TLB test results - Githubissues

quiver-team / quiver-feature

High performance RDMA-based distributed feature collection component for training GNN model on EXTREMELY large graph

Apache License 2.0

48 stars 5 forks source link

TLB test results #17

Closed ZenoTan closed 2 years ago

ZenoTan commented 2 years ago

256GB feature

Warp-based: 13.3GB/s
Warp-based + sort: 20.2GB/s
Warp-based + sort + multi-kernel: 21.1GB/s
Block-based: 12.4GB/s
Block-based + sort: 17.9GB/s
Block-based + sort + multi-kernel: 22.9GB/s
Upper bound (cudamemcpy): 26.3GB/s

64GB feature

Warp-based: 19.9GB/s
Warp-based + sort: 22.7GB/s
Warp-based + sort + multi-kernel: 22.0GB/s
Block-based: 16.3GB/s
Block-based + sort: 19.4GB/s
Block-based + sort + multi-kernel: 22.2GB/s

16GB feature

Warp-based: 20.2GB/s
Warp-based + sort: 22.1GB/s
Warp-based + sort + multi-kernel: 22.4GB/s
Block-based: 16.8GB/s
Block-based + sort: 19.5GB/s
Block-based + sort + multi-kernel: 22.3GB/s