Advanced search

Message boards : Number crunching : Single Precision/ Double Precision

Author Message
Jorge Alberto Ramos Olive...
Send message
Joined: 13 Aug 09
Posts: 24
Credit: 156,684,745
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwat
Message 27041 - Posted: 2 Oct 2012 | 6:36:14 UTC

Hi:

- In what type of computation is currently GPUGrid involved? Single Precision or Double Precision in the GPU?

(next question, mostly intended for developers:)

-In the forseable future, will the programing model for GPUGrid change? (e.g. if currently running SP and intend to change to DP) Any benefits of doing this?

:)

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27042 - Posted: 2 Oct 2012 | 11:11:33 UTC - in response to Message 27041.
Last modified: 2 Oct 2012 | 11:12:23 UTC

Hi:

- In what type of computation is currently GPUGrid involved? Single Precision or Double Precision in the GPU?

(next question, mostly intended for developers:)

-In the forseable future, will the programing model for GPUGrid change? (e.g. if currently running SP and intend to change to DP) Any benefits of doing this?

:)


Its all single precision. Nvidia and AMD have both restricted the double precision capabilities of their cards. The current lot of Nvidia cards do 1/16th DP. If you want DP performance you need to buy the expensive Tesla cards.

I can't see the projects programming model changing much in the coming years. They will probably move to a later cuda version, but they already have issues getting people to update their drivers.
____________
BOINC blog

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27043 - Posted: 2 Oct 2012 | 18:54:21 UTC - in response to Message 27042.
Last modified: 2 Oct 2012 | 18:56:13 UTC

I agree that the research will probably remain single precision for the foreseeable future. GPUGrid is very much dependent on GPU availability. As it's Boinc based there aren't going to be very many Telsa cards available. So developing a project for new Teslas isn't feasible. This isn't to say that the researchers don't do other research in house, and should such cards be available and a suitable project present itself then of course they would be interested. However, for crunchers there isn't much of a reason to get a Tesla card.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Wrend
Send message
Joined: 9 Nov 12
Posts: 51
Credit: 522,101,722
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwat
Message 37726 - Posted: 24 Aug 2014 | 18:16:09 UTC
Last modified: 24 Aug 2014 | 18:28:44 UTC

This thread is a bit dated, but it was the first result I found in a Google search on this subject, and the subject is still relevant.

For reference, I started this thread over at Overclock.net: http://www.overclock.net/t/1509123/is-using-double-precision-actually-better

So, now to the point:

Would it currently be better if I allot more double precision computing performance (lowering single precision) on my two EVGA (/NVIDIA) Titan Black Superclocked cards in the NVIDIA Control Panel?

Here are some relevant specs: "Nvidia states the Titan Black will have single-precision performance of 5.1 TeraFlops and double-precision performance of 1.3 Tera Flops." -http://www.bit-tech.net/news/hardware/2014/02/18/nvidia-gtx-titan-black-launched/1

Please advise, and thanks!
____________
My BOINC Cruncher, Minecraft Multiserver, Mobile Device Mainframe, and Home Entertainment System/Workstation: http://www.overclock.net/lists/display/view/id/4678036#

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 37727 - Posted: 24 Aug 2014 | 19:04:25 UTC - in response to Message 37726.

Would it currently be better if I allot more double precision computing performance (lowering single precision) on my two EVGA (/NVIDIA) Titan Black Superclocked cards in the NVIDIA Control Panel?

No.
Hope that helps :)

Wrend
Send message
Joined: 9 Nov 12
Posts: 51
Credit: 522,101,722
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwat
Message 37730 - Posted: 24 Aug 2014 | 21:04:17 UTC - in response to Message 37727.

Yes, thanks for the confirmation.

We eventually arrived at that conclusion in the Overclock.net thread as well, but I appreciate hearing the same from another source.

Now... to crunch some tasks and rack up some credits for this project to try and bring in up to speed with some the other projects I crunch for.
____________
My BOINC Cruncher, Minecraft Multiserver, Mobile Device Mainframe, and Home Entertainment System/Workstation: http://www.overclock.net/lists/display/view/id/4678036#

RaymondFO*
Send message
Joined: 22 Nov 12
Posts: 72
Credit: 14,040,706,346
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 37739 - Posted: 25 Aug 2014 | 15:28:00 UTC - in response to Message 37730.

To give you more background, even though the technical NVIDIA DP specifications are good for all the TITANS, the real problem is the clock speed is lowered when DP is enabled. At GPU Grid, you will only impair your performance when you enable DP.

Post to thread

Message boards : Number crunching : Single Precision/ Double Precision

//