2369.5329.7293.5263.6230.4187.9148.2114.987.763.362.342.4
TTFT P95 (ms)156.7112.8119.5128.1190.1885.2568.4877.3782.3799.26115.8752.41025.82472.0
Utilization0.3%0.7%1.4%2.4%4.3%7.5%13.1%20.2%31.8%46.2%65.9%81.5%83.6%100.0%
Configuration Summarygpt-oss-120b-sil_tep8_sys8_ng8_nc8_g1_c1_fe1 · online · c8gpt-oss-120b-sil_tep4_sys17_ng17_nc4_g1_c1_fe1 · online · c17gpt-oss-120b-sil_tep2_sys35_ng35_nc2_g1_c1_fe1 · online · c35gpt-oss-120b-sil_tep2_sys70_ng35_nc2_g2_c1_fe1 · online · c70gpt-oss-120b-sil_tep2_sys140_ng35_nc2_g4_c2_fe1 · online · c140gpt-oss-120b-sil_tep4_sys272_ng17_nc4_g16_c2_fe1 · online · c272gpt-oss-120b-sil_tep4_sys544_ng17_nc4_g32_c4_fe1 · online · c544gpt-oss-120b-sil_tep4_sys1024_ng16_nc8_g64_c2_fe16 · online · c1024gpt-oss-120b-sil_dep8_sys2048_ng8_nc8_g256_c16_fe8 · online · c2048gpt-oss-120b-sil_dep4_sys3840_ng15_nc12_g256_c8_fe15 · online · c3840gpt-oss-120b-sil_dep8_sys7168_ng7_nc16_g1024_c32_fe7 · online · c7168gpt-oss-120b-sil_dep2_sys12288_ng25_nc22_g512_c8_fe25 · online · c12288gpt-oss-120b-sil_dep2_sys12800_ng25_nc22_g512_c8_fe25 · online · c12800gpt-oss-120b-sil_tep1_sys22528_ng48_nc24_g512_c32_fe48 · online · c22528
Config Typeonlineonlineonlineonlineonlineonlineonlineonlineonlineonlineonlineonlineonlineonline
Config Region
Target QPS
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score
Benchmark Version629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc629c1bc
Result Version0.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.00.1.0
Archive URIgs://mlperf-endpoints-blob/runs/cad766b3-9ddc-43f4-b3b7-d3e7fb77682b/archive.tar.gzgs://mlperf-endpoints-blob/runs/f33b91e1-bc12-4ad7-924d-040f209814a1/archive.tar.gzgs://mlperf-endpoints-blob/runs/4abe5c9c-6f73-4932-8bc9-18bb33783b10/archive.tar.gzgs://mlperf-endpoints-blob/runs/88c4775c-259c-4702-a325-2099bbd217b6/archive.tar.gzgs://mlperf-endpoints-blob/runs/d18a2c56-620e-4e13-8789-2ac6e18b1cba/archive.tar.gzgs://mlperf-endpoints-blob/runs/8acf867c-9a7c-4430-8839-1ddf462d057c/archive.tar.gzgs://mlperf-endpoints-blob/runs/8666872a-2d6a-481c-b3eb-d22fa6b39a0b/archive.tar.gzgs://mlperf-endpoints-blob/runs/736ea21d-18f6-437c-9aa1-5777550e2ee4/archive.tar.gzgs://mlperf-endpoints-blob/runs/655970e0-1055-419f-be6b-ccbb11ff87b7/archive.tar.gzgs://mlperf-endpoints-blob/runs/e77820f2-0af4-41da-977e-d6685bbe23d0/archive.tar.gzgs://mlperf-endpoints-blob/runs/4d0be1c5-ead7-47d4-9ce0-8e1fa40bdfcd/archive.tar.gzgs://mlperf-endpoints-blob/runs/20e46720-5c96-4cf0-bd03-a39ca7b985bb/archive.tar.gzgs://mlperf-endpoints-blob/runs/6d2df9ef-ccd1-4331-ab57-37c7058ab993/archive.tar.gzgs://mlperf-endpoints-blob/runs/1c6cf0dd-dcda-4835-9ab3-54475b13d533/archive.tar.gz
QPS2.63765.444210.522718.747533.394358.2258101.8297156.4203246.5849358.7057511.1075632.2840648.4210775.5894
Total Output Tokens2,723,1135,513,08810,798,22719,469,44236,811,32364,381,935112,256,121173,690,797266,528,001394,994,669548,465,566659,982,592666,236,494796,780,859
Average Output Tokens1,223.321,227.861,229.171,2311,230.531,231.291,230.721,230.371,230.881,2301,230.421,230.621,230.41,230.33
Run Duration (s)843.96824.73834.86843.63895.81898.02895.73902.50878.14895.26872.14848.20835.07835.00
Total Queries2,2264,4908,78515,81629,91552,28891,212141,169216,535321,134445,756536,302541,478647,616
Total Queries Issued2,2264,4908,78515,81629,91552,28891,212141,169216,535321,134445,756536,302541,478647,616
Total Queries Failed00000000000000
Time to First Token (TTFT) P95 (ms)
Minimum73.927.427.728.528.430.138.436.936.939.252.952.754.961.1
Average137.787.489.989.9111.6244.1288.4398.4457.6515.52053.4639.6723.4945.8
P50105.462.362.866.686.6101.9237.5333.8390.5458.91303.4471.1493.9371.0
P90135.898.3101.8110.2159.6614.2462.5710.5658.6693.34545.9646.7806.21170.4
P95156.7112.8119.5128.1190.1885.2568.4877.3782.3799.26115.8752.41025.82472.0
P991929.71003.91039.9197.4293.61568.4893.11282.91198.52897.59945.89441.610023.115715.4
P9992176.22346.42472.52752.03122.13782.14688.54140.67052.29497.215657.115539.716258.027075.1
Maximum2262.02435.92546.22875.53798.54022.26165.14936.47892.710793.516244.217152.817855.328043.3
Time Per Output Token (TPOT) (ms)
Minimum1.431.371.301.651.692.162.143.024.304.924.567.397.318.56
Average2.392.462.622.953.303.584.094.986.328.229.5815.115.222.4
P502.392.472.622.953.293.564.084.986.318.269.5815.115.322.5
P902.432.532.733.103.443.734.175.046.458.379.7915.215.423.0
P952.442.562.763.143.483.744.205.056.498.399.8415.315.423.1
P992.462.612.843.243.553.774.245.086.568.459.9515.415.523.3
P9992.502.662.943.363.663.784.295.116.808.489.9915.415.623.5
Maximum2.792.683.164.265.274.374.375.957.3613.618.523.931.436.5
Query Latency (ms)
Minimum1276.01289.01254.71437.91526.91753.52077.32557.7674.13869.44718.66827.51216.97716.8
Average3025.83114.43319.33723.24175.24654.55318.56514.98226.310616.013827.219147.919424.328436.1
P502996.93058.53261.63663.04111.34572.75245.46437.48118.610463.113467.718845.519111.027928.3
P903806.43907.64200.24727.25230.05863.86621.08109.410226.613213.717780.623958.124305.835742.1
P954063.24212.54531.45067.65628.36317.07088.58671.510945.414165.219425.925804.126156.938610.5
P994691.14970.45194.45847.56483.17365.58211.09919.612728.816614.723100.030546.331064.846507.8
P9995543.15828.46168.67236.87879.38578.610185.711864.815836.820800.828044.737447.338081.657631.6
Maximum6144.36781.17099.68216.99591.810540.312881.916847.325519.328471.937424.766361.960524.893108.6