Skinny matrix speedup in magma_zgemm and magma_zgetrs_gpu

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Post Reply
yvirgus
Posts: 1
Joined: Fri Jun 05, 2015 4:44 pm

Skinny matrix speedup in magma_zgemm and magma_zgetrs_gpu

Post by yvirgus » Fri Jun 05, 2015 5:15 pm

Hi,

I am trying to perform matrix multiplications with magma_zgemm and solving linear equations with magma_zgetrs_gpu for long skinny matrix. However the speedup is significantly reduced (especially for magma_zgetrs_gpu). Here's the timing info for each run

magma_zgemm

transA = No transpose, transB = No transpose
M N K MAGMA Gflop/s (ms) cuBLAS Gflop/s (ms) CPU Gflop/s (ms) MAGMA error cuBLAS error
===============================================================================================
25 25 200000 15.63 ( 63.97) 110.84 ( 9.02) 5.99 ( 166.96) 2.50e-14 1.76e-14 failed (slow)
2112 2112 2112 340.97 ( 221.03) 340.62 ( 221.26) 7.98 (9447.92) 1.67e-15 1.67e-15 ok

magma_zgetrs_gpu

M NRHS MAGMA Gflop/s (s) CPU Gflop/s (s)
============================================
25 301088 9.74 ( 0.266) 3.70 ( 0.402) (really slow)
5184 5184 285.63 ( 4.41) 7.89 ( 141.19)

Is there a way to speedup these calculations in MAGMA?

Thank you

Post Reply