Benchmarks Cloudlab Server (2 socket)

Date Revision ResNet50 Inception GPT2 Bert Comments
22-08-2022 large-batches ResNet50 Inception GPT2 Bert Eager vs graph for PT
29-09-2022 large-batches-tcmalloc ResNet50 Inception GPT2 Bert Default vs tcmalloc for PT
19-12-2022 pytorch-ht ResNet50 Bert Cores vs HT for PT
21-11-2022 torchserve ResNet50 TS with inefficient pre-processing
21-11-2022 torchserve-opt ResNet50 Bert TS with efficient pre-processing + TS vs PT
17-01-2023 torchserve-allocators ResNet50 Inception GPT2 Bert Default vs tcmalloc for TS
03-04-2023 pytorch-granular ResNet50 Inception GPT2 PT with granular BS (1-32 & ^2)

Benchmarks Fiddle Server (2 socket)

Date Revision ResNet50 Inception GPT2 Bert Comments (for PT)
24-06-2022 efb6225 ResNet50 Eager vs channel last vs IPEX
29-06-2022 6eb235b ResNet50 Inter vs intra heatmaps
08-07-2022 cdf1d77 ResNet50 Inception GPT2 Bert Eager vs graph
18-07-2022 numa ResNet50 Threads=1 to 36
20-07-2022 numa-interleave ResNet50 Interleaved thread mapping
20-07-2022 numa-interleave-no-omp ResNet50 Interleaved thread mapping without OMP flags
28-07-2022 flops ResNet50 FLOPS vs batch size
29-07-2022 large-batches ResNet50 Inception GPT2 Bert BS=1 to 1024 + opt=graph + interleave vs sequential
10-08-2022 local-vs-remote-mem ResNet50 Local vs remote memory access
17-08-2022 thread-assignment ResNet50 OMP vs numactl vs none for thread assignment