Welcome to MilkyWay@home

nVidia GPUs and double precision

Message boards : Number crunching : nVidia GPUs and double precision
Message board moderation

To post messages, you must log in.

AuthorMessage
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 46306 - Posted: 19 Feb 2011, 18:02:15 UTC

I am an ATI GPU cruncher mainly, but for certain projects it might be useful to use the occasional nVidia GPU.

For all project application, including Milkyway, what is the oldest series nVidia GeForce GPU card that provides double precision?

For example it it might be the GTX2xx series or higher.
Go away, I was asleep


ID: 46306 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Werkstatt

Send message
Joined: 19 Feb 08
Posts: 350
Credit: 141,284,369
RAC: 0
Message 46307 - Posted: 19 Feb 2011, 19:20:26 UTC - in response to Message 46306.  

I crunched MW on GTX260 and GTX460.
ID: 46307 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 46308 - Posted: 19 Feb 2011, 20:44:49 UTC - in response to Message 46307.  

I crunched MW on GTX260 and GTX460.



May I ask what sort of outputs they individually give, and what their power draw is estimated to be, Werkstatt?

I was mainly thinking of adding a additional NV GPU to one or both of my quads (then reach the limit of PCIe), but aim these GPUs at PrimeGrid. But with the flexibility to be used here or any of the GPU projects (even SETI MB and Einstein).
Go away, I was asleep


ID: 46308 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Werkstatt

Send message
Joined: 19 Feb 08
Posts: 350
Credit: 141,284,369
RAC: 0
Message 46309 - Posted: 19 Feb 2011, 21:18:46 UTC - in response to Message 46308.  
Last modified: 19 Feb 2011, 21:25:12 UTC



May I ask what sort of outputs they individually give, and what their power draw is estimated to be, Werkstatt?

I was mainly thinking of adding a additional NV GPU to one or both of my quads (then reach the limit of PCIe), but aim these GPUs at PrimeGrid. But with the flexibility to be used here or any of the GPU projects (even SETI MB and Einstein).


I can not remember the time the GTX260 took; I gave that card a friend.
The GTX460 needs ~11 minutes for one OpenCL-Task; power-consumption is slightly higher than that of my HD6950 in the same machine.
Edit: GTX260 was long before the OpenCL, it was CUDA-time!
ID: 46309 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
HalfEmpty

Send message
Joined: 16 Nov 10
Posts: 17
Credit: 17,999,210
RAC: 0
Message 46340 - Posted: 22 Feb 2011, 7:36:29 UTC

Take a look at the CUDA wiki page, and a couple of screens down you will see a table with the compute capability of the more popular card families. Anything with a compute capability of 1.3 or higher will do doubles.
ID: 46340 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
europa

Send message
Joined: 29 Oct 10
Posts: 89
Credit: 39,246,947
RAC: 0
Message 46346 - Posted: 23 Feb 2011, 22:53:45 UTC - in response to Message 46308.  

If price is a major consideration, another possibility would be to go with what I call a "baby Fermi" like the GTX-430. They run about half the price of the GTX460 and have a speed of 179 GFlops vs. 673 for the 460.

As you can see thouth, you get a lot more bang for the buck with the 460. Since the GPU's are close generationally, you don't have problems running them together in the same machine as I'm doing in one of mine.

The 460 runs at about 51-56oC while the 430 runs at about 40oC.

Regards,
Steve
ID: 46346 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Haris Dublas

Send message
Joined: 25 Feb 10
Posts: 49
Credit: 10,137,837
RAC: 0
Message 46348 - Posted: 25 Feb 2011, 7:00:29 UTC
Last modified: 25 Feb 2011, 7:03:02 UTC

Or get a 440, its basically a 430 (a little bit higher clocks) with ddr5. It will perform better in Collatz than the 430 when MW have troubles. The OEM one have 112 shaders while the retail one is only a pumped up 430 (96 shaders).
ID: 46348 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : nVidia GPUs and double precision

©2024 Astroinformatics Group