Folding Forum

Seriously, thank you to all the people joining the project to help fight this latest threat to all of us humans. Please note that the sudden mass consumption of all available Work Units has caused a heavy strain on the F@H network, but that is to be expected. It will probably take some time for things to level out, but in the time since I wrote the article (this afternoon) and right now (9 PM), the work units are back.

https://greenfoldingathome.com/2020/03/ ... ork-units/

Good job F@H Consortium & Volunteers keeping this thing going. Let's make a dent in that virus.

Keep Calm and Fold On

That's incredible news. Hopefully it will keep growing.

Nice blog.

Could you make a post about any of Nvidia Quadro GPUs? I’ve heard that Quadro is almost twice more efficient in double-precision ( FP64 ) calculations than non-Quadro equivalent. FP64 is what you need for Folding. It’s very interesting.

Let's not say anything that suggests something that's not really true.

FAH does not need anyone to spend the extra money on a Quadro. GeForce is sufficient. ... but you certainly can use one if you already have one.

FAHCore_21 and FAHCore_22 use Mixed Precision ... not exclusively Single Precision or Double Precision. For many years, the FAHCores used purely Single Precision. but as project grew in complexitiy, there was some loss of precision. Using exclusively Double Precision solved that problem but it was significantly slower so the code was rewritten to use Double Precision where it was needed and continue to use Single Precision elsewhere. In fact the use of Double Precision. while important, amounts to a relatively small percentage of the operations accounting for a relatively small percentage of the time sp

ent folding.

Yes, Quadro is more efficient for those operations, but because it's a small percentage, speeding up Double Precision doesn't make a big enough difference to matter much. FAH was designed to run on home computers with GPUS that are affordable for the home user.

If someone can find a GeForce and a Quadro G{Is with identical Single precision GFLOPS but with different Double Precision GFLOPS they're welcome to do some benchmarking on the same FAH workload. Report the difference in throughput and the difference in price. I'd be interested in the results.

bruce wrote:If someone can find a GeForce and a Quadro G{Is with identical Single precision GFLOPS but with different Double Precision GFLOPS they're welcome to do some benchmarking on the same FAH workload. Report the difference in throughput and the difference in price. I'd be interested in the results.

Me too. That's why I ask Paragon to do that, If he can afford any of Quadro.

I've got a bunch of Quadro P4000's and K5000's at work, and a few new RTX ones coming in, but no real way to test them there. I don't think they'd be too happy if I smuggled one home, haha.

I haven't tested these mostly because of cost and availability on the used market, but if I could get my hands on one I would. Most of what I've read shows that they can help for certain projects that require more double precision, but this doesn't happen that often. For the majority of work, single precision is fine. CPU folding, I believe, can help when double precision is required, as well as when special calculations are needed that GPUs aren't as suited for.

Still, it would be awesome to see a head to head comparison. I'll keep my eyes out for any Quadro deals. I might be able to get an old Maxwell-based one and compare it against a GTX 9xxx card. I'm especially interested to know the power and efficiency numbers, since workstation cards might prioritize stability and heat vs. turbo-boosted performance.

GeForce and Quadro have the exact same chips and features. Quadro usually have more memory (which is useless in FAH) and drivers aimed to optimize professional software (CAO/rendreing/...) while GeForce are optimized for games.

There's absolutely no benefit in using a Quadro in FAH instead of the GeForce counterpart.

I'm new to folding I'm trying to fold from a workstation with a Quadro P4000 but it doesn't seem to be downloading any WUs for it. Just the CPU.

Something I'm doing wrong maybe?

Maybe, but more likely issues with the servers getting more requests than they can fill. Servers have been expanded to the point they send out and receive about 10x the assignments the did 2 weeks ago, but the number of people/machines requesting work has gone up 20x. More servers are being prepared, but it takes a bit of time each, and there is a very small staff to do so.

toTOW wrote:GeForce and Quadro have the exact same chips and features.

Do GeForce cards have ECC RAM? I thought only Quadros had that.

Denisok wrote:Nice blog.

Could you make a post about any of Nvidia Quadro GPUs? I’ve heard that Quadro is almost twice more efficient in double-precision ( FP64 ) calculations than non-Quadro equivalent. FP64 is what you need for Folding. It’s very interesting.

Most of F@H is still single-precision, and note that in any case most recent Quadro GPUs do not have high-performance double-precision. Only the Quadro GP100 and GV100 do among the current generations after Kepler, and they are incredibly expensive. None of the other Maxwell, Pascal or Turing Quadros have anything beyond single-precision.

nivedita wrote:

Denisok wrote:Nice blog.

Could you make a post about any of Nvidia Quadro GPUs? I’ve heard that Quadro is almost twice more efficient in double-precision ( FP64 ) calculations than non-Quadro equivalent. FP64 is what you need for Folding. It’s very interesting.

Most of F@H is still single-precision, and note that in any case most recent Quadro GPUs do not have high-performance double-precision. Only the Quadro GP100 and GV100 do among the current generations after Kepler, and they are incredibly expensive. None of the other Maxwell, Pascal or Turing Quadros have anything beyond single-precision.

which means an Nvidia Quadro GP100
is more effective than a GTX1060 ?

if so, the question is, can i put it in the same slot of a midi tower or is the Quadro GP100 too big for that?

i ask because i don't want to spend money senselessly, because otherwise i save the money i invest the money in a GTX1660 Ti with 6 gb ?

Since most of the calculations F@h does in the GPU core are SP, the fast DP of a Quadro GP100 will not speed up things much, possibly not even enough to be detected.

Now if you were doing modeling that used a lot of or only DP, the GP100 will definitely be more effective than the 1060.

only1jva wrote:I'm new to folding I'm trying to fold from a workstation with a Quadro P4000 but it doesn't seem to be downloading any WUs for it. Just the CPU.

Something I'm doing wrong maybe?

off topic in this thread but did you create a GPU slot in FAHControl? Configure > Slots tab > +Add

VAcharonD1 wrote:

toTOW wrote:GeForce and Quadro have the exact same chips and features.

Do GeForce cards have ECC RAM? I thought only Quadros had that.

No, but FAH barely use VRAM, so it's not needed as long as you keep your memory clock in manufacturer's margins.