Problems with my newly upgraded AMD X6 1090T host
22 Dec 2010 1:28:47 UTC
Topic 195530
(moderation:
Have received over 40 "marked as invalid" task within the last couple weeks after upgrading my hardware to an AMD Phenom X6 1090T processor. Nearly half of all my E@H tasks are getting this invalid status. Not seeing any issues with my other projects, only E@H. Is there a compatibility issue with the latest AMD processors on BOINC or E@H?
Cheers,
Gary.
Problems with my newly upgraded AMD X6 1090T host
)
Which date did you swap the new processor in, and what was the previous processor? Was it 64bit?
Cheers, Mike.
( edit ) 'There are only 10 kind of people in the world, those that understand binary and those that don't!' - I just got that. :-) :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I think my problem is solved.
)
I think my problem is solved. I down-graded back to BOINC 6.6.41 and haven't had a single "marked as invalid" task for several days compared with a nearly 50% failure rate on BOINC 6.10.58. Nice!!!!! Happy Computing to All and to All a Good Night!!!!!
There are only 10 kind of people in the world, those that understand binary and those that don't!
RE: I think my problem is
)
Well Done!! That's what we like. :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: RE: I think my
)
Well, partly.
I'd also like to see someone try and work out which of the differences between v6.10.41 and v6.10.58 caused this apparent effect - on his machine, if no-one else's.
If we can identify the problem, then we can watch out for it in the v6.12.xx line when it comes out. So far, I'm not seeing any problems like this with v6.12.8, but then I didn't see them with v6.10.58, either.
RE: I down-graded back to
)
I disagree. I see 1, 2, 3, 4, 6, 7, 8, 9, 10 tasks in trouble. As far as I can see, all your tasks are still not validating, they go from "Completed, waiting for validation" to "Completed, validation inconclusive" to "Completed, marked as invalid".
I have the feeling your CPU is damaged. If you can re-run your benchmarks, while other than BOINC Manager, not much else is running, and post the results here, that would be nice. In the mean time, I found 3 other AMD 1090Ts that have way better numbers than yours has: 1 T, 2 T, 3 T.
12/17/2010 8:48:39
)
12/17/2010 8:48:39 PM Running CPU benchmarks
12/17/2010 8:48:40 PM Suspending computation - running CPU benchmarks
12/17/2010 8:49:11 PM Benchmark results:
12/17/2010 8:49:11 PM Number of CPUs: 6
12/17/2010 8:49:11 PM 3259 floating point MIPS (Whetstone) per CPU
12/17/2010 8:49:11 PM 8046 integer MIPS (Dhrystone) per CPU
I agree, I spoke too soon. It's still hitting the fan.
There are only 10 kind of people in the world, those that understand binary and those that don't!
Now, while BOINC benchmarks
)
Now, while BOINC benchmarks can be easily influenced by outside influences, (I see a big drop in both floating point and integer when just leaving Winamp loaded in memory) it doesn't explain why your floating points are higher than the norm, while those integers are too low for your CPU. Are you under- or overclocking it in any way? I'd check in the BIOS and see if everything has been set up correctly.
12/18/2010 6:11:25
)
12/18/2010 6:11:25 AM Benchmark results:
12/18/2010 6:11:25 AM Number of CPUs: 6
12/18/2010 6:11:25 AM 3257 floating point MIPS (Whetstone) per CPU
12/18/2010 6:11:25 AM 8001 integer MIPS (Dhrystone) per CPU
12/18/2010 6:11:26 AM Resuming computation
Not currently over clocking, running at the standard 3.2 speed. Here is a second pass at the benchmark, similar numbers. No other programs actively running but a small number of programs in memory (no WinAmps) - this is primarily my number crunching machine. I ran some SiSoft Sandra tests last night and didn't find anything that looked abnormal. Any suggestions on better software for testing? Thanks for your help.
I used to run 1500 credits a day and have dropped to 300 with no end in sight. This is my favorite BOINC project and I'm frustrated that I cannot contribute after purchasing a 6-core monster CPU. Bad Carma!
There are only 10 kind of people in the world, those that understand binary and those that don't!
What you can do is run
)
What you can do is run Prime95 on it, which is a similar stress testing program. It'll test out what combination of numbers are Mersenne primes using your CPU, thereby stressing it out enormously. If that runs without problems (for a day), your CPU should be sound and it's something else (memory, motherboard, PSU, cosmic rays).
Prime95.
Don't run BOINC and Prime95 at the same time. Make sure to stop running BOINC while you test with Prime95. Let us know the outcome of that test run. Any errors, or automated reboots by the system are things we're very interested in. (As a healthy techie interest... I am not affiliated in any way or form with EAH. ;-))
I ran Prime95 torture test
)
I ran Prime95 torture test with small FFTs for hours with no errors, large FFTs generate an error in about 30 minutes, the blend torture test option also generates errors in a short period of time. I ran Memtest86 for 6 hours with no errors.
I'm not an overclocker so this is all new to me. Seems like the small FFT test is saying the CPU is okay and the Memtest86 is saying the RAM is okay. So what is in the middle that generates errors so quickly with large FFT testing? My AMD1090T X6 processor temperature is running between 52-56C during the torture test which is lower than the 60-62C I used to run with my AMD9850 X4 processor running fully compute bound. My AMD9850 never had an issue running E@H tasks but my AMD1090T is turning out to be a complete disaster!!
I cannot believe that I have to start tweaking core voltages and other parameters to get E@H to work on an off the shelf processor/mobo combination!
There are only 10 kind of people in the world, those that understand binary and those that don't!