Chris Elrod comments

Repositories
Issues
Comments

Results 832 comments of


                                            Chris Elrod

trafficstars

Suboptimal `dgemm` OpenBLAS performance and dual-socket scaling

> @chriselrod Are you able to reproduce this on some of our machines? Not really on deapsea2, which I believe is 32 cores x 2 sockets: ```julia julia> using LinearAlgebra...

Suboptimal `dgemm` OpenBLAS performance and dual-socket scaling

```julia julia> using LinearAlgebra julia> BLAS.get_num_threads() 64 julia> M = K = N = 16_000; julia> A = Matrix{Float64}(undef, M, K); B = Matrix{Float64}(undef, K, N); C = Matrix{Float64}(undef, M,...