Advice for an efficient folding GPU... or no GPU folding

[jwyount]

In order to run games at 1920x1080 with a reasonable amount of anti-aliasing your going to need a GTX560Ti or above. . . the 460s just don't cut it on directX 11 and the 480s generate a lot of heat.

Thanks Eagle, and others pointing in that direction.

The GTX560Ti is looking increasingly likely. Looks very good for initial cost, PPD/watt, my modest gaming needs, longevity till obselecense... And my general strategy of buying roughly 12-month-old tech. That strategy has surely saved me boatloads of cash.

I always browse Tom's and Anandtech when re-educating myself for my infrequent rebuilds. They and other sources had already convinced me that a good folding GPU will be quite enough for my gaming tastes.

The Sabertooth z77 that's on its way here has two "PCIe 3.0/2.0 x16 (x16 or dual x8)" Which I understand to mean two cards would run at PCIe3 8x/8x which I understand to be as good as PCIe2 16x/16x... I think it's extremely unlikely I'll ever have more than one GPU, but it's nice to know the option is there. And I'm pretty sure my Cooler Master UCP 900W PSU has all the connectors I'll need.

[{RaW}Eagle1]

The Sabertooth z77 that's on its way here has two "PCIe 3.0/2.0 x16 (x16 or dual x8)" Which I understand to mean two cards would run at PCIe3 8x/8x which I understand to be as good as PCIe2 16x/16x... I think it's extremely unlikely I'll ever have more than one GPU, but it's nice to know the option is there. And I'm pretty sure my Cooler Master UCP 900W PSU has all the connectors I'll need.

Looks like a damn good mobo, but adding a second card with it would not be a good idea. Splitting the bandwidth is unlikely to effect gaming in any way that really matters but may lead to limited results with F@H: A lot of people have had problems with GPU folding recently on projects with high atom counts due to what is believed to be a latency issue. (GPUs were running at around 85% load in most instances). PCIe 3 has been heralded as the solution to the issue, although its worth noting that you'll need a third generation intel CPU to take advantage of it: Check out the ASUS page for your choice of mobo.

For the moment the GTX560Ti seems to strike a balance between outlay, gaming performance and PPD / watt. I guess what really matters is how much you weight each factor in your descision: Is initial cost more or less important than PPD / watt? and just how much does longevity matter to your GPU choice in terms of gaming?

Eagle

[jwyount]

Looks like a damn good mobo, but adding a second card with it would not be a good idea. Splitting the bandwidth is unlikely to effect gaming in any way that really matters but may lead to limited results with F@H: A lot of people have had problems with GPU folding recently on projects with high atom counts due to what is believed to be a latency issue. (GPUs were running at around 85% load in most instances). PCIe 3 has been heralded as the solution to the issue, although its worth noting that you'll need a third generation intel CPU to take advantage of it

I knew it needed a 3rd gen for PCIe3... but that's what I'm getting. This all started with picking the i7-3770k.

Do you believe that "adding a second card with it would not be a good idea" applies even with the 3rd gen Intel Core cpu? Not clear from your post.

For the moment the GTX560Ti seems to strike a balance between outlay, gaming performance and PPD / watt. I guess what really matters is how much you weight each factor in your descision: Is initial cost more or less important than PPD / watt? and just how much does longevity matter to your GPU choice in terms of gaming?

PPD/watt matters more that initial cost. I can afford a GTX 690 but want to be convinced it'll deliver performance at least close to the price difference. It's about 4-5x as expensive as the GTX 560 Ti. Will it soon deliver that much more PPD/watt?

Longevity matters quite a bit to me. I upgrade only when I have a pretty serious need.

(In this case, the upgrade was triggered by the inability of my desktop to reliably support 2tb hard drives, which I need for data storage. After testing with 9 different drives with a boot utility and also testing my RAM I decided the problem was likely enough to be with my motherboard that I might as well replace the whole thing.)

[jwyount]

My current choice seems to be coming down to GTX 560ti versus GTX 680. Currently the 680 is about twice the cost, about the same PPD, and significantly better PPD/watt (I'll try to get some power numbers together). The 680 would presumably have more longevity as a folder (no guarantees) and certainly as a gaming GPU.

It's a bit of a coin toss so far, but at least I'm narrowing the candidates with your help.

[P5-133XL]

Personally, I'd avoid Kepler's till they fix the driver issues. Currently the only WU's that work are beta and even then you have to have the right driver version. Nvidia seems to be releasing new drivers weekly and some work while others are a big fail.

[jwyount]

Personally, I'd avoid Kepler's till they fix the driver issues. Currently the only WU's that work are beta and even then you have to have the right driver version. Nvidia seems to be releasing new drivers weekly and some work while others are a big fail.

Thanks. Didn't realize that it was limited to beta WU's for now.

If I do find myself leaning more towards the Keplers I'll at least wait a month or two before purchasing and see what new info there is. My current GPU will eventually end up in a guest machine, but there's no rush for that.

[{RaW}Eagle1]

I knew it needed a 3rd gen for PCIe3... but that's what I'm getting. This all started with picking the i7-3770k.

Do you believe that "adding a second card with it would not be a good idea" applies even with the 3rd gen Intel Core cpu? Not clear from your post.

So long as you've got an ivy bridge CPU then running a dual PCIe 3 card setup on split bandwidth won't be an issue. (There are no cards currently on the market that can saturate it). It would only be a problem if using a CPU that didn't have a PCIe 3 controller onboard (pre-ivy bridge), at which point it would essentially be split bandwidth PCIe 2.

