The GPU Folding Thread... X1900 owners in here.

Joe42

Joe42

Joe42

Soldato
Joined
18 Jan 2005
Posts
4,171
Location
Northants
As most of you probably know, stanford have been working on a gpu client and its due to be released today, afternoon pacific time which is 8 hours behind gmt, so sometime this evening, around 8.

There has been lengthy discussion of the gpu client in the team news, however now that the client is about to be released we need a thread where everything gpu related can be collated and discussed and those with X1900 series cards can compare results.
I'll also be writing a quick guide to folding on your gpu, and as i don't yet have an x1900 series card i'll need help from people who have an X1900 series card and can experiment with the client.
We can also try and answer all the questions we have about the client.


Download it here.

What we know so far:

-You need one of the following graphics cards:
Radeon X1900 GT
Radeon X1900 XT
Radeon X1900 XTX
Radeon X1900 CrossFire Edition
Radeon X1950 XTX
Radeon X1950 CrossFire Edition

-Support for X1800 series and older has been withdrawn, but may be added in future.

-You need the ATI Catalyst 6.10 driver.

-Folding on an X1900 series gpu gives 20 - 40x speed increased over a cpu.

-Points will be adjusted to reflect the massive speed gains of running on a gpu, however i believe points will still be better than running on a cpu. Points may be raised later is stanford feel tis required.

-Making the gpu client required a lot of developement, and as curren ati cards are better for folding a client for nvidia cards won't be available, but may be produced in future.

-The client is currently in beta, however although it has bugs it does work.

-There is a new core based on gromacs for gpus only.

-Folding only draws around 80w from each gpu, so its not as massively power hungry as it could be.

-The client will come in both graphical and command line versions.

-The gui client will be released sometime after the console client.

-An official faq and howto will be released with the client.

-The affect that this has on both users and teams will be very varied, with some teams having far more users with X1900 series than others.

-Gpu temperature when running folding is cool relative to the high temperature gpus are able to run at.



What we don't know, which we intend to answer in this thread:

-What effect does the ram on the graphics card have compared to speed i.e is it better to get a slower card with more ram or a faster card with less.
?

-Can you fold on multiple gpus?
Rumour has it than you will be able to, in the same way that you can with the cpu client, however this is unconfirmed.
Edit: The pc perspective article, says dual gpu support coming soon, the slides in the article are presumable from the stream computing conference.
?

-Can you adjust gpu utilisation?
?

-How is it affected by pci-e bandwidth? i.e could you run it from a card in a 4x pci-e slot with no slowdowns or does it require a 16x slot to work at full speed? If it can be run from a 4x or even a 1x or 2x slot this allows for many users to have a 3rd or even 4th card.
?

-What is the affect of folding long term on the lifetime of the card and the cooling system?
?

-What are the avantages and disadvantages of the gui client over the console client?
?

-Will it work with fahmon?
Yes

-Will the client automatically give up resources when required i.e. gaming etc?
?

Bugs:
A few, minor, read thread for detail.


Links:

Ati catalyst driver 6.10 beta.

The folding at home client download page (currently the download links for the gpu clients point to the old 5.02 cpu client)

Ocuk X1950 series page.

Ocuk X1900 series page.

Ocuk this week only page featuring one X1950 XTX and one X1900 XT

Stanford folding at home news.

Folding@Home Petaflop Initiative (FPI)

Folding@Home on ATI GPU's official FAQ

Folding@Home high performance client FAQ (with some gpu info)

Ati stream computing press release (including the official announcement of folding on gpu)

Anandtech: The GPU Advances: ATI's Stream Processing & Folding@Home

Pc perspective have a nice article on gpu folding here.

I intend to incorporate most of this into the quick guide to gpu folding which i've started and can be found here.


Questions to you:

Any suggestions for things we know, and is there anything else you want me to add to the above list of things we want to know?

Who has an X1900 series card, and which of you will be able to report back when the client is released to answer some of these questions above?

Who intends to buy an X1900 series card as soon as the client has been tested, or as soon as you know which card is best value for folding?

Personaly i can't wait to see what this client will be like, its been a long wait for anything exciting to happen with folding developement but this is definately interesting.
As you may have noticed, our team output is already at the highest level it ever been, thanks to the dedication of team members, and this gpu client can only make things better. ;)
 
Last edited:
I intend to keep the quick guide as small and 'quick' as possible, so once things have settled down a bit if you could do a full guide covering everything i think that would be great.

