Lets get some gpu folding done

**Sparky** · 10-03-2006, 04:10 PM

Originally Posted by Bloody_Sorcerer

man this wu is slow

been running over 24 hours now and im at 31. something sounds wrong...

Yeah it should only take around 12-14 hours. Do you have a full CPU core dedicated to it? You can't share a core between the CPU client and GPU client I found.

**ewitte** · 10-03-2006, 05:38 PM

Originally Posted by thewildblue

Wonder if we could mod an INF file to report the X1600XT as an X1900 purely just to crunch ?

I just used a hex editor to change their beta to 1800. It gives me early work units. I'm downloading the recommended driver version to test that.

**ewitte** · 10-03-2006, 05:43 PM

I seem to have it running now on an X1800

No telling how fast though.

Edit. "completed 2" so far 16 minutes each on 2725.

**enroscado** · 10-03-2006, 08:46 PM

ewitte, can you upload your edited beta so in can give it a try with my x1800?

Edit: Btw, what driver version and what client are you using?

**jimwah** · 10-03-2006, 11:51 PM

My x1900gt is currently running a 2711 (Run 1, Clone 222, Gen 0) at about 11mins20secs a frame. XT's sound like they are doing about 8mins a frame, and your x1800 is doing 16mins a frame (if these projects are comparable, certainly have the same credit, 220pts). My GT result maybe a little lower than usual tho' 'cos mine is under stock clocks iirc - 75mhz under on the mem or GPU, came out of the box like that

I let mine run again overnight, fully expecting an EUE, but it ran ok & was at frame 78 when I got up - I'm on the 6.10's here.

**Sorrow13** · 10-03-2006, 11:58 PM

Im in, been running for a bit.
cat 6.10s, latest dx9, console version. DLd the 'fancy graphics version' just to see how it looks. reminds me of the ps3 client screens. using stock clocks for now, make sure i know how the system works and then ill crank it up a bit.

think core speed or mem speed makes a bigger dif in crunch speed? Id say core, but ive no idea how the internals of the program work.

**ewitte** · 10-04-2006, 02:12 AM

Originally Posted by enroscado

ewitte, can you upload your edited beta so in can give it a try with my x1800?

Edit: Btw, what driver version and what client are you using?

I'll send you a PM. If anyone wants it send me a PM. I don't want to put it public. Its really easy to do yourself though. Use a hex editor and search for X1900 and change the first one to X1800. The second time it shows up is just the message you get if you don't have one. I backed up to catalyst 6.5, it would not work with what I had (6.9?)... also I needed the April DX SDK update (there is a link on the folding website).

BTW changed to my 3D clocks and voltage and its now 13 minutes per frame. I'm on frame 36.

**3NZ0** · 10-04-2006, 08:25 AM

6.8/9 have no support what-so-ever

Done 2-3 wu's now on the gpu, working perfectly

**enroscado** · 10-04-2006, 09:25 AM

OK, i got it running. So far 19 minutes for each step on my x1800xt. Does that sound about right for a 2725?

**pythagoras** · 10-04-2006, 09:40 AM

Any alteration of the f@H client is a violation of the eula, this normally ends in the user accounts being zeroed

Regards

John.

**cadaveca** · 10-04-2006, 10:00 AM

Originally Posted by nn_step

good advice. Anyone tried Crossfired X1900XTs yet?

Just slows things down. abt a 40second penalty.

**enroscado** · 10-04-2006, 10:15 AM

Originally Posted by pythagoras

Any alteration of the f@H client is a violation of the eula, this normally ends in the user accounts being zeroed

Regards

John.

TBH, i haven't read the EULA. Going back to old "just cpu" F@H client.

**jimwah** · 10-04-2006, 10:17 AM

Completed a 2711 today on the Cat 6.10's

It halted at 85 frames then jumped back into life a few hours later & uploaded a WU.

It also downloaded a new core before starting a new WU, so we'll see how this one does.

