Folding@Home - Turning Up the Speed!

Departments of Chemistry and of Structural Biology, Stanford University ^ | August 4, 2008 | Dr Vijay Pande

[texas booster]

With the 6.20 (classic & GPU) and 6.22 (SMP) clients out, we (Dr. Pande and the F@H team) can start looking forward to the next steps in client development. We still have some last bits of work to completely unify the clients, but the hard part is already completed there for the most part.

The 6.2x series introduces several new features for donors, but in time, the clients have been getting gradually more and more complex to use. The Win/SMP and multi-gpu setups are examples of very challenging setups. Our primary plans for the future are to make setup much easier, especially for the very challenging clients (Win/SMP, multi-gpu). Our dream client install would be a single installer that would install either the classic, SMP, or GPU client based on donor choices during the install process, and the installer would take care of complex install situations (such as multi-gpu).

It may take a while to get to a client installer like that. In the mean time, we are looking to improve our installation instructions to make the process easier with existing clients.

We have spent a lot of effort developing these new clients and putting all of our effort into getting them to run well (the GPU2 client/core has gotten a great deal of effort, which has paid off in the transition from GPU1 to GPU2). It's now time to concentrate on usability and ease of installation.

Our hope is that with this completed, we will have a very advanced set of clients, but with very easy installation. That's easier said than done, but that's where we're headed.

[astyanax]

I've got a couple dozen machines, the units I'm considering being higher end, duo-cores w/ 8800's(512M), 2 and 4 gig system memory. It sounds like a very productive way to spend a Monday (certainly beats working!)
What would you recommend for an O.S.?
I've got images of WinNT through XP (Vista on 1 box) and numerous flavors of Linux. Newest image is Ubuntu (Heron), I believe (it's been a few months,) but I have access to everything and anything.
Any and all advice is certainly welcome and appreciated.

[astyanax]

Wow, a Dell iMac...
LOL!
My first thought also, but I (wisely) remained silent...

[astyanax]

Ah, the voice of experience!
Yes, it’s amazing how poorly electronics fare when you let out that “magic smoke”...

[texas booster]

If these are work systems then you have to dance with the OS that brung you.

Install XP and secure it thru SP3 for each system that can run an 8800 video card. The NVidia GPU will get you the most points.

Otherwise drop a Linux distro on them. I am not much help in Linux, and my son (Linux guru) has just moved off to college. We have several Linux pros around that can help.

Make sure that you use the same name and team number (36120) for every box. Please use your FReepname if you can.

For the systems that have an 8800, check the NVidia chart above and see where those cards place. Then move the cards to a quad core if possible, and dedicate a core to the cards. It takes one core to feed a GPU.

Look here for a new GPU installer that is beta but solves a few problems:

http://foldingforum.org/viewtopic.php?f=50&t=4806

For this month just get an 8800 running in each system. Dual GPU systems can wait for next month.

Once each system has a working GPU, then count the cores left over.

If you have any quad cores install the latest 6.xx SMP code from the main Download page.

For every dual core then install the 6.2x console version onto that system, into the folder F@H1.

Important notes:

You need to install each F@H package into its own folder. I use F@H 1, F@H 2, F@H SMP and F@H GPU.

ALWAYS select Advanced options when asked. On most other questions you can accept the default.

Set the “Accept Advanced Methods” to “y”.

Make sure that you keep a pattern to each box and which Machine ID that you assign them. I always set my GPUs to ID 4, and then console 1, console 2 etc to match the name of the folder.

Do not forget your FReepname and team 36120.

I will heartily recommend that you install the script from Egon in his above post. Works wonders.

[texas booster]

On your Vista installs always run these programs with the “Run as Administrator” choice from the right click.

Much misery and wasted time otherwise.

Please ping for questions.

[papasmurf]

“Are any of your old buddies from the truckers forum still folding?”

It appears that three of them are still with us.

Dylan has been up and down and, as of last month, was back in the hospital. I haven’t had an update in a while, and only hear 2nd hand of his current status. Finance have forced his Mom (Smile) to move to Southern Indiana to be closer to the medical facilities he needs.

[astyanax]

Will do, and thank you for all the assistance.
Let me get the cluster set up, do some tweaking, and then I will seek advice on optimizing.
What’s the word re:XP SP2? I’ve been (intentionally) avoiding SP3... (other than a stand-alone for compatibililty/ regression testing.)
I have a bank of boxes (mostly Win and Linux) at work, that although older (generally P4 or equiv.), might also be suitable.

[texas booster]

IMO, Win XP2 is the preferred evil empire distro.

Since you have a gang of fast NVidia cards, Winders is the way to go.

If you get 5 systems all running with GPUs, you will be the top points person on our team.

Klutz_Dohanger is the current overall points leader, but the energy czars at his work have cut him way back. If he can get the systems running again then he can get 80,000 - 100,000 points per week.

It takes about a day for the first GPU work unit to cycle through, and about a week for the first CPU core to register.

[astyanax]

Roger, Roger!
I’ll see what I can do.
If it’s a matter of bandwidth vs. “the energy czars” I should be fine. ;o)
80-100K points/ week?
How many machines does that entail?

