Menu
Back previous/next report: k j

Intel 4x Arc(TM) Pro B70

Submitted by Intelon 2026-06-26

Submission Info
Submitter Intel
Submission ID 482a9d35-4e93-4643-b6af-d5dac03316cd
Submission Status PUBLISHED
Submission Date 2026-06-26
Publication Date
Division Standardized
Model
Model Name Llama 3 8B
Division Standardized
Model Precision FP16
Calibration Link GitHub
Model Notes W4A8 quantization with vLLM's quantization tools, with fp8 kv cache. No other model transformations applied.
Datasets
Dataset Name CNN/DailyMail
Dataset Type performance
Average Input Tokens 870
Dataset Link Hugging Face

System TPS vs. Tokens/s per User

1.0k2.0k3.0k4.0k5.0kSystem TPS010203040Tokens/s per User

System TPS vs. Concurrency

1.0k2.0k3.0k4.0k5.0kSystem TPS05001,0001,500# of queries

Time to First Token P95 vs. Concurrency

05101520TTFT P95 (s)05001,0001,500# of queries

Tokens/s per User vs. Concurrency

010203040Tokens/s per User05001,0001,500# of queries
SUT Summary
Intel
System Name 4x Arc(TM) Pro B70
System Availability Available
System Category Datacenter · —
System Size 4x Intel(R) Arc(TM) Pro B70 Graphics
Node Types 1
Total Node Count 1
Max Supported Concurrency 1,536
Node Type ID 0
Node Count 1
Software
Ubuntu
Framework vLLM 0.1.dev11435+gfa77e1a63
Serving Framework vLLM 0.1.dev11435+gfa77e1a63
Inference Backend Not detected: CUDA/ROCm/XPU runtime not found
Driver Not available
Link to Container
Operating System linux debian ubuntu 24.04
Filesystem ext4 vfat
Other Software
Software Notes
Hardware

Processor

Processor Model Name Intel(R) Xeon(R) 698X
Processors per Node 1
Cores Per Processor 86
VCPUs Per Processor 172

Accelerator

Accelerator Model Name Intel(R) Arc(TM) Pro B70 Graphics
Accelerators per Node 4
Memory Type GDDR6
Memory Capacity 32GiB
Accelerator Interconnect Not available
Host-Accelerator Interconnect PCIe 5.0 x16

Host / Storage

Host Memory Capacity 134G
Memory Configuration 8 x 16GB DDR5-6400
Storage Capacity 1.8 TB NVMe SSD
Storage Type NVMe SSD

Networking

Network Interfaces 3x Ethernet
Network Details Ethernet
Network Topology

Other

Other Hardware
Cooling
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
Concurrency163240641282565121,0241,536
System Tokens/Second693.11,281.31,570.22,228.33,446.54,248.84,549.44,613.24,612.3
Tokens/Second per User43.340.039.334.826.916.68.94.53.0
TTFT P95 (ms)192.1213.7222.7249.9289.8433.9773.35946.019687.3
Utilization15.0%27.8%34.0%48.3%74.7%92.1%98.6%100.0%100.0%
Configuration Summaryllama-3.1-8b-concurrency · submission · c16llama-3.1-8b-concurrency · submission · c32llama-3.1-8b-concurrency · submission · c40llama-3.1-8b-concurrency · submission · c64llama-3.1-8b-concurrency · submission · c128llama-3.1-8b-concurrency · submission · c256llama-3.1-8b-concurrency · submission · c512llama-3.1-8b-concurrency · submission · c1024llama-3.1-8b-concurrency · submission · c1536
Config Typesubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmissionsubmission
Config Region
Target QPS35.000035.000035.000035.000035.000035.000035.000035.000035.0000
Result Statecompletecompletecompletecompletecompletecompletecompletecompletecomplete
Measured Accuracy Score
Benchmark Version7251d707251d707251d707251d707251d707251d707251d707251d707251d70
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/6b8a596c-c7b0-4aa4-b3bb-5e21ecf513e5/archive.tar.gzgs://mlperf-endpoints-blob/runs/15f2b51a-7273-4481-8832-afe6c9679aae/archive.tar.gzgs://mlperf-endpoints-blob/runs/1f9661d8-b56d-4f3a-b590-3a2a42eae969/archive.tar.gzgs://mlperf-endpoints-blob/runs/5d828820-2e60-4025-a443-0a22b2f1694d/archive.tar.gzgs://mlperf-endpoints-blob/runs/14fafa04-daf1-4b13-b003-71291f5310ce/archive.tar.gzgs://mlperf-endpoints-blob/runs/f14bf3bb-3659-4e2d-aa16-6e5712310d46/archive.tar.gzgs://mlperf-endpoints-blob/runs/5b8cc8de-8990-4db3-9300-ad974d418f8b/archive.tar.gzgs://mlperf-endpoints-blob/runs/074db4fd-49d0-43ed-99f1-b50a92b4718c/archive.tar.gzgs://mlperf-endpoints-blob/runs/f5d6ff9f-c7dc-4f65-98c1-859a5e2ef09c/archive.tar.gz
QPS5.414910.010312.267217.409026.925633.194035.542036.040736.0337
Total Output Tokens1,711,1021,711,1021,711,1021,711,1023,422,2043,422,2043,422,2043,422,2043,422,204
Average Output Tokens128128128128128128128128128
Run Duration (s)2,468.731,335.421,089.74767.88992.96805.45752.24741.83741.97
Total Queries13,36813,36813,36813,36826,73626,73626,73626,73626,736
Total Queries Issued13,36813,36813,36813,36826,73626,73626,73626,73626,736
Total Queries Failed000000000
Time to First Token (TTFT) P95 (ms)
Minimum51.052.151.751.855.556.068.3119.7118.1
Average114.7121.0122.3136.0160.4251.3473.22449.113772.5
P50106.9110.9111.1120.4135.3209.5350.81684.413808.5
P90175.7183.1187.4209.8247.4367.8630.85124.119041.5
P95192.1213.7222.7249.9289.8433.9773.35946.019687.3
P99251.7280.2292.0340.2429.5784.84918.518034.828677.5
P999333.4534.1745.81162.72103.14612.710059.223257.936133.6
Maximum1219.3864.01002.41525.42652.75404.011239.825102.538841.1
Time Per Output Token (TPOT) (ms)
Minimum20.220.420.420.822.122.525.130.438.4
Average22.324.224.727.836.158.6109.4203.3222.5
P5022.324.124.527.735.957.3104.0224.7226.2
P9023.525.826.530.240.573.0149.1229.5229.8
P9523.926.427.131.041.977.5164.9230.0230.4
P9924.727.428.332.544.886.8183.1230.9231.2
P99926.228.729.734.047.693.5197.6231.9231.6
Maximum35.729.630.836.450.097.6200.6232.4232.5
Query Latency (ms)
Minimum2627.62665.02689.62787.32977.42964.93491.54337.111227.7
Average2953.13193.73255.93668.94746.57695.114370.028269.742030.5
P502942.83184.13237.93654.14707.97520.313644.330227.142515.9
P903105.83414.13500.13985.95324.19586.419728.433711.347536.2
P953162.23489.23585.24100.05518.010168.321771.034600.848150.4
P993279.13619.43753.44317.15963.411442.624018.743136.256803.3
P9993492.33810.83982.04681.36629.712275.026147.850903.364281.4
Maximum4676.83893.74313.85106.87393.412694.026872.553271.366466.7