Edit: What other monitoring programs are there apart from fahmon? I've added fahmon to the list of questions... does that electron microscope program or whatever it is use the same method of using the unitinfo file?
 
Last edited:
Bigstan said:
Another thing for the need to know list.
Will the client automatically give up resources when required i.e. gaming etc.
Click to expand...
Added :)

That link to anandtech has an extra http at the front so it doesn't work. I've added a working link to the links section of my first post.
Edit: Beaten ;)

Interestingly enough we will be stomping team Anandtech in 1055 Days. :p
 
Last edited:
rich99million said:
Joe42 may already be working on it
Click to expand...
I hadn't started on it, so if Bti wishes to do it thats fine by me. :)
If you need any help with it let me know. You will need to remember that the majority of people in the graphics forum won't know what folding is, and they won't want to read more than a few lines, so you'll face the usual problem of explaining what it is and how to do it and why to do it without everyone getting bored and not bothering to read it. ;)

I'll add a bugs section to the first post, and update links to the new ati drivers in both the first post and the quick guide.
 
lay-z-boy said:
Dont mince your words, just they that most of them are dumb enough to argue over 3fps in a synth benchmark (stupuid) ;)

Most will give up after 3 lines, if it doesnt have an installer or results in little or no personal gain they will forget about it.

Its a fact, i cba to learn php, i gave up after 7 lines into the manual. :o
Click to expand...
The last two pimping threads i did took at least an hour each to write, you have to keep slimming it down more and more as you go until its acceptable. And still many people didn't read it.
This is why i work on the quick guide, but as i haven't finished it bti is going to have to make his own, unless he wants to wait till the quick guide is done and link to it... or perhaps even better both write a custom guide for the thread and link to the quick guide...

The problem is you only really get one chance, once you've made an impression on them anything with the word folding turns them away.
I would also suggest that you do not put folding in the title, as it doesn't interest people until they find out what its really like.
 
Last edited:
SiriusB said:
Just been having a thought while messing around with installing the new drivers.

Since both the GPU client AND the ATi Drivers that make the GPU client work are BETAs I would strongly recommend stopping your CPU clients from folding. You should do this via "services.msc" then back up any F@H folders. You can then carry on folding after you have sorted out the GPU client.
Click to expand...
I hoping yourself and others can try running both together and see if there are any problems.

Edit: Just spotted this:
Stanford said:
We will release an open beta of our client for GPU's on Monday Oct 2 in the early afternoon (Stanford time).
Click to expand...
Earlier rather than later. Stanford time is currently 12:30...
 
Last edited:
Its here!

And there are 3 versions, including a gui without fancy graphics.

Edit: The first bug: People without an ati x1900 series who try and run the console get an error message but the window closes too quickly for it to be read. Anyway, i get the gist of what its trying to say. :p
 
Last edited:
The gui client looks exactly the same as the cpu one, and does seem to include gpu utilisation adjustment. However its still labeled as cpu utilisation which makes me wonder if it actually works.

Thats about all i can do now guys, the rest is up to you lucky X1900 owners. Not a moment to loose i say. :p


Edit: Bug 2: If you press cancel on the gui client it pops back up again, you have to press ok to get rid of it.

Someone with an X1900, can you tell us what the options are for the command line client?
 
Last edited:
Is yours working now?

What might be good is if a few of you give your gpus their own username so we can compare outputs. I could keep a census of which is which and we can compare different models against each other.
 
This is the place to keep an eye on.

I'm very dissapointed. Not only is it buggy as hell, the ppd is less than many of the cpus it has to rob rendering it a step backwards in terms of output.
 
I suppose it is a bit of an experiment so you just have to try things until you're happy with it.
You could try lowering the gpu utilisation a bit more and see if that helps.
It occurs to me that Vista might be a spanner in the works for this, as it will be using the gpu all the time for rendering its fancy(read: pointless) effects.
 
I'd say this isn't stable enough for mass adoption and a pimping thread, so that'll have to go on hold i think.

Seems like they are making good progress on the bug fixing over at stanford thanks to concorde and others, so thats good.

My only issue is the poor ppd, in most cases it seems its better to leave the gpu client off and fold on the cpu core that it would otherwise rob, which is a shame considering the amount of effort required to get the thing working in the first place. Hopefully they will increase the points and make it more worthwhile.
 
So is there a proven formula to get it working it does it require individual tweaking in every case?

So if they raise the points and fix a few of these issues its all good. The points is the really annoying thing, we were promised big points.
 
You must log in or register to reply here.
Back
Top Bottom