validation inconclusive
18 Nov 2011 8:10:57 UTC
Topic 84778
(moderation:
this wu http://albertathome.org/workunit/11481 maybe hides some problems: i was unable to validate against a cpu but got validated against another ati opencl.
Maybe the brp3sse version really did some mistakes, maybe we both with ati are wrong and he was right...
validation inconclusive
)
I got some "validation errors" (wu marked as invalid) on my host.
It works ok, never had a problem with boinc or opencl (i work with gpgpu, so opencl runs well on my machine!), so maybe those are referring to some bugs in the ati opencl app?
I hope you are collecting and analyzing them...
A reply would be appreciated.
RE: i work with gpgpu, so
)
How do you know for certain? As that's what we're testing here at Albert. You cannot compare the running of other projects their OpenCL application with running one here. Different apps, different methods of compiling, different requirements, different use of what the validator should look for.
Since you had some validate errors, OpenCL does not run well on your machine. Not here, at least. Or the validator did not think they were well enough. Examples:
this task shows you paired against ATIOpenCLLion, but no match, while it and CUDA do match according to the validator.
this task shows you paired against one CUDA, but with too much problems between you two, so the task went out to two more CUDA, both of whom validated against each other. You and the original CUDA got none there. Interesting thing for the developers.
this task shows you paired against 3 other CUDA, one of which has a definite validate error, while the other two CUDA match and walk away with the credit.
Your ATIOpenCL work will validate against CPU (SSE) work and other ATIOpenCL work, it will also validate against other CUDA work, just apparently not against all ATIOpenCLLion (Macintosh) and CUDA work. That's something the developers here are looking for and into to see where they should fine tune the validator further.
It all depends on the wide variety of drivers used by you and the others.
RE: RE: i work with
)
just because I was not referring to BOINC at all, but to some code I run on my workstation related to my work. I was just saying that OpenCL "as a platform" works on my computer and seems to produce valid numbers if the code is valid (I'm unable to write bugfree code too!) and the drivers are good (11.11 had some problems, maybe some inconclusives were related to those?)
Anyway, I'm unable at the moment to use the card more on Albert. But I think that there're too many inconclusives and/or invalid just to be able to and say "the code is always correct in this new opencl app".
And yes, I'm sure the devs here are always working, I was just asking for a reply just to be sure about that.
I have 2 units in the
)
I have 2 units in the "validation inconclusive" state and a third one in "pending" state but the third or second unit is still "unsent".
Tullio
All these WUs have been
)
All these WUs have been canceled. But I finished a LineVeto search v1.03 SSE2 and it is still pending with no wingman. A BRP3SSE is running in high priority.
Tullio
I am wondering if there is a
)
I am wondering if there is a problem with validation related to Xeon processors?
Or certain Hosts?
I have an AMD Phenom X6 running Windows 32 bit XP and it has returned what appear to be 4 valid results but were marked "Validation Inconclusive" with the end result being I was marked as "Invalid".
A second AMD phenom X4 32 bit XP machine has another "Validation Inconclusive" still pending.
As does a AMD Phenom X6 running 64 bit Linux which has 2 "Validation Inconclusive" both still pending.
As all appear to have an Intel Xeon processor as the CPU that caused the initial "Validation Inconclusive" state, I thought that it might be related.
Also the same Hosts appear to be involved.
The other "Validation Inconclusive" work units still pending I expect to be invalidated as well.
It is confusing as all those hosts do not have a problem running the work units of this or other projects and the result outputs appear to my untrained eye to be fine with no errors.
See WU 184930
WU 186147
WU 186659
WU 186663
Have all been marked as invalid
Also WU 183758
WU 192546
WU 192545
Are all pending but expected to be marked Invalid.
Can these be looked at please?
Conan
I have a validation
)
I have a validation inconclusive on LineVeto on my AMD APU E-450 running Linux. No problem so far on my Opteron 1210 also running Linux. All 32-bit.
Tullio
Yep, all my invalid WUs are
)
Yep, all my invalid WUs are coming from "Gravitational Wave S6 LineVeto search (extended) v1.04 (SSE2)" running on Xeon E5620 CPUs (8 invalid / 674 valid / 532 pending).
It seems to be Xeon-related problem.
RE: Yep, all my invalid WUs
)
`Very likely.
Tullio
A message by Bernd in
)
A message by Bernd in Einstein@home says:
We found a problem in S6BucketLVE validation. Until this has been tracked down and fixed (or worked around) the validator is disabled.