Posted on 01/07/2007 11:28:30 AM PST by texas booster
Also, please feel free to bump the fund raiser and get us over the FReegoal!
https://secure.freerepublic.com/donate/
Thanks!
Bump, on a slow Sunday.
We finally passed AOA Forums for # 56!
Guess that Free Republic is flying higher than the pilots.
Actually, you're thinking of AOPA.
In tracing back to AOA's origins, I found them here.
Just got an ATI 1950 and seem to have followed the instructions, but I keep getting EUE (or is it UEU) on every work unit immediately (went from 600 WUs to about 1300 overnight (no points from them). Any idea what might be happening.
I went through Stanford's website to get to ATI and the only available driver was 7.1. Is this the problem?
I'll defer to someone else on this. I have no clue about he ATI cards.
The problem is the Catalyst 7.1 driver. Check out the comments on Catalyst versions:
"This is relevant to anyone folding with a GPU at present.
"The new Catalyst drivers released today 2007/01/10, version 7.1 (8.33) do not work with regard to folding on your GPU. The new drivers will cause the folding GPU core to continuously EUE on WUs, producing an "mdrun_gpu returned 114" error.
"The only versions of Catalyst that work with the GPU client are:
6.5
6.10
6.11
"The GPU client will work with these drivers with a significant performance hit:
6.6
6.7
"The following drivers do not work at all with the GPU client:
6.8
6.9
6.12
7.1"
And for software gurus that enjoy writing low level drivers, here's a little tidbit for an ATI engineer:
"It appears to be a compiler bug that's still standing. I thought we had a fix going into 7.1, but it apparently won't make it until 7.2. The issue seems to be that drivers go to QA long before they get released, and if we don't catch the bug early enough along, we miss getting a fix in the release. The other issue is that we've been running down other issues, like attempting G80 support, so we've been lax about updating the regression test we've given AMD. Just to annoy us and them, the 7.1 driver does pass the regression we gave them, but the current WUs take a slightly different code path."
Drop back to 6.11 and all should be fine.
From the Folding Team at Stanford:
"1/22/2007 Planned shutdown temporarily today (again)
We may have to shut down some FAH elements (stats updates and/or web pages) due to a major power overhaul in our server room, yet again. More info as we get our final schedule."
And everyone knows that there are NEVER any problems with power usage in a data center. Why, just add an extension cord ...
Stepped back to 6.11 and the EUEs have ended, but the GPU Server doesn't seem to be starting (Starting GPU Server text, but nothing following).
Any advice?
JosephW
Well, my run has ended. My brother is losing control of the two Macs.
Bring it on, JosephW!
;-)
I'm trying, but that ATI card is causing me grief. How am I ever expected to pass you, if my new toys don't work? (another new toy later this week - dual core, but lower speed laptop). The desktop with the ATI was built by my students in an A+ class this last week (slave labor)
JosephW
Which ATI card do you have? X1950 Pro? How many pixel shaders (36 or 48)? How much RAM (512 MB is faster than 256 MB)?
Uninstall the GUI folder. Reboot Install fresh from Stanford the GUI 5.91 beta 2.
Fire it off. Make sure that the user name is Texas Booster - I can see the points more easily that way to provide the best support. :}
Probably a clean start is what you need. Make sure that (for now) just load the one GUI client and the ATI card have 100% access to the CPU. Noother console or anything.
You should be processing a frame every 8 - 12 minutes, and a work unit every day.
Make sure that you have Egon's monitor loaded with a GUI description so we can watch it go.
Unfortunately, it's the Pro with 36 shaders and only 256MB.
Cleared out the old and installed the 5.91 beta 2 in console mode.
Will install Egon's monitor as soon as it runs right.
Tried to use Texas Booster, but the system just didn't like that. Thought it was some kind of virus, or plague (whatever it was it wasn't good :) )
JosephW
Looks like it finally started. I'm also running the CPU console version but at a lower priority than the GPU version.
Doesn't look like the GPU really uses that much of the CPU and I'd hate to waste those clock cycles when catching up to Egon :)
I suspect that the only real problem except the slow start was having the username set to Texas Booster. Seems like this is reserved to test EUE's and not meant for real computing :p
If you should find that test name in use change it to josephw and I'm sure we'll see that name rise quicker in the stats.
BTW any news on the PS3 client (such as an actual date - I might need to get my boys a toy)
Great! Seems that the SMP/GPU team is continuing it's asinine method of reporting the project name/number incorrectly in the unitinfo file.
It's reporting 2158 GPULambda which is the name and number of atoms of exactly 9 different WUs. Sigh.
If you can look in your FAHLog.txt file and give me the actual project number, I can have FAHMonitor refer to it correctly in the future.
Project 2736
New problem. GPU crashes when starting up as a service. Looks like I'll have to run it in a command prompt until the next beta revision :(
JosephW
Bummer. If it runs then we will be happy with that.
I know that one needs at least 25% of the CPU just to process requests into the GPU core. I would probably put the console at 50% and the GPU using the rest.
I haven't heard of GPU service crashes but I'll look into it later.
Well I gave the GPU version above normal priority and the CPU version Low. I figure the GPU will take the CPU whenever it needs to (since it's faster, it's welcome to it).
Bump for later reading.
Bogus!
I updated my "bad project names" list. On its next reporting, Constellation should correct itself. Thanks!
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.