ati 4870 failing immediately

[Al]

Just bought a used 4870 for my 2nd rig for the express purpose of crunching MW. It works fine on Collatz, but fails all MW wu immediately. With updated drivers it would make it to 80% complete, then get a computational error. With the original disk drivers it fails at 1 second. I'm a Nvidia guy, so any help from the ati folks would be helpful. BTW, it is not OCed. Ask if you need more info, I'm sure I didn't give you everything you need for a diagnosis.
Al

xp 32 - Intel Quad q6600

[Crazybob]

Are you running or have you run any of the optimized aps? If there is an old ap_info.xml file floating around, that will do it. I am running 11.3 driver version for my 4870. Not sure if that has anything to do with it. I remember having that same problem, but don"t remember the cause, because I was dealing with a few issues at that time. I do run the same Quad though. I'll keep it in mind and if I think of anything else, I'll let you know.

[Al]

no app_info file. I did move it to another rig, with a better power supply thinking that might be the issue. No luck. All fail at between 50 & 75%.

[Crazybob]

What kind of error message does it show in the messages tab?

[Al]

7/15/2011 2:27:10 PM | Milkyway@home | Aborting task ps_separation_13_3s_free_2_1051303_0: exceeded elapsed time limit 98.80 (2960409.21G/29964.59G)

It aborts, finishes and then says "Computational Error."

[Crazybob]

That's interesting. Never seen that one before. I'm sure that the project doesn't care how long it takes you to crunch, so that can't be it. Must be something where it finishes one part of the WU and then needs to switch to something else, but is hanging up. Where's Maxwell or Crunch3r when you need them.

Well since I'm not helping, my advice would be to detach. Make sure that you delete the MW directory and then reattach. Maybe that will excorsize the ghosts out of the machine.

[Bryan]

IIRC there was a thread on MW discussing that several months ago. I'm thinking that the time limit was something of a protection thing because some wu were hanging up GPUs. I don't remember if there was a work around or whether they thought they had fixed it on the wu side.

[Crazybob]

IIRC there was a thread on MW discussing that several months ago. I'm thinking that the time limit was something of a protection thing because some wu were hanging up GPUs. I don't remember if there was a work around or whether they thought they had fixed it on the wu side.

I was just about to PM you on this one. Figured you were about the most knowledgable about it. If I get time I'll search on the MW site.

[Crazybob]

This thread suggests an app_info.xml file might fix the problem. It might be worth a try.

http://milkyway.cs.rpi.edu/milkyway/...?id=2468#49488

[trigggl]

7/15/2011 2:27:10 PM | Milkyway@home | Aborting task ps_separation_13_3s_free_2_1051303_0: exceeded elapsed time limit 98.80 (2960409.21G/29964.59G)

It aborts, finishes and then says "Computational Error."

Have you tried looking at the stderr and stdout files in the 'slot' folder while it's crunching? I've found some useful info doing that before.