Folding@Home Weekly Team News - 28th September 2006

Joe42 · 29 Sep 2006 at 21:12

I don't know if anyone else was aware of this:

Stanford said:
We are primarily beta testing the ATI GPU client software internally at the moment, but gradually releasing the client to collaborators and other closed beta testers. We will make a formal announcement of the client at the ATI Stream Computing Event in San Francisco on Friday September 29, 2006 (this is open for press personnel, not for the general public), and then (assuming everything looks reasonable), start an open beta for the console client on Monday October 2 with the GUI GPU client (with real time visualization) to follow.

Is this the same beta program for the wus or a separate thing?
Is anyone on this beta and able to get it in the next few days to try it out?

Edit: The official press release from ATI is here.

Edit2: So who actually has an ATI X1800 or X1900 gpu and will be able to install the new gpu client as soon as it becomes available?

This could make thigs very interesting both with team competition and user competition. Apparenty the ppd of these gpus will be very high indeed.

lay-z-boy · 29 Sep 2006 at 22:03

x1900xtx, x1300pro are the ati cards i have that should be able to do it.

If nvidia gets good support i have a go7900gtx & 7600gt.

Cant wait.

Joe42 · 29 Sep 2006 at 22:04

An update from the press release:

ATI said:
"To get involved with Stanford's University's Folding@home program and take advantage of the accelerated research capabilities of ATI processors, a Radeon X1900 GT, Radeon X1900 XT, Radeon X1900 XTX, Radeon X1900 CrossFire Edition, Radeon X1950 XTX, or Radeon X1950 CrossFire Edition is necessary along with the new Foldingkhome (graphics processing unit) GPU client and the ATI Catalyst 6.10 driver from ATI. Both the Folding@home GPU client and ATI Catalyst 6.10 driver are expected to be available Monday, October 2nd. "

If it does turn up on monday i think a thread in the graphics forum would be a good idea...

lay-z-boy said:
If nvidia gets good support i have a go7900gtx & 7600gt.

Cant wait.

They don't plan on doing anything just yet with nvidia cards sadly.

I'm going to add a section in the quick guide in anticipation of this appearing soon.

Edit: I've just noticed that press release sugegsts you can only run it on an X1900 series, not an X1800 series or anything else.

lay-z-boy · 29 Sep 2006 at 22:11

Joe42 said:
They don't plan on doing anything just yet with nvidia cards sadly.

I'm going to add a section in the quick guide in anticipation of this appearing soon.

Hence my 'if', *pokes you*

Joe42 · 29 Sep 2006 at 22:12

lay-z-boy said:
Hence my 'if', *pokes you*

Ow! *rubs side*

You won't be able to run it on your x1300pro, X1900 series only it seems.

rich99million · 29 Sep 2006 at 22:12

From the High Performance FAQ - here

Which GPUs will be supported? We have not made any final decisions on this issue. However, our software will likely require the very latest GPUs from ATI (especially now that the newest ATI GPUs support 32 bit floating point operations). Previous work of ours used NVIDIA GPUs as well, but we have now concentrated on ATI GPU's as they allow for significant performance increases for FAH over NVIDIA's GPU's (at least at the current generation). Our GPU cluster has 25 1900XT's and 25 1900 XTX's. We find a considerable performance increase of 1900XT's even over 1800XT's, due to the architectural differences between the R580 and R520 GPU's. Our code will run on R520's, but considerably more slowly than R580. We're very much looking forward to trying out R600's.

Other than that and the information you've already posted there is only really the specualtion in the GPU client / core issues forum

I haven't checked it out since my 9800Pro is going to be no good at all

From the sounds of it the client will be available to all since that section of the forum is also open

edit: Joe if you do go pimping please remember this is still in testing - you could do more harm than good getting a load of people running newly released clients

Joe42 · 29 Sep 2006 at 22:14

rich99million said:
<snip>

Thats old news, the quote from the press release i just posted was released in the last couple of days and they obviously haven't had time to update the faq yet.

Edit: The press release was made today.

