Question[s] regarding various GPUs for folding

Come join the... uh... er... fold.

Moderators: just brew it!, farmpuma

Question[s] regarding various GPUs for folding

Postposted on Tue Dec 10, 2013 9:19 pm

Is there much of a difference in the Points Per Day potential of current generation cards from AMD or Nvidia? Is double precision arithmetic important at all (would a Titan earn more PPD than a 780 Ti)? Has Nvidia or AMD traditionally been better for folding, or have they leapfrogged one another every so often? What is the speculation regarding the next generation cards from both camps, if any?

Would any of you cringe at the thought of folding with a reference R9 290 twenty four hours a day?

I'm tentatively planning on upgrading from my 660 Ti in the first or second quarter of next year, and I'd like to make F@H PPD one of the main factors to consider.
Dysthymia
Gerbil
 
Posts: 17
Joined: Thu May 01, 2008 4:54 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Tue Dec 10, 2013 10:18 pm

Lack of folding testing in the Tech Report GPU review and comparison articles is an annoyance for me, and I have never gotten any response from TR. I am left to believe that management is bored with it or doesn't think we need or want this information. To me this is a huge lacking, and any coverage not including it is like a disk drive review not including capacity, RPM, or cache information, or CPU reviews not containing cache, clockspeed, or hyperthreading information.

I need more comprehensive information with GPU reviews, and I feel I'm not getting it on TR. I need folding information. If I should find a site that regularly includes folding or GPGPU information in GPU reviews, then I will start going there first for all of my future graphic card reading and education.

Because of this, I cannot answer your question; there's no way to know how a 290 or 290X would perform compared to a Titan or a 7980, 780 GTX, etcetera. All you can do at this point is buy whatever you think will serve you in your OTHER interests, and then when you receive the card, do your own testing to see how it performs for folding.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1565
Joined: Tue May 25, 2004 7:41 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Wed Dec 11, 2013 12:47 am

When I was testing it, the Titan was giving me a slightly better PPDs than 290x, with much better efficiency (it wasn't trying to turn my room into sauna)... Be aware that it can still be pretty inconsistent depending on the type of work unit and a lot of it also depends on particular drivers (for example, a couple of latest Nvidia's versions have significantly decreased the PPDs on certain GPUs). If you want more details about whole thing and about the performance of particular GPU - I suggest finding a very active Folding-related forum sections. HardOCP's forums have one, so does OCN forums.
My subscription allows you people to exist on this site and makes me a better human being than you'll ever be
JohnC
Gerbil Jedi
Gold subscriber
 
 
Posts: 1886
Joined: Fri Jan 28, 2011 2:08 pm
Location: NY/NJ/FL

Re: Question[s] regarding various GPUs for folding

Postposted on Wed Dec 11, 2013 1:27 am

Thank you for sharing your experience, John. It's difficult to find any info and whatever people offer is helpful.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1565
Joined: Tue May 25, 2004 7:41 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Wed Dec 11, 2013 6:21 pm

The problem is, the software probably isn't coming close to fully utilizing a 290X so it's just too hard to say. Pande Group's software took years (and several major rewrites, we're now using the GPU3 client) before my GTX 480 would show 99% load utilization under GPUZ. I'm sure it will take at least a year before it does a good job maximizing the 290X's potential.

Regarding the other question, Folding@home only uses single-precision computations. So double-precision is meaningless for comparisons with F@H. FAHBench offers single, double, explicit, and implicit tests, but the only test that matters is single-precision implicit. Unless they've begun experimenting, last I heard F@H only uses implicit, single-precision calculations in all GPU work units.
Kougar
Gerbil Team Leader
 
Posts: 298
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: Question[s] regarding various GPUs for folding

Postposted on Fri Dec 13, 2013 8:46 pm

Well, I asked around and was able to find this:
http://compdewddevelopment.com/projects ... ByPpd=true
Stats from 10 Nvidia and 8 AMD cards from the past three generations plus some older stuff.
Dysthymia
Gerbil
 
Posts: 17
Joined: Thu May 01, 2008 4:54 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Fri Dec 13, 2013 9:43 pm

BIF wrote:Lack of folding testing in the Tech Report GPU review and comparison articles is an annoyance for me, and I have never gotten any response from TR. I am left to believe that management is bored with it or doesn't think we need or want this information. To me this is a huge lacking, and any coverage not including it is like a disk drive review not including capacity, RPM, or cache information, or CPU reviews not containing cache, clockspeed, or hyperthreading information.