[texas booster]

With CPU clients that is around 300+ systems.

With the new high performance clients one can do that with 25 systems, sometimes less.

[astyanax]

Well, I’ve been trying to justify a new quad-core or two...

[anymouse]

I keep my system on too much as it is, why would I want to pay a higher electricity bill just to let some science hackers take over my system in the dead of night? SETI at home was a novel fade, but that was in the days of cheap power.

[texas booster]

Well, the science generated by that higher electric bill might be a key to saving your life or the life of a loved one.

I did the SETI@home and Genome@home, as well as a couple of others years ago. I worked on G@H until it was completed, and I dropped SETI and other distributed computing solutions when my father started showing signs of Alzheimer's Disease. Folding@home has the ability to make a personal difference in our lives.

These programs for protein folding will not be a novel fade but will expand dramatically as we realize that some simulation problems cry out for lots of slower, cheap computers instead of a gigantic supercomputer.

You do not need to keep your system on 24/7 if you don't want to. Yes it gets more work completed but, as you noted, power isn't so cheap anymore.

So turn it off when you are finished with the computer and don't worry about it. The work units will still get completed in plenty of time to give credit for it and to help science progress. Plenty of folks only run their PS3 on weekends and they just fold then.

FWIW, the average computer running F@H will add between $3 and $10 per system over the course of a month, per DOE studies.

[houeto]

Bump.

[agooga]

It’s a worthy cause. I fold on my laptop at work— my company pays for the power, heh, heh!

[HKMk23]

Still got 4 CPU’s bangin’ away out here on the Left Coast.

Little by little does the trick.

Looking to configure a new system for the home office, though, so I will likely add an NVIDIA GPU to my folding ranks in the next couple of months. Still researching configurations to achieve optimal preformance/$$ ratio. I can accept that I probably won’t hit it dead-on, but I’d like to get close as far as my selections for the MB/CPU/GPU triad are concerned.

[texas booster]

It is certainly a moving target.

I favor the new Penryn Quad processors, as long as it is not my checkbook that is paying for it.

The ATI cards are supposed to have price decreases in the next month, which will put pressure on NVidia to lower prices ...

I also think that a new installer for SMP for Windows will be out by then.

A single quad core system with a fast GPU should be able to run SMP and a GPU kernel and get 7,000 ppd.

At least in theory.

[HKMk23]

A single quad core system with a fast GPU should be able to run SMP and a GPU kernel and get 7,000 ppd.

At least in theory.

That's a theory I just might have to put to the test.

[texas booster]

Here is a reason to buy that new power supply:

http://www.newegg.com/Product/Product.aspx?Item=N82E16814103064

DIAMOND 4870PE5512OC Radeon HD 4870 512MB 256-bit GDDR5 PCI Express 2.0 x16 HDCP Ready CrossFire Supported Video Card - Retail

$289.99 after rebate.

However, this card would rank among the worst in the price/performance category because it is priced for early adopters.

Early adopters = suckers like me.

[texas booster]

Thoughts on the NVidia vs ATI religious war, from the F@H forum:

If we look at worst case scenario figures, comparing like for like:

nVidia G92 (full version, not cut down GT/GS versions) = 128 shaders (unified - each shader can execute all 5 instruction classes)
ATI R700 = 800 shaders (specialized - each shader can execute one of the 5 instruction classes, which makes it equivalent (worst case scenario, assuming the workload is unparallelizable or the compiler is totally useless of taking advantage of this) to 160 shaders).

So: nVidia G92 = 128 shaders, ATI R700 = 160 shaders. nVidia G200 = 240 shaders.

Assuming all optimizations and compilers are of equal performance and ATI GPU cannot gain advantage from the compiler aligning parts of the execution out-of-order to use all available shaders all the time, this should roughly be the ratio of the performance between these GPUs, clock-for-clock. In reality, ATI's 800 specialized shaders should put it at least 2x ahead of the G200 because the unified equivalence is reached only in the worst case scenario where effectively only one same instruction is used repeatedly (unlikely in the extreme for any plausibly useful work).

Personally, I think we're all likely to die of old age before ATI make a compiler decent enough to leverage their GPUs to anything like their true potential - or for that matter, by the current evidence, even to the worst case scenario potential, as mentioned above.

Another thing worth considering is that nVidia are getting much higher clock speeds out of their 55nm chips (G92b in the GTX+ card) than ATI are getting out of theirs in the R700. Vertical scalability (clock speeds) is linear, whereas horizontal scalability (more pipelines / greater parallelism) is logarithmic - another fact that works against the R700. nVidia's brute-force approach is easier to optimize for and higher clock speeds provide better performance scaling.

The ATI 4870 has tremendous potential which will become realized as the Pande group writes better cores for the GPU utilization. Today it is a toss up in F@H production values between NVidia and ATI.

Of course, both cores are getting bonus points for their beta status so comparisons are difficult.

Our best guess is that either of the very high end GPU cards should provide about 110 GFLOPs of crunching prowess. That's a lot of power, and compare that to the recent past.