mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   GPU Computing (https://www.mersenneforum.org/forumdisplay.php?f=92)
-   -   GTX 680 vs. 580 (https://www.mersenneforum.org/showthread.php?t=16766)

Rodrigo 2012-04-29 05:23

GTX 680 vs. 580
 
I'm flirting with the idea of getting a high-end GPU for TF, and when I visited the [URL="http://mersenne-aries.sili.net/mfaktc.php?sort=ghdpd&noA=1"]benchmark page[/URL] I saw that the Nvidia GeForce GTX 580 has higher productivity ratings than the newest model, the GTX 680.

This was a surprise to me, as the 680 signifcantly outperforms even overclocked 580's in the majority of [URL="http://www.maximumpc.com/article/features/kepler_unveiled_nvidias_gtx_680_benchmarked_-depth"]Maximum PC's tests[/URL], and in the rest runs at worst about even.

What accounts for this?

Also, can someone explain why the 680 is given on the comparison chart as drawing less power "at full throttle" (as Maximum PC puts it) than the 580 AND doing twice as many GFLOPS, yet it rates considerably lower in both GHz-days/day and (GHd/d)/W?

Maybe in normal usage conditions (neither idle nor maximum) the 680 draws more power than the 580? Or is Compute 3.0 that much less efficient for TF than 2.0?

Just trying to understand the numbers here, thanks.

Rodrigo

Karl M Johnson 2012-04-29 05:32

All the answers are in [url=http://mersenneforum.org/showthread.php?t=16508]this thread[/url].
It's in the same subforum:smile:

In short, it's a great gaming GPU, but that's it.

Dubslow 2012-04-29 05:32

[url]http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/17[/url]

(The GeneferCUDA is written by msft :smile:)

sdbardwick 2012-04-29 05:38

The [URL="http://mersenneforum.org/showthread.php?t=16508"]Kepler thread[/URL] has some explanation, but here's the TL;DR version:
Optimized for single precision operations (gaming) at the expense of double precision (computing).

Rodrigo 2012-04-29 17:18

1 Attachment(s)
Thanks guys. It looks like I'd be skipping the 680.

Considering how well y'all keep on top of GPU developments, it was weird that forum searches based on "680" and "GTX 680" had turned up zero results.

ATH 2012-04-29 18:36

[QUOTE=Rodrigo;297877]Considering how well y'all keep on top of GPU developments, it was weird that forum searches based on "680" and "GTX 680" had turned up zero results.[/QUOTE]

use google, search like this:
gtx 680 site:mersenneforum.org

shows several of the threads.

Dubslow 2012-04-29 20:23

[QUOTE=Rodrigo;297877]
Considering how well y'all keep on top of GPU developments, it was weird that forum searches based on "680" and "GTX 680" had turned up zero results.[/QUOTE]Or Kepler, the architecture's code name. You'll notice that's what the thread is called.

Rodrigo 2012-04-29 20:44

[QUOTE=ATH;297883]use google, search like this:
gtx 680 site:mersenneforum.org

shows several of the threads.[/QUOTE]
Golly, Google Search works better than our own forum's search function.

Thanks for the tip!

Rodrigo

Xyzzy 2012-04-30 01:30

[url]http://www.mersenneforum.org/showthread.php?t=15556[/url]

Rodrigo 2012-04-30 05:13

[QUOTE=Xyzzy;297934][URL]http://www.mersenneforum.org/showthread.php?t=15556[/URL][/QUOTE]
Thanks, Xyzzy -- appreciate it!

Another benefit of using the Google function is that we don't have to wait 60 seconds between searches, if the first one didn't come up with anything.

What's the purpose of that waiting period? [Forum Feature Request:] It's one reason I didn't try more, different search terms when I was looking up the GTX 680 -- got tired of sittin' there waiting for the obligatory minute to be up.

Rodrigo

Dubslow 2012-04-30 05:40

[QUOTE=Rodrigo;297956]
What's the purpose of that waiting period? [/QUOTE]I'd guess that such a search function is computationally expensive, by far the most straining thing the average user can do around here. Were there not a limit, it would be relatively easy to launch gazillions (relatively speaking) of search requests and take down the server (analogous to a DDoS, except because each request is that much more work serverside relative to just a page load, you need far less total requests to have an effect). Perhaps someone has already done something of the sort?


All times are UTC. The time now is 13:05.

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