**nn_step** · 10-04-2006, 10:19 AM

Originally Posted by cadaveca

Just slows things down. abt a 40second penalty.

ouch, sounds like there is no advantage to use a crossfire board...
Which means I have some nVidia chipsets that could use some extra x1950XTX goodness

**Bloody_Sorcerer** · 10-04-2006, 01:12 PM

my x1900GT is doing around 11m30s per % on a 2725. stock clocks of 500/600 and surprisingly low temps with fan @ 100%

**hoax** · 10-04-2006, 04:42 PM

I just finished my first gpu folding attempt

It took 8 minutes per step on a 2725 (total 85 steps).. x1900xtx...stock clock and fan..

apparently it went without any problems...but I was sleeping so I dont know for sure..

**jimwah** · 10-05-2006, 12:50 AM

Have a look through the FAHlog.txt file in your GPUclient's folder...

[11:37:07] Folding@home Core Shutdown: FINISHED_UNIT
[11:37:10] CoreStatus = 64 (100)
[11:37:10] Sending work to server

I think a CoreStatus = 64 (100) means that all went as planned, certainly did here anyway

**sc00p** · 10-05-2006, 08:36 AM

Have you guys with gpu folding checked (&set) with AtiTool voltage to be higher (what was is 1.4V... shait I'm in hurry... do not have time to get facts straight) Might prevent some of the EUEs

**Scimitar** · 10-05-2006, 09:22 AM

Originally Posted by pythagoras

Any alteration of the f@H client is a violation of the eula, this normally ends in the user accounts being zeroed

How exactly does Stanford know if you've modified the GPU client? I've been running a hacked / optimized client for CPU folding for 5 years now and never been zeroed out.

**cadaveca** · 10-05-2006, 10:06 AM

Originally Posted by nn_step

ouch, sounds like there is no advantage to use a crossfire board...
Which means I have some nVidia chipsets that could use some extra x1950XTX goodness

I think it's becasue the cpu is then busy feeding two cards instead of one, as you can see temps increase a bit on the card that isn't used. I think it just replicates the data and does no work...as the composting engine is external to the gpu, there's no way for load-balancing and merging of the work to happen. Maybe Rv570 might make this a bit different, but then again, maybe not. I tihnk the framebuffers need to be shared, or at least have access to one another, for this to work right.

**Sparky** · 10-05-2006, 10:25 AM

Originally Posted by Scimitar

How exactly does Stanford know if you've modified the GPU client? I've been running a hacked / optimized client for CPU folding for 5 years now and never been zeroed out.

lol you just busted yourself

**Bloody_Sorcerer** · 10-05-2006, 02:18 PM

Originally Posted by cadaveca

I think it's becasue the cpu is then busy feeding two cards instead of one, as you can see temps increase a bit on the card that isn't used. I think it just replicates the data and does no work...as the composting engine is external to the gpu, there's no way for load-balancing and merging of the work to happen. Maybe Rv570 might make this a bit different, but then again, maybe not. I tihnk the framebuffers need to be shared, or at least have access to one another, for this to work right.

they should run as a multicore system would; the two clients wholly independant of one another.

**nn_step** · 10-05-2006, 02:24 PM

Originally Posted by Bloody_Sorcerer

they should run as a multicore system would; the two clients wholly independant of one another.

sounds about right to me, however you have to make sure you don't have the system set on "crossfire mode"

**cadaveca** · 10-05-2006, 02:27 PM

Uh, I agree, however it still doesn't work. I'd love it if it would...1000ppd it would be!If they could ease the cpu load so you could run a cpu client as well...

**STEvil** · 10-05-2006, 07:40 PM

or you need to only be running the GPU clients since it seems to take one cpu core per gpu core used?

Thread: Lets get some gpu folding done

Thread Tools

Search Thread

Rate This Thread

Display

Bookmarks

Bookmarks

Posting Permissions