I need more comprehensive information with GPU reviews, and I feel I'm not getting it on TR. I need folding information. If I should find a site that regularly includes folding or GPGPU information in GPU reviews, then I will start going there first for all of my future graphic card reading and education.

Because of this, I cannot answer your question; there's no way to know how a 290 or 290X would perform compared to a Titan or a 7980, 780 GTX, etcetera. All you can do at this point is buy whatever you think will serve you in your OTHER interests, and then when you receive the card, do your own testing to see how it performs for folding.


the problem with folding performance is that the work units aren't standardized. PPD data collected would only be valid for a few months (assuming they are valid at all).
Diplomacy42
Gerbil
 
Posts: 63
Joined: Sat Sep 01, 2012 2:56 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Sat Dec 14, 2013 9:34 am

The problem I have with hardware reviews and such when it concerns folding is they all come down to PPD numbers. From what I can tell the PPD numbers you see on your folding client are pie in the sky numbers. From what I understand the PPD number means I have to run a work unit for 24 hours with no down time to get that number. Well there is down time even if its just for downloading and uploading files then switching over to the new WU. PPD and what points you get in reality are never the same thing.

For the last year or so Nvidia was consider the better choice for folding. But we now have new hardware from AMD and there is a issue with Nvidia drivers and changes they made to OpenCL, which is what F@H uses. People with GK 104 and 110 GPU's are all seeing a big drop off in PPD. So you have a choice, go back to older drivers and take a hit in performance if you game on the same machine.

I tried a driver roll back myself and my numbers for folding did go up. But the down side was most of my newer games were about unplayable due to stutter and low FPS. I went back to newer drivers, complained on Nvidia's forums about it and Am hoping they do something but the general consensus over at the F@H forums is Nvidia isn't going to fix it soon if at all.

So right now it seems to be a toss up on which is better, AMD or Nvidia. If it was me, I would wait until AMD gets it act together on R9 drivers and for some third party cooling solutions show up for the R9 series cards. If those things happen then AMD is probably the better choice for the money you spend. Then again, as things stand currently, I am considering a GTX 780 Ti. Faster than the R9 290 yet cheaper than the Titan. But keep in mind, I game on the same system as I fold on.
Khali
Gerbil XP
 
Posts: 336
Joined: Thu Dec 20, 2012 5:31 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Sat Dec 14, 2013 10:11 am

BIF wrote:Lack of folding testing in the Tech Report GPU review and comparison articles is an annoyance for me, and I have never gotten any response from TR. I am left to believe that management is bored with it or doesn't think we need or want this information. To me this is a huge lacking, and any coverage not including it is like a disk drive review not including capacity, RPM, or cache information, or CPU reviews not containing cache, clockspeed, or hyperthreading information.


Little harsh aren't we?

The time it takes to compile necessary information for this would not even come close to being cost effective for TR. New drivers, new WUs, F@H issues (i'm sure everyone has experienced downtime if they have ever folded for a while), and the number of data points needed for the relatively small audience that wants this all make this an issue. It's just not worth their time I suspect.

Now don't get me wrong, I think it'd be awesome if this info was included in reviews. I however do not think it's nearly as awful as you make it out to be.
i7 860 - GA-P55-USB3 - 8GiB - HD7850 - SSD - 3.64TB HDD - Xonar D1 - U2410 - Win7 Pro x64.
DancinJack
Minister of Gerbil Affairs
 
Posts: 2045
Joined: Sat Nov 25, 2006 3:21 pm
Location: Austin, TX

Re: Question[s] regarding various GPUs for folding

Postposted on Fri Dec 27, 2013 8:08 pm

DancinJack wrote:
BIF wrote:Lack of folding testing in the Tech Report GPU review and comparison articles is an annoyance for me, and I have never gotten any response from TR. I am left to believe that management is bored with it or doesn't think we need or want this information. To me this is a huge lacking, and any coverage not including it is like a disk drive review not including capacity, RPM, or cache information, or CPU reviews not containing cache, clockspeed, or hyperthreading information.


