Calling ALL nVidia 200 series GPU crunchers

Message boards : Number crunching : Calling ALL nVidia 200 series GPU crunchers
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1288370 - Posted: 27 Sep 2012, 10:29:24 UTC

Let's see if this gets anyone's attention...

Contrary to common knowledge, it looks like running 2 tasks on ANY 200 series GPU when using the 270 family drivers is actually a good idea. Use Fred's Performance Tool and find out for yourself:)

Optimize your GPU. Find the value the easy way.

Thanx again Fred!
ID: 1288370 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1288393 - Posted: 27 Sep 2012, 11:16:25 UTC - in response to Message 1288370.  

See the reply to your general comment in the other thread.

Remember that Fred's tool is an artificial benchmark environment, with an identical WU running in each thread, and both WUs started at the same second (which will have an effect on the early CPU-based startup times before GPU crunching gets going fully).

Give it a test in the real world too, before you come to final conclusions.

(Oh, and BTW, the packaged WU in Fred's tool last time I looked was an old stock reference WU, which pre-dated the update to setiathome_enhanced)
ID: 1288393 · Report as offensive
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1288395 - Posted: 27 Sep 2012, 11:31:53 UTC - in response to Message 1288393.  

Richard, I sure hope you didn't take offence to my posts.

And I hope you don't shoot the messenger when I point out that Zule's 260 saw a 13% boost in real-world conditions.
ID: 1288395 · Report as offensive
tbret
Volunteer tester
Avatar

Send message
Joined: 28 May 99
Posts: 3380
Credit: 296,162,071
RAC: 40
United States
Message 1288500 - Posted: 27 Sep 2012, 17:29:57 UTC - in response to Message 1288393.  



(Oh, and BTW, the packaged WU in Fred's tool last time I looked was an old stock reference WU, which pre-dated the update to setiathome_enhanced)



What would you expect to be the effect of a pre-enhanced work unit's use, if predictable at all, without regard to the 200, 400, 600 series issue?
ID: 1288500 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1288517 - Posted: 27 Sep 2012, 18:15:09 UTC - in response to Message 1288500.  

(Oh, and BTW, the packaged WU in Fred's tool last time I looked was an old stock reference WU, which pre-dated the update to setiathome_enhanced)

What would you expect to be the effect of a pre-enhanced work unit's use, if predictable at all, without regard to the 200, 400, 600 series issue?

I asked Joe that at the time. I hope he doesn't mind me quoting his answer:

It's the unmodified reference WU from the project source. So it has the old Classic SaH beam width of 0.1, AR is 0.775, chirp resolution is 0.333 etc.

Although that now differs from current work, I'm inclined to think the test environment differences you noted are more likely causes of suspicious outcomes.

- the 'test environment differences' being WUs starting together (not overlapping the CPU phases of different WUs), always the same WU, etc.

What I've read since then suggests that newer revisions of Fred's tool may include varied WUs with current processing parameters, but I don't know if he's found a way of staggering the start times.
ID: 1288517 · Report as offensive
tbret
Volunteer tester
Avatar

Send message
Joined: 28 May 99
Posts: 3380
Credit: 296,162,071
RAC: 40
United States
Message 1288519 - Posted: 27 Sep 2012, 18:20:54 UTC - in response to Message 1288517.  
Last modified: 27 Sep 2012, 18:25:40 UTC


Although that now differs from current work, I'm inclined to think the test environment differences you noted are more likely causes of suspicious outcomes.


- the 'test environment differences' being WUs starting together (not overlapping the CPU phases of different WUs), always the same WU, etc.

What I've read since then suggests that newer revisions of Fred's tool may include varied WUs with current processing parameters, but I don't know if he's found a way of staggering the start times.


I think I understand the implications of what you have explained, thank you.
ID: 1288519 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1288606 - Posted: 28 Sep 2012, 0:11:21 UTC - in response to Message 1288370.  

Tried the benchmark tool on my old GT240 with driver 285.58

One at a time = 10:25
Two at a time = 09:26

I tried two at a time back when it was my new GT240 and it crapped out on me, very unresponsive system. I'll keep it going @ 1x
</Tazz>
ID: 1288606 · Report as offensive
Profile shizaru
Volunteer tester
Avatar

Send message
Joined: 14 Jun 04
Posts: 1130
Credit: 1,967,904
RAC: 0
Greece
Message 1288736 - Posted: 28 Sep 2012, 8:55:43 UTC - in response to Message 1288606.  

I think that may have more to do with the Celeron than the 240. During the Great Seti Outage of '09 I gave Rosetta a shot on my 2.8GHz Celeron and the poor thing choked. The Atom I'm running now is slightly better at not freezing up on me, but it prefers to run at around 85-90% instead of full throttle. Maybe you could try dialling that Celeron of yours down a notch, if you haven't already. Regardless of the 1task-2task thing...

Question:
When I find some time, I'd like to run Fred's tool using drivers 270, 275, 280 and 285. Any ideas how I could get a more "typical" WU into Fred's tool?

Richard? Anyone?
ID: 1288736 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1288744 - Posted: 28 Sep 2012, 9:26:42 UTC - in response to Message 1288736.  

I think that may have more to do with the Celeron than the 240. During the Great Seti Outage of '09 I gave Rosetta a shot on my 2.8GHz Celeron and the poor thing choked. The Atom I'm running now is slightly better at not freezing up on me, but it prefers to run at around 85-90% instead of full throttle. Maybe you could try dialling that Celeron of yours down a notch, if you haven't already. Regardless of the 1task-2task thing...

Question:
When I find some time, I'd like to run Fred's tool using drivers 270, 275, 280 and 285. Any ideas how I could get a more "typical" WU into Fred's tool?

Richard? Anyone?

I would have to download the new version and see what it's got inside already - but today is going to be a bit busy on Beta. Remind me if I don't get back to you, please.
ID: 1288744 · Report as offensive

Message boards : Number crunching : Calling ALL nVidia 200 series GPU crunchers


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.