Xeon Phi and ClMagma

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Locked
quanghieu
Posts: 2
Joined: Wed Jul 05, 2017 4:46 am

Xeon Phi and ClMagma

Post by quanghieu » Wed Jul 05, 2017 9:31 pm

I have an issue when i try clMagma in my algorithm using Xeon Phi co-procesor, lets say my algorithm takes 2.6s to run, when i try clmagma for gpu that time its reduced to 0.8s, thats what i can expect, but when i run it inside Xeon Phi the times goes to 792 seconds, so its there any issue using this platform for ClMagma? (clmagma is recompiled for this platform before any test are run).

Thanks you for your time
Last edited by quanghieu on Fri Nov 24, 2017 12:59 am, edited 3 times in total.
Top
mgates3
Posts: 918
Joined: Fri Jan 06, 2012 2:13 pm

Re: Xeon Phi and ClMagma

Post by mgates3 » Wed Jul 26, 2017 11:58 am

We haven't tested clMAGMA with the Xeon Phi, and don't recommend it. To my knowledge, Intel hasn't updated OpenCL for the Xeon Phi in a while. Simply using the vendor-optimized Intel MKL for the Xeon Phi will yield much better performance.
-mark
Top
Locked

Return to “User discussion”