One thing I am not clear on (and maybe someone of greater technical knowledge can enlighten me) is whether pre-6xx series graphics cards (which are all PCIe 2) would see any benefit from being in a PCIe 3 slot? or would they be unable to utilise the additional bandwidth? I feel it worth asking in this thread as it may have bearing over jwyount's decision.

[P5-133XL]

The issue with the PCI-e bus is from one set of projects p7640-p7644. They do not use the GPU at 100% and the speculation is that they have a latency issue with the bus (not a BW issue). The PCI-e 3.0 has 2x the BW of 2.0 but I have absolutely no idea if the latency has changed.

[{RaW}Eagle1]

I am aware of the issues regarding those projects, as I was one of the first to identify the issue as latency rather than bandwidth. The question was whether a PCIe 2 graphics card will see any benefit from being in a PCIe 3 capable mainboard?

Thanks

[MARSTG]

there is no such thing as efficent GPU folding : GPU folding will always take a lot more electricity than SMP, but simply put is this way : you either make a small investment in the video card and pay more on your hydro bill, or make a larger investment and get a good CPU for SMP folding. Example : A GTX 295 will generate around 17k PPD and with the additional heat, max power on that card is 289W, and will draw all of them while folding, I know from experience, had the 9800GX2 (11k PPD). You can find it for around 120$ on ebay. A Core i7 2600k @ 4GHz will do 27k PPD consuming 95W. True the cost of a 2600k+mobo will be around 420$, 300$ more than the GTX295. From my experience, the 9800GX2 was costing me 20$/month (folding 24/7) so you can have a pretty good idea what GPU folding will do to your hydro bill. I live in Quebec, electricity is relatively cheap compared to other canadian provinces and even US, but if electricity costs a lot where you live my advice is get SMP folding 24/7 and from time to time (winter) GPU folding, especially if you need to pay gor the heating, a dual GPU from nVidia will play the part of a room heater quite well. Avoid the latest series 600 for folding, we don't know when it will be optimized.

[Devlin85]

These days, you get far greater PPD/Watt than CPU.. As an example, I am folding with 2 GTX 760's, those cards by themselves pull in up to 68k PPD EACH and only use a MAXIMUM of 150 Watts. That's 453PPD/Watt! The CPU I use is a i7-4820k which brings in up to 25k PPD using AT LEAST 130 Watts (OC'd), only 192PPD/Watt. Non/OC it's only 15k PPD and that's an even weaker 115PPD/Watt. Clearly GPU folding is the way to go.

And if you are running GPU folding with an Overclocked CPU you are losing points (A LOT), when I was overclocking sure the PPD on my processor went up, buy my GPU folding PPD was cut off at the knees! I was lucky to bring it a total of 40kPPD. After setting back to factory specs for week, 141,571 PPD is my new best! So be carefull with those setting!

[7im]

Hello Devlin85, welcome to the forum. That's all very accurate for folding today.

Please note that with necromancing a thread 2 years old, the folding situation has changed a bit since the original posts were made.

Please also note that when AVX support comes to folding, it will likely double the CPU performance, putting the CPU PPD/watt back to about equal with GPU folding. Should be interesting to see what gets posted 2 years from now.

[new08]

And if you are running GPU folding with an Overclocked CPU you are losing points (A LOT), when I was overclocking sure the PPD on my processor went up, buy my GPU folding PPD was cut off at the knees! I was lucky to bring it a total of 40kPPD. After setting back to factory specs for week, 141,571 PPD is my new best! So be carefull with those setting!

Devlin- what is the cause of this and can it be quantified on any particular rig in a simple way -other than extensive benching?

[7im]

And if you are running GPU folding with an Overclocked CPU you are losing points (A LOT), when I was overclocking sure the PPD on my processor went up, buy my GPU folding PPD was cut off at the knees! I was lucky to bring it a total of 40kPPD. After setting back to factory specs for week, 141,571 PPD is my new best! So be carefull with those setting!

Devlin- what is the cause of this and can it be quantified on any particular rig in a simple way -other than extensive benching?

This happens if you oversubscribed the number of fahcores to more than the number of CPU cores. I wouldn't be too concerned about the details when folding with a config that isn't recommended.

[Nathan_P]

These days, you get far greater PPD/Watt than CPU.. As an example, I am folding with 2 GTX 760's, those cards by themselves pull in up to 68k PPD EACH and only use a MAXIMUM of 150 Watts. That's 453PPD/Watt! The CPU I use is a i7-4820k which brings in up to 25k PPD using AT LEAST 130 Watts (OC'd), only 192PPD/Watt. Non/OC it's only 15k PPD and that's an even weaker 115PPD/Watt. Clearly GPU folding is the way to go.

And if you are running GPU folding with an Overclocked CPU you are losing points (A LOT), when I was overclocking sure the PPD on my processor went up, buy my GPU folding PPD was cut off at the knees! I was lucky to bring it a total of 40kPPD. After setting back to factory specs for week, 141,571 PPD is my new best! So be carefull with those setting!

Welcome to the forum.

On a single cpu config I would have to agree with you but its not true at the higher end of cpu folding. Dual E5 2692v2's @2.4 using an AX650 PSU pull 278w from the wall and earn 470k PPD, that's 1,690PPD/w. Dual X5670's get 488PPD/W and dual E5-2665's get 939PPD/W.