mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2013-11-18, 00:44   #1
Fusion_power
 
Fusion_power's Avatar
 
Aug 2003
Snicker, AL

7×137 Posts
Default GPU advances including massive memory running Prime95?

Arguably the best way to speed up Prime95 would be to run it on a GPU card. What are the chances of adapting the software to one of these?

http://www.amd.com/us/press-releases...2013nov14.aspx

Last fiddled with by Fusion_power on 2013-11-18 at 00:44 Reason: spelling
Fusion_power is offline   Reply With Quote
Old 2013-11-18, 01:06   #2
kracker
ἀβουλία
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

32·241 Posts
Default

LL testing programs exist for CUDA and OpenCL.
kracker is offline   Reply With Quote
Old 2013-11-18, 01:17   #3
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·5·941 Posts
Default

Quote:
Originally Posted by Fusion_power View Post
Arguably the best way to speed up Prime95 would be to run it on a GPU card.
Beautifully funny.
chalsall is offline   Reply With Quote
Old 2013-11-18, 02:46   #4
flashjh
 
flashjh's Avatar
 
"Jerry"
Nov 2011
Vancouver, WA

1,123 Posts
Default

Will be quite the revelation once it's working well . Some progress is forthcoming and there should be some significant and exciting improvements in the near future...
flashjh is offline   Reply With Quote
Old 2013-11-18, 03:57   #5
Fusion_power
 
Fusion_power's Avatar
 
Aug 2003
Snicker, AL

7×137 Posts
Default

Lets see. I think I posted in the software section. I think I asked about Prime95, not mlucas, not cuda.

The last I checked I think Prime95 supports up to 4 cores working independently. How would it perform on a multicore gpu? Especially one with 12 GB memory? What would be required to adapt the software? and what advantages would be gained.
Fusion_power is offline   Reply With Quote
Old 2013-11-18, 04:03   #6
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

22·32·167 Posts
Default

Quote:
Originally Posted by Fusion_power View Post
Lets see. I think I posted in the software section. I think I asked about Prime95, not mlucas, not cuda.

The last I checked I think Prime95 supports up to 4 cores working independently. How would it perform on a multicore gpu? Especially one with 12 GB memory? What would be required to adapt the software? and what advantages would be gained.
Prime95 is an x86 program. Your question is like asking a dog to speak Chinese with a chicken. This is why things like CUDA and OpenCL exist; so that we can run code on the GPUs.
retina is online now   Reply With Quote
Old 2013-11-18, 04:30   #7
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

110101111112 Posts
Default

Yup. CUDALucas and CLLucas ARE the "ports" of Prime95 to GPUs which support CUDA or OpenCL computation. They just don't have the same GUI as Prime95 does, or nearly as much functionality.

To clarify Retina's statement in case it wasn't direct enough, code doesn't "just run." Different hardware uses different assembly languages and what is designed for one piece of hardware does in fact sound like Chinese to another. Prime95 is written in x86, CLLuas is used for OpenCL (AMD's GPUs) and CUDALucas is used for CUDA (Nvidia's GPUs), because your CPU does not understand CUDA and an HD 7970 does not understand x86.

To overcome this difference in assembly languages, drivers essentially translate the language but that isn't sufficient to account for the massively different architectures between a GPU and CPU (which are the actual reason for GPUs being so insanely much faster). Instead, a separate program is written to work on a CUDA compatible device or an OpenCL compatible device.


If you were thinking of running Prime95.exe off a GPU instead of off the CPU, exactly the way it is (to run a blend stress test, do P-1 factoring or automatically contact the server for work, for example) then you're out of luck.
TheMawn is offline   Reply With Quote
Old 2013-11-18, 15:00   #8
Manpowre
 
"Svein Johansen"
May 2013
Norway

3·67 Posts
Default

The only thing that can change this, is with introducing of ARM processors on the same dye as the GPU processor. That means we dont have to memcopy data from gpu to processor to get processor cycles. we can rather have the ARM cpu process it right there and then.
Manpowre is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Prime95 P-1 memory leak? TheMawn Software 5 2013-12-12 01:22
Running other programs while running Prime95. Neimanator PrimeNet 14 2013-08-10 20:15
i7/prime95 memory question stars10250 Hardware 115 2012-10-24 15:24
Prime95 Memory Usage jimmyhua Software 7 2005-07-10 07:37
Icons, including favicons, for ElevenSmooth wblipp ElevenSmooth 3 2003-11-13 12:11

All times are UTC. The time now is 22:41.

Tue Jan 26 22:41:00 UTC 2021 up 54 days, 18:52, 0 users, load averages: 2.73, 2.44, 2.51

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.