Little harsh aren't we?

The time it takes to compile necessary information for this would not even come close to being cost effective for TR. New drivers, new WUs, F@H issues (i'm sure everyone has experienced downtime if they have ever folded for a while), and the number of data points needed for the relatively small audience that wants this all make this an issue. It's just not worth their time I suspect.

Now don't get me wrong, I think it'd be awesome if this info was included in reviews. I however do not think it's nearly as awful as you make it out to be.

Stanford does have a GPU folding benchmark that should not take that long to run. They don't measure in PPD, but at least it is a 3rd party (not AMD or Nvidia) program measuring something. But then again so many sites are running them now I am not sure what value TR can add to that. Perhaps if we dangle the *pageviews!* flag in front of Damage/Geoff that may work. :P
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24417
Joined: Mon May 24, 2004 2:19 am

Re: Question[s] regarding various GPUs for folding

Postposted on Mon Dec 30, 2013 10:40 am

Flying Fox wrote:Stanford does have a GPU folding benchmark that should not take that long to run. They don't measure in PPD, but at least it is a 3rd party (not AMD or Nvidia) program measuring something. But then again so many sites are running them now I am not sure what value TR can add to that. Perhaps if we dangle the *pageviews!* flag in front of Damage/Geoff that may work. :P


You're referring to FAHBench. The problem is only one of the four benchmarks it gives are applicable to actual F@H work, and even then the benchmark doesn't appear to properly utilize newer hardware it doesn't recognize. It was reasonably accurate when it first came out but now I'm not sure it is, specifically some of the rankings look wrong when compared to current F@H PPD numbers.
Kougar
Gerbil Team Leader
 
Posts: 298
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: Question[s] regarding various GPUs for folding

Postposted on Tue Dec 31, 2013 7:48 pm

Hi fellas.

Yeah, I understand the difficulties; I just don't care. :wink:

Meaning this: If we can come up with ways to benchmark FPS, dropped frames, jitterry timing issues, and whatall crazy other things we've seen reported in the GPU space, then we can CERTAINLY come up with some standardized baseline for benchmarking F@H performance, WU differences notwithstanding.

We have benchmarking tools for EVERYTHING these days, be them CPUs, GPUs, Hard and Solid State disk drives, network devices, and even low-flush toilets!

And really, who better than TR to do this (the GPU thing, not the toilet thing)? Who would you trust more to come up with a way that produces meaningful comparison information? My money is not on Tom's Hardware, Valve, or the manufacturers! There is a void here that should be filled, and so I shall continue to voice my concern for this.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1565
Joined: Tue May 25, 2004 7:41 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Wed Jan 01, 2014 3:02 pm

BIF, I agree but don't hold your breath. I couldn't get them to clean up the old posts from back in 2005 that are still in this part of the forum. I'm pretty sure they wont take the time to do benchmarks either.

Then again the attitude is prevalent on most other tech sites with a folding team. Heck, the people doing bitcoin mining get more attention than those who are actually doing real worth while science with their computers.
Khali
Gerbil XP
 
Posts: 336
Joined: Thu Dec 20, 2012 5:31 pm

Re: Question[s] regarding various GPUs for folding

Postposted on Wed Jan 01, 2014 8:20 pm

One issue is the folding algorithms and methods change over time. Something that is meaningful today will not be a year from now. I used notfred's bootable USB to help me select a quad core intel chip. Tinker / Amber methods are in the past. SMP may be declining. The points awarded reflect this trend.

Maybe we need a few dedicated individuals who will work to create and maintain an appropriate benchmark.
PhilipMcc
Gerbil
Gold subscriber
 
 
Posts: 90
Joined: Thu Feb 05, 2009 10:15 am
Location: Pittsburgh


Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: Yahoo [Bot] and 4 guests