Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
-
roalmar2
- Posts: 20
- Joined: Thu Jul 03, 2014 6:06 am
Post
by roalmar2 » Tue Sep 23, 2014 12:29 pm
Hello,
I am working with multigpus system.
In
http://icl.cs.utk.edu/magma/news/news.html?id=345 at pdf:
http://icl.cs.utk.edu/projectsfiles/mag ... MA_1.4.pdf at the page 24, you expose dsytrd function using it with several gpus.
Why you use it? I put different commands and the results are all simillars that when 1 gpu is used.
Like these:
Code: Select all
testing_dsytrd_gpu -N 15000 --ngpu 3
testing_dsytrd_gpu -N 15000 --ngpu 1
N CPU GFlop/s (sec) GPU GFlop/s (sec) |A-QHQ'|/N|A| |I-QQ'|/N
===========================================================================
15000 --- ( --- ) 14.54 ( 309.62) --- ---
Why? Could you explain me where is my mistake?
Thank you!!!
-
mgates3
- Posts: 918
- Joined: Fri Jan 06, 2012 2:13 pm
Post
by mgates3 » Wed Sep 24, 2014 4:28 pm
Use the multi-GPU interface, magma_dsytrd_mgpu, in testing_dsytrd_mgpu.
magma_dsytrd is a CPU interface to one GPU.
magma_dsytrd_gpu is a GPU interface to one GPU.
-mark
-
roalmar2
- Posts: 20
- Joined: Thu Jul 03, 2014 6:06 am
Post
by roalmar2 » Fri Sep 26, 2014 7:20 am
Ok, thank you very much.
Some error in my compilation occurs and I haven't got the executable file.
I will recompile it.