Folding@home GPU2 Beta Released, Examined

ThinSkin writes "Stanford has recently released an update to their Folding@home GPU-accelerated client, which includes notable upgrades such as support for more current Radeon graphics cards and even a visualizer to see what's going on. ExtremeTech takes a good look at the new Folding@home GPU2 client and interviews Director Dr. Vijay Pande about the project. To the uninitiated, Folding@home is a distributed computing project in which hundreds of thousands of PCs and PS3s devote a portion of their computing power to crunch chunks of biological data. The goal is 'to understand protein folding, misfolding, and related diseases.'"

Support for NVIDIA GPUs coming?

[Grokmoo]

From the article:

ET: Whenever someone hears about GPU-accelerated FAH, their first question is why there is no client with support for Nvidia cards. In the past it was said that it had more to do with Nvidia's drivers. Now that the core doesn't use DirectX, couldn't a GPU client use Nvidia's CUDA? Is there any work going on there, and if not, why not? Dr. Pande: We are interested in CUDA and are investigating how well FAH on CUDA would work.

I am awaiting this with some serious excitement. Getting Folding@home working on Nvidia GPUs would definitely add a lot of computing power into the mix. This is especially true now, as it seems that the current crop of high end GPUs seems to favor Nvidia.

From the benchmarks I have seen, it seems that there are currently no games that can effectively utilize, for example, 2 9800 GX2s. If Folding@home releases an Nvidia client, those people who have plunked $1000 into graphics cards may finally be able to put them to use!

Re:Support for NVIDIA GPUs coming?

[magarity]

The first thought that comes to mind is whether the 8xxx and newer with the stream processors would need a completely different programming approach compared to the prior models. This being /., who wants to be first to pretend to know the answer?

Re:Doing this at work?

[Gewalt]

Better check up on power consumption there, factor in 124 hours a week at 240 watts (conservative) x150 machines. Take a look at how much money your company is "donating" in raw power consumption, then triple it, cause for every watt of heat dissipated, it takes 2 watts of AC to remove it. See if your CEO approves of that donation to FAH that he can't even write off for tax purposes. (no receipt)

Re:Doing this at work?

[Gewalt]

There's a HUGE difference between an idle computers power consumption and ones whose CPU and RAM throughput are being taxed to the limit by a process like folding. The 240 watts I mentioned is just the CPU, northbridge, RAM, and internal heat evacuation. Drives and monitor are completely irrelevant.

You execs are right to dismiss the notion of shutting down a computer thats idle. It's NOT consuming much. However, when that same computer is crunching foldings numbers for it.... THAT is a huge cost.

Re:Global Warming!

[boombaard]

hm.. this way you're directly investing in 'new' science, and you know what the goal is.. if you invest in amnesty/OxFam/whatever you know at least 20% is lost due to "overhead", another 10% at least is lost due to corruption, and even then (in the case of oxfam and related charities), there is a chance you're funding an organization that has more than a few members (statistically speaking, based on the amount of cases that have come out over the past 5 years or so) that indulge in sex-for-food programmes while they're doing their work. (That said, i do donate to Oxfam, because there just isn't an alternative i know that i know is better, and i'm hoping they're doing at least something with it that can be called useful.

Anyway, it is of course up to you (and i'll admit i'm somewhat cynical when it comes to those organizations), but if i had to choose, and if i had a choice, i'd rather invest in an @home project.. i find it a lot more intrinsically motivating than knowing i'm keeping a statistic alive that in 10-20 years might start earning their country some money through taxation because he's had his K-6 education.

Re:Global Warming!

[evanbd]

They're a university. I'm sure they have that taken care of.

Re:Global Warming!

[vux984]

hm.. this way you're directly investing in 'new' science, and you know what the goal is..

Fair enough. But its a little dishonest if you don't REALIZE how much you are invested. That's my biggest issue. Once people know what it costs I have no issue if they're still willing to contribute. But it bugs me, especially since I beleive the a very significant proportion of the people contributing to F@H are not the one's paying the bills.

The other part is how much do F@H results actually cost, in aggregate? Is it good value for the science produced? They've consumed between $50 and 100 million in electricity. Could they have made better progress towards their goals if they were given the money directly? At the very least if they built their own super computer and managed the costs directly the waste would be far far less.

Not only would they be paying industrial rates for electricity instead of residential rates, they'd also be using far less of it because they'd have racks of CPUs not all powering hard drives, and what not needlessly.

Hell, just take a look at the from their site: (For the purposes of this I've assumed that it costs 'volunteers' on average $10 to run a cpu per month in electricty.)

190,000 PCs generating 182 TFLOPs. 191k cpus. Total Cost ~1.9M/month. ~$10,494/TFLOP/month
  41,000 PS3 generating 1257 TFLOPs. 41k cpus. Total Cost ~0.4M/month. ~$326/TFLOP/month

What moron would keep the PCs running?

A final note about overhead. You lose 10-20% efficiency right off the top with F@H due to the lack a tax receipt. I can donate $250 to a registered charity at the same cost to me as buying $200 worth of electricty due to the taxes. Or conversely when you donate $200 to F@H -you- pay an extra 20-50 in taxes vs had you given the same $200 to a registered charity.

but if i had to choose, and if i had a choice, i'd rather invest in an @home project.. i find it a lot more intrinsically motivating than knowing i'm keeping a statistic alive that in 10-20 years might start earning their country some money through taxation because he's had his K-6 education.

Between those two I'm inclined to agree. I tend to mostly donate to small local organizations myself.

Re:Idea: F@H to help filter spam?

[religious freak]

Yeah, because Lord knows the spammers don't have spare CPU capacity at hand on all the hijacked machines they control.

It's worth a shot at thinking outside the box, but they have the CPU cycles and can likely hack past any kind of attempt to node lock the work units.

I suppose a minor benefit would be that some kind of work gets done before a spam message was sent out, but there's got to be a way to get past that requirement -- F@H is based on a measure of trust (and some cross-validation) that participants aren't gaming the system. With the email plan, the incentive is to offer incorrect results quickly, rather that accurate results slowly.

But like I said, at least it was a try at a different approach.

Re:Global Warming!

[Random Destruction]

Unless its winter. In that case 100% of the energy going into your computer is going towards heating your house. Sure electricity is more expensive than gas, but it'll drastically change your numbers for winter months.

Re:That's the stupidest idea I've seen all week.

[SuiteSisterMary]

Actually, a very similar system was tried; I don't know if it's still in any sort of wide-spread use (or as wide-spread as it ever got) or not.

Hashcash [wikipedia.org] involved calculating a hash, taking up CPU time, and sticking it in the email header. The recipient could easily verify that you'd spent CPU to send this message, hence, in theory, proving that you're not a spammer.