PSA: Folding@Home now supports nVidia & ATI GPUs

[Foxery]

Posted with permission from ViRGE.
If I left anything out, or could clarify more, let me know and I'll edit as needed.
I'll try to keep this as short and to the point as I can.

Official announcement thread @ FoldingForum

I'm new here / What is this thing?
Folding@Home is a medical research project hosted by Stanford University which uses the combined power of many computers to analyze chemical processes in the human body; namely, the functions and interactions of proteins. Specific diseases/conditions that they hope to learn about include Alzheimers, Parkinsons, Huntingtons, and cancer. With help from nVidia and ATI, they have expanded this successful project to run its simulations on GPUs. (There is even a PlayStation 3 client!)

Any time your computer is idle, F@H grabs those unused CPU/GPU cycles to perform useful work. The program runs at low priority, so it will not typically interfere or slow anything down when you need to use your computer. Projects are split up by Stanford such that each machine receives a small portion of the work to be done, typically taking only a few hours (for GPU units) to a few days (some older CPU units) to complete and send back.

Supported cards
nVidia: GeForce 8000 series, 9000 series, GT260 and GT280 (Quadro support coming in the next version)
ATI: Radeon 2000, 3000 and 4000 series
Laptop GPUs will not run.
*** SLI or CrossFire must be disabled in order to run multiple cards. Dual-GPU cards currently only use 1 GPU, but this may be fixed in the future.

First and before you do anything else, read the FAQ!
NVIDIA FAQ or
ATI FAQ
A longer list of answers for a variety of questions (science details, results obtained, tech specs of the software) can be found here

Drivers
nVidia: Stanford recommends the 174.55 version of the drivers for now. You do not need the other files, just the drivers. Advanced users have reported better results with the 177.xx driver series.
ATI: Official Catalyst 8.5 and higher only. (Catalyst Control Center is not required.) Omega drivers or other 3rd party mods may not run.

Operating Systems
Only 32-bit flavors of Windows XP and Vista are officially supported.
For advanced users only: Link to instructions on running nVidia cards under Vista64.

Username and Teams
By default, the client contributes anonymously. By right clicking the icon in your system tray, you can configure a username for yourself, and join a team. AnandTech's team number is 198, if you would like to join us. (Shameless plug!)

Stanford University hosts the official statistics page for user contributions. You can check to see if your username is free by clicking "Donor Statistics."

For those who enjoy tracking numbers and comparing your results to other members, there are some more detailed stat trackers hosted by Extreme Overclocking and Kakao. Pretty graphs and predictions of future milestones are included.

Known bugs
It's a beta, and yes we meant that, not like Google but like software engineers - expect glitches and updates. The main thing is that the viewer will not work, as we're in the middle of changing how the live coordinates get to the viewer. Other than that, visit our forums to see what's happening.

Direct download link:
6.12 beta 6 ? Same client for nVidia & ATI owners

Official support forum for nVidia users
Official support forum for ATI users
Or, of course, our very own AnandTech Distributed Computing section. (here)

 

[bryanW1995]

this is way better than my thread.

btw, we are TeAm 198. this means you, keysplayr and anybody else with one of those new nvidia folding monsters!!

here's a link to the thread in the DC section: http://forums.anandtech.com/me...=2198265&enterthread=y

 

[Keysplayr]

Originally posted by: bryanW1995
this is way better than my thread.

btw, we are TeAm 198. this means you, keysplayr and anybody else with one of those new nvidia folding monsters!!

here's a link to the thread in the DC section: http://forums.anandtech.com/me...=2198265&enterthread=y

Hehe, I am still on the TeAm Anandtech World Community Grid. Protein folding and such.
I had at one point 12 systems ranging from a laptop to Dual Xeon Servers running that client.
I would be more than happy to run this F@H client on my 280 here.

 

[taltamir]

Only 32-bit flavors of Windows XP and Vista are officially supported.

Why is it so hard to support 64bit OS? it is really not that difficult...

 

[bryanW1995]

Originally posted by: keysplayr2003

Originally posted by: bryanW1995
this is way better than my thread.

btw, we are TeAm 198. this means you, keysplayr and anybody else with one of those new nvidia folding monsters!!

here's a link to the thread in the DC section: http://forums.anandtech.com/me...=2198265&enterthread=y

Hehe, I am still on the TeAm Anandtech World Community Grid. Protein folding and such.
I had at one point 12 systems ranging from a laptop to Dual Xeon Servers running that client.
I would be more than happy to run this F@H client on my 280 here.

sweet!! I'm getting over 2000 ppd with my 3870, a 280 should get well over 5k! now we need to get rollo doing this, I might even take it easy on him if...well, let's don't get crazy or anything...

 

[bryanW1995]

Originally posted by: taltamir

Only 32-bit flavors of Windows XP and Vista are officially supported.

