Skip to content

Latest commit

 

History

History
66 lines (48 loc) · 1.73 KB

README.md

File metadata and controls

66 lines (48 loc) · 1.73 KB

Model Benchmarks

Quick Start

cd codewithgpu/benchmarks
python bench_model.py -f ./results.json

For more usages, see "--help" argument:

python bench_model.py --help

Training Baselines

ResNet50

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 61
torch_vm TITAN V FP16 110 661
torch_vm TITAN V FP32 14.9 289

ViT-Base

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 22
torch_vm TITAN V FP16 110 333
torch_vm TITAN V FP32 14.9 86

MobileNetV3

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 85
torch_vm TITAN V FP16 110 1527
torch_vm TITAN V FP32 14.9 878

Inference Baselines

ResNet50

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 214
torch_vm TITAN V FP16 110 2071
torch_vm TITAN V FP32 14.9 940

ViT-Base

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 61
torch_vm TITAN V FP16 110 1033
torch_vm TITAN V FP32 14.9 262

MobileNetV3

Backend Device Prec Perf (TFLOPS) Time (FPS)
torch_vm M1 Pro FP32 4.6 382
torch_vm TITAN V FP16 110 6504
torch_vm TITAN V FP32 14.9 3807