Edit2: Not pimping it until its released and tested by current team members to check it works.

rich99million · 29 Sep 2006 at 22:17

Joe42 said:
Thats old news, the quote from the press release i just posted was released in the last couple of days and they obviously haven't had time to update the faq yet.

Edit: The press release was made today.

yes and the fact that they don't expect to be using this generation NVidia chipsets is still relevant

Joe42 · 29 Sep 2006 at 22:24

rich99million said:
yes and the fact that they don't expect to be using this generation NVidia chipsets is still relevant

Yes thats still the case, sorry.

rich99million · 29 Sep 2006 at 22:26

Just posted by Vijay

Vijay Pande said:
Tarx said:

That's a bit strange - I would have thought that cards like the X1800XT and X1600XT would work, especially since the X1900GT is listed...

Click to expand...

It's mainly about the processor on the card. The card names are very misleading, as a high number card doesn't mean a powerful processor (as far as FAH is concerned). All the 1900 class cards have the R580 processor. None of the others do. That's the deciding factor right now.

http://forum.folding-community.org/viewtopic.php?p=145364#145364

Joe42 · 29 Sep 2006 at 22:30

So its 100% X1900 series only?

rich99million · 29 Sep 2006 at 22:36

Joe42 said:
So its 100% X1900 series only?

at the moment yes - though I'm not sure what controls there will be on the client, maybe it'll just interrogate the card/drivers and only run if it finds what it's after

it should be interesting to see what happens - I can see myself having a new rig near the end of this year and then a fridge full of PS3's early next year

Joe42 · 29 Sep 2006 at 22:45

rich99million said:
at the moment yes - though I'm not sure what controls there will be on the client, maybe it'll just interrogate the card/drivers and only run if it finds what it's after

Seems likely as Ati are releasing a new driver to make it work. I would guess anyone can download it but it will only work on X1900 series. With the fast pace of gpu technology and upcoming new gpus i can't see them taking the time to develop it for older ati cards or current nvidia cards, although hopefully they will get it working on next gen nvidia cards.

I've been holding off on a gpu upgrade for a while now so once i see how this goes i think its upgrade time.

I should think initially there won't be many options, but the most important one for me would be a percentage utilisation setting as i doubt these cards will last long if run on full whack 24/7.

lay-z-boy · 29 Sep 2006 at 22:50

Tbh, as good as it would be to pimp the gpu client, i smell problems.
Overclocks, huge increase in heat, powersupply's, crashing etc.
I predict that is what will happen if you pimp it to people who think oc'ing to instability but stable enough for what they do is good.

I, myself will return to stock just so i know that that doesnt affect the client in anyway.

Still, its far too early to say anything or lay down rules etc.
Support is very, very small and its probably going to be hogged by many other folders with high end hardware, lets see what happens

Joe42 · 29 Sep 2006 at 23:00

lay-z-boy said:
Tbh, as good as it would be to pimp the gpu client, i smell problems.
Overclocks, huge increase in heat, powersupply's, crashing etc.
I predict that is what will happen if you pimp it to people who think oc'ing to instability but stable enough for what they do is good.

Yes it will be a bit of an experiment and i'm not going to be pimping it until you and others have had a chance to give it a go and see what its like. You don't mind blowing up your pride and joy for the good of the team do you?

Tbh I can't see it being stressed more by folding than by games in the short term, its the 24/7 or 12/7 operation that'll cause the problems i reckon. Which should be fine for most of the enthusiast crowd as they upgrade every year or two which won't give things like the cooling system time to fail.

I mean if you look at the cpu client, in the majority of cases if a game will crash folding will crash, and its only in a minority of cases where folding crashes but a game will not. I think the gpu client will be similar, but that minority will be even smaller as games stress the gpu more than they stress the cpu.

Are they folding the same wus as the cpu client or does this require different wus?