Why is it so hard to support 64bit OS? it is really not that difficult...

one of our TeAmmates (Philippart) posted the workaround for 64 bit os. for some reason the 64 bit os's have some major driver signing issues still.

 

[Schadenfroh]

No linux support

 

[bryanW1995]

hey foxery, what about cat 8.6? do I need to keep it on 8.5 for now?

http://forums.anandtech.com/me...=2198747&enterthread=y

 

[Foxery]

Originally posted by: taltamir

Only 32-bit flavors of Windows XP and Vista are officially supported.

Why is it so hard to support 64bit OS? it is really not that difficult...

It's a university, not a large corporation... limited funding & manpower, and only a few ATI/nVidia staff helping out. I linked to a Vista64 hack in the OP, though.

Originally posted by: Schadenfroh
No linux support

Part of this is due to the Linux display drivers not supporting the features they need - at least with ATI's CAL. I'm not sure if it's the same case for CUDA or not. But again, limited resources, so they focus on the most popular platforms.

Oddly enough, some of the CPU code is the exact opposite - written on Linux first due to availability of certain programming tools, and then ported to Windows.

Originally posted by: bryanW1995
hey foxery, what about cat 8.6? do I need to keep it on 8.5 for now?

Yes, 8.6 will work.

Nighty night! :moon:

 

[adlep]

I have a question:
What are the crunch times of the WU when compared to the "regular" CPU client?
Is the GPU ver much faster than the CPU ver?
I imagine that the GPU ver will act similarly to the PS3 client, so it will be very fast....
CUDA is shaping up to be a nice idea. I wish AMD will adopt it for their GPUs as well for some new, universal "Direct X" type of instruction set for all GPUs.

 

[Cutthroat]

I used the workaround posted by Philippart found here to get it running on my Vista 64.

Wow is it ever fast on my 8800GTX which is at default clocks. FAHmon says 4500ppd for project 5001. It should take about 1/2 hour to complete a WU, and only taking about 20 sec/frame.:shocked:

Unlike I'd expected it is not using an entire core of my CPU, highest CPU usage I've seen so far is 15% on core 2, the other cores are idle. I think I can probably run 2 SMP clients as well.

EDIT: Yes I can run 2 SMP clients at the same time as the GPU2 client, and it does 8000 ppd on my PC

 

[ViRGE]

Originally posted by: bryanW1995

Originally posted by: taltamir

Only 32-bit flavors of Windows XP and Vista are officially supported.

Why is it so hard to support 64bit OS? it is really not that difficult...

one of our TeAmmates (Philippart) posted the workaround for 64 bit os. for some reason the 64 bit os's have some major driver signing issues still.

The problem appears to be driver related, rather than a problem with the application. It's probably just some stupid bug in the Vista64 drivers that no one found until now. It's not a signing issue.

 

[Keysplayr]

Originally posted by: bryanW1995

Originally posted by: keysplayr2003

Originally posted by: bryanW1995
this is way better than my thread.

btw, we are TeAm 198. this means you, keysplayr and anybody else with one of those new nvidia folding monsters!!

here's a link to the thread in the DC section: http://forums.anandtech.com/me...=2198265&enterthread=y

Hehe, I am still on the TeAm Anandtech World Community Grid. Protein folding and such.
I had at one point 12 systems ranging from a laptop to Dual Xeon Servers running that client.
I would be more than happy to run this F@H client on my 280 here.

sweet!! I'm getting over 2000 ppd with my 3870, a 280 should get well over 5k! now we need to get rollo doing this, I might even take it easy on him if...well, let's don't get crazy or anything...

Ok, I'm running the client for team 198. How do we check ppd?
It seems to be kicking an awful lot of ass.

Ah, FAH Mon. Well, it says 6513 PPD.

FAHMon Pic

 

[Foxery]

Originally posted by: adlep
What are the crunch times of the WU when compared to the "regular" CPU client?
Is the GPU ver much faster than the CPU ver?

The work units aren't quite the same, but in general, GPUs can crunch numbers 60-100x faster than an Intel CPU. After all, modeling the movement of 3D objects (proteins) is exactly what GPUs are built to do anyway!

I imagine that the GPU ver will act similarly to the PS3 client, so it will be very fast....

Some of the code is actually the same between the PS3 and GPU clients.

CUDA is shaping up to be a nice idea. I wish AMD will adopt it for their GPUs as well for some new, universal "Direct X" type of instruction set for all GPUs.

I don't think nVidia licenses CUDA to anyone else. AMD's equivalent is called CAL, which oddly enough, was originally created by nVidia and they sort of abandoned development in favor of CUDA.

The old GPU client, which ran on Radeon 1800/1900 cards, did use DirectX and was riddled with problems. Worst of all, it gave bad results. They shut it down shortly after releasing this version.

 

[Foxery]

Originally posted by: keysplayr2003
Ok, I'm running the client for team 198. How do we check ppd?
It seems to be kicking an awful lot of ass.

Ah, FAH Mon. Well, it says 6513 PPD.

