GPU-Insights

GPU Insights, repository for all kinds of code related to GPUs

Default benchmark

python main.py -m -s 512 -e 2 -mt resnet50 -bs 256 -dt FP32 -gpu 0
python main.py -m -s 512 -e 2 -mt resnet50 -bs 256 -dt FP16 -gpu 0

How to understand the results

This performance test evaluates the performance of the hardware device in a training scenario, and the output is a score. Score reflects the unit time to complete a given training task (ResNet50). Thus, a higher score means higher computational performance. Note that the score is affected by both the video memory bandwidth and the PCIe bus bandwidth.

Model Params

23.5 M

Results

For better visual experience, visit this website.