nicejunjie / scilib-accel

automatic GPU offload for scientific libraries
MIT License
10 stars 0 forks source link

babelstream example #1

Open jinz2014 opened 4 months ago

jinz2014 commented 4 months ago

Could you please put Babelstream examples in the repository so that people may understand how CPU/GPU select DDR or HBM accesses described in the paper ? Thanks.

nicejunjie commented 4 months ago

I may do that later. But here is the data I measured on Grace-Hopper (120GB LPDDR5x model). If you'd like to measure the GPU to LPDDR5 bandwidth, use this code: https://github.com/nicejunjie/CUDA-Cpp-STREAM

Screenshot 2024-07-16 at 11 06 57 AM