GPU2 folding?

Snarkasm

I never, EVER get above 10% on the core I assigned to the GPU fold, but FAHmon is still reporting around 4k ppd, so I figure it's doing what it's supposed to. That's a very interesting comparison, good fodder for my anti-anti-Vista campaign.

sgstair

well besides it being essentially series 9 hardware I checked, looks like the biggest difference between the 8800GTS and 9800GTX is the clock speeds.
Obviously the GPU stuff is weighted really high at the moment, that will probably change, (weighted at 1500ppd on benchmark machine, as opposed to ~100-200 based on priority for CPU units) but for now the GPU is happily chewing through 480pt units in just over 2 hours.
But as long as folding doesn't affect other stuff I'm doing (videocard fan is still completely silent with this going on - heh), I really can't complain

_k

Ok so something interesting. I just got a new mobo to replace the same one I had that was making noise. One GPU client runs like it should per percentage though as soon as I kick in the second one it freaking kills the ppd. Currently I actually have no OC on anything, parts in sig.

Leonardo

Have you assigned core affinities for each GPU client?

_k

yes sir, it was harder to get the new ones set up to run together than last time.

_k

Ok interesting problem, just needed to delete more things to get both GPU2 up. When I started both of them they were stuck on the same core and I could not change the processes independently. Whenever I changed one the second took the affinity as well, never saw that happen with the beta. Had to give them two different cores and they seperated....thoughts or similar experiences.

Leonardo

When I started both of them they were stuck on the same core and I could not change the processes independently.

Process Lasso is your friend. You can lock processes to specific cores. You can also set it so that the processes you identify will lock to designated cores on process startup. The software works perfectly.

_k

Um I am using lasso, whenever I tried to just change the CPU affinity in there it wouldn't changed and I had to change the default setting to get it to move but it would change both instances.

Snarkasm

The problem is he's running two GPU clients, Leo. Since they both use the same process name, he can't change one and not the other using the default names.

_k

Actually I didn't have this issue before with the old clients, its one of the problems the new client has generated for me. And last time when I created the new folders I just let the duplicated WUs work out and pull new ones but this time I had to delete all of the logs and such to get the second instance to launch correctly and work.

Leonardo

Since they both use the same process name

OK, I'll smack myself now for lack of thinking it through.



(ouch)

_k

I am not sure why both instances were taking the same affinities even after changing one but not the other, some weird bug from the new version since the GPU and SMP seem to be more sensitive then their previous versions.

everyguy

_k_ wrote:

I am not sure why both instances were taking the same affinities even after changing one but not the other, some weird bug from the new version since the GPU and SMP seem to be more sensitive then their previous versions.

Maybe you could use the NV_FAH_CPU_AFFINITY environmental variable feature to allow the GPU clients to share all cores (or maybe two of them), and just set the smp client with process lasso to use just two of the cores. If you set the priorities on the GPU higher than the smp, you should be okay. The GPUs don't actually have to have their own personal core to work. There just need to be some core available when they need it.

_k

I finally loaded up the 177.35 beta drivers after getting both of the new clients running together and stable. Two 88 G92 running 675/1704/975,GPU/Shader/Mem, are putting out 10368.00 ppd effective on my rig. I found if you go up on the next divider, shader on air it errors to many WUs. mmm WU WU boarding the PPD train.

Snarkasm

Have you figured out how to run Riva on the 177.35, or are you doing it some other way, k?

_k

I am running Riva on the 177.35 for everything excluding when I play TF2 and FarCry if I ever go back to it. I was going to flash the bios on both cards to something like 675/1700/975 and not worry about it but I got lazy and couldn't find any kind of GUI or cmd flash that I could get to work so I gave up.....I might write a long post about it all tonight or in the next few days.

_k

Is there some way to put a preference on what project you want to do, cause 5504 takes less time than any other project I have done.

Leonardo

No, there is no way to register a preference.

bothered

Until _k_ suggested I try GPU2 folding I wasn't aware of it, I fold but (shame mode) I don't really monitor the forum for folding information, updates etc. I think this is true for most folders. Maybe GPU folding needs to be 'pushed' or at least advertised in the forum more. The difference in PPD is brilliant and a few more GPU folders could make a difference to team 93.

_k

Um so does the thread want an article to be submitted for GPU2 folding?...cause I started just never went past the intro.

Leonardo

does the thread want an article

Not a bad idea, at all! Contact Keebs and Prime.

_k

someone want to help me, i am actually typing something up keeb said to go with i just need some help editing and cutting down.

edit and post it back up highlight what you did or are thinking. This is a quick draft I did up just to get started its wordy and very strong.

Garg

I just started GPU folding on my laptop today. No problems so far, although I'm only a couple hours into it. I feel kind of fortunate to get it running since I didn't read the instructions

FahMon says it's making better PPD than the SMP client did. That surprised me a little since this is a mobile card. If it's more stable than the SMP client, I'll be especially happy.

BTW, I'm seeing the really low CPU utilization in Vista people were talking about. Usually 2-7%.

everyguy

bothered wrote:

Until _k_ suggested I try GPU2 folding I wasn't aware of it, I fold but (shame mode) I don't really monitor the forum for folding information, updates etc. I think this is true for most folders. Maybe GPU folding needs to be 'pushed' or at least advertised in the forum more. The difference in PPD is brilliant and a few more GPU folders could make a difference to team 93.

_k_ was talking about doing an article about it. I think that would be a great idea. Put it where everyone on Icrontic could see it. Probably a high percentage of Icrontic readers have high-end graphics cards that could run the GPU client.

sgstair

I was just lucky and bought a capable video card and reformatted my computer right after the client came out; am now seriously getting a second capable video card for another computer I have here (also for diabolical CUDA plans later on)

Leonardo

Hey, let's get this thread going again! For that matter, let's pack some more enthusiasm into Team 93!

If you have 3870 ATI or 8600 or higher Nvidia video card, join get it folding already!

As for enthusiasm... I just ordered two EVGA 9800GTs on sale and rebate at Newegg. Two more GPU2 folders will hopefully come online this weekend!

BuddyJ

Heck, I'll try it on my Lanbox. Lets see what an HD 3870 can do.

sgstair

I have a second 9800GTX on the way...

_k

Has anyone else seen that WUs are taking several attempts to get turned i. Mine are usually taking 4 or 9 attempts, while the SMP cores are being accepted first try.

Leonardo

It's server problems at Stanford. Pande Team is in the process of moving over all the assignment and collection servers from the Folding lab to Stanford's main server complex. They are experiencing difficulties with this.