mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   GPU Computing (https://www.mersenneforum.org/forumdisplay.php?f=92)
-   -   EVGA GTX 580 memory (https://www.mersenneforum.org/showthread.php?t=17142)

ET_ 2012-09-05 13:53

EVGA GTX 580 memory
 
Ok, it's time to look for a new GPU, cc >= 2.0, so I decided to "adopt" a GTX 580.

I just noticed that the clock of this model ranges from 770 to 800 MHz, while the onboard memory is either 1536 or 3072MB.

I'd like some indication about the more reliable manufacturers (ASUS, Zotac, EVGA...), and some information related to the use of 3GB of onboard memory with the actual programs we all know (mfaktc, cudalucas, msieve,, gpu-ecm...): is it worth having more memory, or the latency/bandwidth problems tend to minimize its use?

Luigi

firejuggler 2012-09-05 14:13

Gigabyte/Asus never failled me (yet).

kladner 2012-09-05 14:50

I am running a Gigabyte GTX 460 (2 fan) and a Gigabyte 570 (3 fan) in the same system. There are questions about the 570, which I have not been able to get to run CudaLucas. It has been suggested that this may be due to a memory problem, though I have been unable to demonstrate this with any test utility I have found.

Both of these models dump a lot of their hot air inside the case. I don't know how the models/brands with a single centrifugal blower compare in cooling ability. These [U]appear[/U] to exhaust more of their heat to the outside, but I have no experience with them. Regardless, you are likely to see considerably more heat in your case, especially since you are talking a GTX 580 which draws more power than a 570.

Note that cc=2.0 is the most desirable level. On the memory question, my understanding is that larger amounts would mainly be called for if you plan to run CudaLucas with very large exponents. My GTX 460 (1 G RAM) is currently using 146 MB on a low 46M exponent in CL.

Dubslow 2012-09-05 18:42

Yes, kladner's right. Memory has almost zero effect on mfaktc, and CUDALucas doesn't use too much memory for exponents in the current wavefronts. If you plan on running 332M+ expos with CUDALucas, 1.5 GB may or may not be enough (I have no idea what the exact memory requirements are).


All times are UTC. The time now is 12:56.

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