• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

GPU tanks SMP2 performance

M

Mr. Pedantic

[H]ard|Gawd
Joined
Sep 19, 2009
Messages
1,707
That's the long and the short of it. My total ppd halves when I add my 4870 into folding, from 15k-ish to about 8k-ish. I've decided to run it in -smp 7 and dedicate a whole thread tot he 4870 (in which case I get about 350 ppd extra over just running SMP). I'm running the new core, 2.15, by the way; wasn't that supposed to help with SMP2 performance with GPU clients running?
 
Thats an ATI issue.

Really just run SMP if thats the case.
 
ATI card require significant CPU overhead if i remember correctly, like a core just to itself to achieve less ppd then a 9800gt. I really hope ati is contributing to the move to opengl.
 
It's not only an Ati issue. It's pretty much the same for me and I'm running nVidia clients. Massive drop in PPD on the order of half or more. Ati does use more CPU cycles but there are some nVidia WUs like the 353 that consume a lot, and if you have multiple cards it just gets worse. It's the way the bonus is calculated. It amplifies PPD after a certain performance level is reached. Below that performance level PPD drops off a lot. With the older cores, the PPD drop in conjunction with GPU clients was small because there was no bonus and WU points were very high. Now, the base points are very low and if your SMP performance is corresponding low, you will receive little bonus points to bridge the difference between A3 and A2 PPD performance.

Some architectures greatly benefit from the new system, though. Nehalem is the big winner here.
 
My nV cards do not tank any of my cpu clients (either VM -bigadv or SMP2); I disagree that it's not an ATi only issue.
 
flogge said:
There are some settings you can change to drastically reduce the CPU overhead consumed by an ATI card. In my sig system the 2 4890s together probably reduce my SMP output by 1000 ppd average.

http://www.hardforum.com/showthread.php?t=1419852&highlight=ATI+reduce+CPU
Click to expand...

I've never seen that before, to be honest. I've just set it up, and it looks like CPU usage is down to about 1-2%, down from 10-12%. Definitely a big improvement. I'll wait a while, see how the ppd is affected.
 
Last edited:
Playing with the environment variables can take time but it does help. I tweaked my kid's 4850 and the CPU utilization is much better.

Now I just need to be convinced that I'm not going to take a big hit going from my VM + 2 Nvidia GPUs to SMP2 + 2 GPUs.
 
zero2dash said:
My nV cards do not tank any of my cpu clients (either VM -bigadv or SMP2); I disagree that it's not an ATi only issue.
Click to expand...
I have three systems that say otherwise. Let's see where do I begin? Current PPD output: 4 core Opteron system @2.8GHz = ~1200 PPD with 2 GPUs; 4 core Xeon system @ 2.5 GHz = 1700 PPD with 2 GPUs; 8 core Xeon system @2.6GHz = 650 PPD with 4 GPUs. These are figures using HFM.net. You don't want to see the FahSpy estimates.
 
You have issues, I don't. Other people here don't have those issues either. If someone like Tigerbiten can run 2 GX2's and see roughly a loss of a measly 500ppd per card then IMHO there's no reason to believe nV cards tank cpu PPD like some of you think they do. I don't speak for anyone, I said "my cards do not tank my cpu clients". If EVERYONE had issues, then you can make that broad sweeping generalization, but the fact is it's a hit or miss problem that only affects some people and in most cases, it only affects ATi users.

There's only one way to find out the reality of the issue (or non issue) and that's for everyone to run the same hw under the same environment with the same variables and same setup in terms of clients, flags, os and hardware. Seeing as if that's impossible, it will continue to be a "some people have problems, others don't" sort of issue. Be that as it may, I refuse to make that broad sweeping generalization. That's the same sort of generalizations that Apple comes up with in their ads regarding Window PC's and that's also a load of baloney.

My experience has been contrary to these sorts of posts. I may lose a couple hundred PPD by running gpu clients at the same time, however I come out far and away on the plus side of the spectrum because my gpu clients each gain several thousand PPD in return of taking a few hundred PPD off the cpu client side.
 
Last edited:
zero2dash said:
You have issues, I don't. Other people here don't have those issues either.
Click to expand...
I agree that there are issues with my systems. There's no argument there!

If someone like Tigerbiten can run 2 GX2's and see roughly a loss of a measly 500ppd per card then IMHO there's no reason to believe nV cards tank cpu PPD like some of you think they do.
Click to expand...
Because some people have issues and most may not does not eliminate the possibility conflicts may still exist with certain hardware setups. Absence of evidence does not equate to evidence of absence.

I don't speak for anyone, I said "my cards do not tank my cpu clients". You have issues, I don't. If EVERYONE had issues, then you can make that broad sweeping generalization, but the fact is it's a hit or miss problem that only affects some people.
Click to expand...
By the same token you cannot claim issues do not exist either. I did not intend to make a blanket statement and was only relating my observations. Sorry if you read it that way.
 
APOLLO said:
I agree that there are issues with my systems. There's no argument there!
Click to expand...
:D Heh I mean no offense but I've felt like crap the last few days with the flu and that gave me a good chuckle...thank you. :) LOL

APOLLO said:
Because some people have issues and most may not does not eliminate the possibility conflicts may still exist with certain hardware setups. Absence of evidence does not equate to evidence of absence.

By the same token you cannot claim issues do not exist either. I did not intend to make a blanket statement and was only relating my observations. Sorry if you read it that way.
Click to expand...

True (re: claiming issues don't exist). Not trying to do that either...my experience (and what I've read) has been less people have issues than those who do have issues. I'd like to say it's like a 70/30 non-issues/issues ratio, but I dunno. I'm just glad I haven't really had a lot of issues myself, but I know there's people out there who seem to have nothing but troubles (Kendrak comes to mind...not to pick on him or anything. :p LOL)
 
Theres a lot of unkowns right now and it helps to post what people experience with their systems, but APOLLO, yo stuff is screwed up:p
 
nitrobass24 said:
Theres a lot of unkowns right now and it helps to post what people experience with their systems, but APOLLO, yo stuff is screwed up:p
Click to expand...
I know, LOL. I never see my full production in F@H. I admit I do have more problems that the average person from what I've seen over the years. My hardware selection is eclectic and somewhat unusual, unlike the norm in F@H. More problems surface due to these choices.
 
You must log in or register to reply here.
Back
Top