Intel perf table: ResNet-50 v1.5 INT8 on Flex 170, OpenVINO, 3297 img/s (b1)
3297.0 inferences/s
Fields
id
benchmarks/intel-perf-flex170-resnet50-int8-b1Metric
tps_aggregate
Value
3,297
Unit
inferences/s
Model
ResNet-50 v1.5 (ImageNet2013)
Methodology
VENDOR-MEASURED (Intel AI performance table, measured 2023-10-24). 1x Intel Data Center GPU Flex 170 on Xeon Platinum 8360Y 2S-36C-72T host, 128GB DDR4-3200, Ubuntu 22.04. OpenVINO 2023.2, INT8 precision. Unit = images/sec (inferences/sec), NOT tokens - CV inference rows for the uncovered accel record. Batch size 1.
Measured by
vendor
notes
Batch-1 latency-class row in the same vendor table. Vision/generation rows - first measured coverage for this part.
Notes
Batch-1 latency-class row in the same vendor table. Vision/generation rows - first measured coverage for this part.