Mikhail Shiryaev

Results 5 comments of Mikhail Shiryaev

run_mkl_igemm_ia.sh - this scripts targets Xeon Phi machine with MCDRAM in flat mode, in that case MCDRAM is recognized as separate numa node and has index "1".

Hi @Tron-x, could you please specify how do you launch IntelCaffe over OpenMPI? As far as I know IntelCaffe uses MLSL only for multi-node communications. MLSL uses Intel MPI under...

@XapaJIaMnu - CCL relies on multiple threads (CCL workers) to parallelize communication of current process. Having each CPU core mapped on CCL rank (not depending whether all ranks within single...

@alexbatashev - could you try again with code from master branch? Compute backend dpcpp_level_zero was renamed to dpcpp, so now COMPUTE_BACKEND=dpcpp is the only possible value. https://oneapi-src.github.io/oneCCL/introduction/installation.html#enable-sycl

@shioyadan - actual execution of store operation happens after signal from "retire" stage, so "--store_completions" helps to see store latency which may be useful for example for cache write policy...