MGP · 29 Sep 2006 at 23:19

rich99million said:
I'd be looking at the RAM then maybe

what messages do you get - or does the whole system just crash?

edit: actually what are your temps like? Gromacs uses SSE but Amber does not use any optimisations
try running StressCPU which uses part of the Gromacs code and see if that replicates the problem for you

It's just folding that seems to reset, stops the WU, sends it back incomplete and goes on with another.

logs show things like:

Code:

[00:56:50] Folding@Home Gromacs Core
[00:56:50] Version 1.90 (March 8, 2006)
[00:56:50] 
[00:56:50] Preparing to commence simulation
[00:56:50] - Assembly optimizations manually forced on.
[00:56:50] - Not checking prior termination.
[00:56:54] - Expanded 3059285 -> 16583069 (decompressed 542.0 percent)
[00:56:54] - Starting from initial work packet
[00:56:54] 
[00:56:54] Project: 2414 (Run 62, Clone 14, Gen 8)
[00:56:54] 
[00:56:55] Assembly optimizations on if available.
[00:56:55] Entering M.D.
[00:57:03] Protein: p2414_Ribo_tryptophan280
[00:57:03] 
[00:57:03] Writing local files
[00:57:07] Extra SSE boost OK.
[00:57:08] Writing local files
[00:57:08] Completed 0 out of 250000 steps  (0)
[00:59:11] Quit 101 - Fatal error: NaN detected: (ener[13])
[00:59:11] 
[00:59:11] Simulation instability has been encountered. The run has entered a
[00:59:11]   state from which no further progress can be made.
[00:59:11] This may be the correct result of the simulation, however if you
[00:59:11]   often see other project units terminating early like this
[00:59:11]   too, you may wish to check the stability of your computer (issues
[00:59:11]   such as high temperature, overclocking, etc.).
[00:59:11] Going to send back what have done.
[00:59:11] logfile size: 7565
[00:59:11] - Writing 8128 bytes of core data to disk...
[00:59:11]   ... Done.
[00:59:11] 
[00:59:11] Folding@home Core Shutdown: EARLY_UNIT_END
[00:59:14] CoreStatus = 72 (114)
[00:59:14] Sending work to server

I was seeing messages in the logs

By using the -forceasm flag, you are overriding safeguards in the program. If you did not intend to do this, please restart the program without -forceasm. If work units are not completing fully (and particularly if your machine is overclocked), then please discontinue use of the flag.

that parameter had been set in the registry according to the sticky. I've turned forecasm off now to see if that makes a difference.

As I said it's not overclocked and temps don't seem an issue. Every thing else on the PC seems to keep going OK. It's an Athlon XP3500+, with 1GB DDR SDRAM (unbranded) on a Gigabyte KA GNS Ultra 939 Motherboard. I did have problems when I first put the PC together that it wasn't stable in some games until I upped the memory voltage by one point, a suggestion from somewhere on this forum, which seemed to cure it. Maybe that points to a memory fault somewhere?

rich99million · 29 Sep 2006 at 23:50

MGP said:
Maybe that points to a memory fault somewhere?

yes almost certainly it's memory related - usually if you are seeing NaN errors they can be remedied by backing off the memory timings a little
that would explain why upping the voltage has helped before as it has made the memory just stable enough for what you needed it to do at the time

you can probably keep the actual ram bus speed as it is - just slacken off the "tCAS-tRCD-tRP-tRAS" timings a little and then test

download and run CPU-Z first of all to establish what settings your RAM is currently working with and if there are any recommended settings programed in to the chips in the SPD section

Joe42 · 29 Sep 2006 at 23:54

MGP said:
Maybe that points to a memory fault somewhere?

Best bet is to run memtest a see if it passes ok.

A.N.Other · 30 Sep 2006 at 00:02

I've got a X1900XT and I'm on the F@H beta team, so I'll be logging on to get it as soon as possible. Monday is my first day at uni, though, so freshers parties may get in the way

.

A.N.Other · 30 Sep 2006 at 00:33

From a guy interpreting some data released about the GPU stats ...

OK, this looks to me like we're going to see the GPU production roughly equal to the performance of 4 core 2 duo's.

That's far from set in stone, but it is truly amazing. I have to get my card crunching now. I didn't know that it's actually an open beta, so everyone can download it as soon as it's released. With the number of techies in this team compared to others, it should give us a real boost.