Kick ass

http://fahmon.net
Here's the homepage for FahMon ("Folding@Home Monitor") which looks up your times in log files for any client, and adds up your Points Per Day (PPD).

EOC Stats and KakaoStats can generate all kinds of info about your past and future results. It takes about a week for their averages to make sense - i.e. when "zero" days stop being included.


Added to OP:
*** SLI or CrossFire must be disabled in order to run multiple cards. Dual-GPU cards currently only use 1 GPU, but this may be fixed in the future.

 

[Keysplayr]

I wonder how this would perform with 2x9800GTX's non SLI.
I'd have 256 shaders instead of 240. I'm going to give that a try in another rig. See how it goes.

Ah, FAH Mon. Well, it says 6513 PPD.

FAHMon Pic

EDIT: PPD seems to be going up. 7056 PPD. Does it depend on the WU?

FAHMon Pic2

 

[LOUISSSSS]

my FAHMON tells me that my 8800gt is only getting 1323.00* PPD

Monitoring Preferences is:
*Check* auto reload clients
Reload interval (mn) 1

*Unchecked* use experimental reload system

Calculate PPD based on Last 3 frames
*Check* Ignore asynchronous clocks

 

[Cutthroat]

Originally posted by: LOUISSSSS
my FAHMON tells me that my 8800gt is only getting 1323.00* PPD

Monitoring Preferences is:
*Check* auto reload clients
Reload interval (mn) 1

*Unchecked* use experimental reload system

Calculate PPD based on Last 3 frames
*Check* Ignore asynchronous clocks

If you open the viewer, how long does it say to complete a WU? FAHmon probably doesn't have enough info yet to make an accurate estimate.

 

[LOUISSSSS]

Originally posted by: Cutthroat

Originally posted by: LOUISSSSS
my FAHMON tells me that my 8800gt is only getting 1323.00* PPD

Monitoring Preferences is:
*Check* auto reload clients
Reload interval (mn) 1

*Unchecked* use experimental reload system

Calculate PPD based on Last 3 frames
*Check* Ignore asynchronous clocks

If you open the viewer, how long does it say to complete a WU? FAHmon probably doesn't have enough info yet to make an accurate estimate.

currently the F@hViewer says

Name: p5002_supervil
Progress: 60&
Performance: 800-1000(ish) iter / sec
Time Left: 41 mins

8800gt

 

[Cutthroat]

Originally posted by: LOUISSSSS

Originally posted by: Cutthroat

Originally posted by: LOUISSSSS
my FAHMON tells me that my 8800gt is only getting 1323.00* PPD

Monitoring Preferences is:
*Check* auto reload clients
Reload interval (mn) 1

*Unchecked* use experimental reload system

Calculate PPD based on Last 3 frames
*Check* Ignore asynchronous clocks

If you open the viewer, how long does it say to complete a WU? FAHmon probably doesn't have enough info yet to make an accurate estimate.

currently the F@hViewer says

Name: p5002_supervil
Progress: 60&
Performance: 800-1000(ish) iter / sec
Time Left: 41 mins

8800gt

That does seem kinda low, do you have it set to use 100% GPU?

 

[bryanW1995]

I used the workaround posted by Philippart found here to get it running on my Vista 64.

Wow is it ever fast on my 8800GTX which is at default clocks. FAHmon says 4500ppd for project 5001. It should take about 1/2 hour to complete a WU, and only taking about 20 sec/frame.

Unlike I'd expected it is not using an entire core of my CPU, highest CPU usage I've seen so far is 15% on core 2, the other cores are idle. I think I can probably run 2 SMP clients as well.

EDIT: Yes I can run 2 SMP clients at the same time as the GPU2 client, and it does 8000 ppd on my PC

I am locked at 25 % cpu usage (3 cores run seti 24/7). When I tried seti on all 4 cores and the gpu2 client on the side, I only got 500-750 ppd and seti didn't go down quite as much, but it was still less output than 4 cores. I did have a lot of eue's however, and I haven't seen any recently now that I've got a core dedicated to f@h.

my FAHMON tells me that my 8800gt is only getting 1323.00* PPD

Monitoring Preferences is:
*Check* auto reload clients
Reload interval (mn) 1

*Unchecked* use experimental reload system

Calculate PPD based on Last 3 frames
*Check* Ignore asynchronous clocks

that sounds awfully low. what cpu are you running?

 

[Diogenes2]

Where do I get FAH Mon to look for the GPU files ?

 

[Cutthroat]

Originally posted by: Diogenes2
Where do I get FAH Mon to look for the GPU files ?

here you go

 

[Foxery]

Cutthroat: The graphical viewer slows down the crunching, so FahMon is a "safer" way of checking.

Louissss: What driver version? Grab ForceWare 177.35 if you're on the 174 series, and watch it fly. Your system should be much, much higher.

Bryan: It's in his sig

 

[bryanW1995]

that's what I get for trying several replys at once...