@ TimeLord04 - I am the doctor trapped inside the TARDIS...
I have read that the OpenCL bug on OSX makes OpenCL completely unusable as kernels don't compile anymore. Looks you "just" have some invalid results, right?
For invalid tasks, I think the issue is A+B+C is not always equal to A+C+B. A, B, C and the SUM are computed in parallel. Each vendor (Nvidia/AMD/Intel) has its way of spreading the parallel tasks. So I though in case A+B+C != A+C+B, Nvidia card would validates themselves. Same for ATI. But it does not seem to be always the case. Definitely need more investigations.
@ TimeLord04 - I am the doctor trapped inside the TARDIS...
I have read that the OpenCL bug on OSX makes OpenCL completely unusable as kernels don't compile anymore. Looks you "just" have some invalid results, right?
For invalid tasks, I think the issue is A+B+C is not always equal to A+C+B. A, B, C and the SUM are computed in parallel. Each vendor (Nvidia/AMD/Intel) has its way of spreading the parallel tasks. So I though in case A+B+C != A+C+B, Nvidia card would validates themselves. Same for ATI. But it does not seem to be always the case. Definitely need more investigations.
I would actually go here: TBar's CUDA Apps to replace OpenCL on MAC at SETI.
Here, TBar, (and others), CLEARLY state that on MOST MAC Laptops, and some Desktops, and ALL Hackintoshes; that OpenCL being used in "Darwin 15.4.0" and newer, (El Capitan 10.11.4 onwards), that these computers experience HIGH numbers of Inconclusives, and INCREASED numbers of Invalids.
The remedy was TBar's CUDA75 App, for my system. No such App exists here at Einstein. Bernd, (in a prior post here), states he will investigate the MAC issues once he is through with the Windows FGRPB1G issues here.
Therefore, the problem IS OpenCL. AGAIN, I will continue to monitor and report.
[EDIT:]
Prior to OpenCL Units being crunched, (here and at SETI), my MAC/Hackintosh system NEVER had Invalids, PERIOD...
Under V1.16 running 1 per gpu/cpu with R9270x x2 with a FX 8350 the wu's are getting to aprox 20% then finishing, they are showing as valid. Anyone else seeing this?
Under V1.16 running 1 per gpu/cpu with R9270x x2 with a FX 8350 the wu's are getting to aprox 20% then finishing, they are showing as valid. Anyone else seeing this?
Yeah, it seems normal. My 270X is completing those tasks 2x in 11-12 minutes and 1x in 9 minutes. Looks like same as yours.
Under V1.16 running 1 per gpu/cpu with R9270x x2 with a FX 8350 the wu's are getting to aprox 20% then finishing, they are showing as valid. Anyone else seeing this?
Yes, actually on my EVGA NVIDIA GTX-760, on Win XP Pro x64, I get into the high 30%-age area, (roughly 34 to 35 Min. into crunching), and then it shoots to 100% and begins Uploading. All have validated so far...
As mentioned prior, I crunch 2 Units at a time.
[EDIT:]
Scratch that, I was thinking of my MAC timings on the two GTX-750TI SC cards... The 1.12, 1.13, and 11.4 Units are the ones that hit the high 30%-age area, (roughly 34 to 35 Min. into crunching), then shoot up to 100% and begin Uploading.
The 1.16 Units on the XP and 760 system run for 1 Hr 2 Min, to 1 Hr 3 Min, and then finish crunching at 100% So, actually, for me - the 1.16 Units act more "normal" under XP x64 than MAC and the 1.12, 1.13, and 1.14 Units.
The reason the percentage done isn't working as it should is because the app doesn't update it's progress to Boinc so instead of showing zero progress for the duration of the task Boinc makes a guess and starts showing progress. Obviously Boinc isn't that good at guessing progress so when the task finishes it jumps to 100%.
There's a file in the slot folder of a running task called boinc_task_state.xml inside that file is a tag called <fraction_done> that should be updated by either the task or Boinc but it always shows zero for the FGRP GPU app.
It currently looks like we will make a CUDA port of the app, but not this year anymore. If the 32Bit build of this works, we'll have a 32Bit app at least for NVidia as well, but probably never for AMD.
I also read that people with GTX750s have this problem so what of your concern about all the little people who contribute to your science project? My guess is that most will turn off the application or gpu as it make their computer unuseable.
I switched my browser Mozilla Firefox to Google Chrome and it works much better. I can play videos in HD it's running fine and scrolling is smooth. But don't know why!? I experience a little bit of lag in some apps but I can tolerate it, my first and biggest problem was the browser thing. It seems resolved.
Another "problem":
2016. dec. 19., hétfő, 12:28:43 CET | Einstein@Home | No work is available for Gamma-ray pulsar binary search #1 on GPUs
2016. dec. 19., hétfő, 12:28:43 CET | Einstein@Home | (reached daily quota of 24 tasks)
I think it's a little bit low for this type of GPU.
I think it's a little bit low for this type of GPU.
The currently configured maximum daily quota is 32*8=128 per core. You probably reported a lot of computing errors lowering your daily quota (this is to prevent misconfigured hosts to trash arbitrary numbers of tasks). Every successful result returned should double your quota.
I just doubled the quota for GPU apps (32*16) for the remaining time of the current "short" tasks (LATeah2003L), will set it back when we start issuing tasks that do run (roughly) as long as the BRP4G do.
@ TimeLord04 - I am the
)
@ TimeLord04 - I am the doctor trapped inside the TARDIS...
I have read that the OpenCL bug on OSX makes OpenCL completely unusable as kernels don't compile anymore. Looks you "just" have some invalid results, right?
For invalid tasks, I think the issue is A+B+C is not always equal to A+C+B. A, B, C and the SUM are computed in parallel. Each vendor (Nvidia/AMD/Intel) has its way of spreading the parallel tasks. So I though in case A+B+C != A+C+B, Nvidia card would validates themselves. Same for ATI. But it does not seem to be always the case. Definitely need more investigations.
choks wrote:@ TimeLord04 - I
)
I would actually go here: TBar's CUDA Apps to replace OpenCL on MAC at SETI.
Here, TBar, (and others), CLEARLY state that on MOST MAC Laptops, and some Desktops, and ALL Hackintoshes; that OpenCL being used in "Darwin 15.4.0" and newer, (El Capitan 10.11.4 onwards), that these computers experience HIGH numbers of Inconclusives, and INCREASED numbers of Invalids.
The remedy was TBar's CUDA75 App, for my system. No such App exists here at Einstein. Bernd, (in a prior post here), states he will investigate the MAC issues once he is through with the Windows FGRPB1G issues here.
Therefore, the problem IS OpenCL. AGAIN, I will continue to monitor and report.
[EDIT:]
Prior to OpenCL Units being crunched, (here and at SETI), my MAC/Hackintosh system NEVER had Invalids, PERIOD...
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
Under V1.16 running 1 per
)
Under V1.16 running 1 per gpu/cpu with R9270x x2 with a FX 8350 the wu's are getting to aprox 20% then finishing, they are showing as valid. Anyone else seeing this?
Mpat120 wrote:Under V1.16
)
Yeah, it seems normal. My 270X is completing those tasks 2x in 11-12 minutes and 1x in 9 minutes. Looks like same as yours.
Mpat120 wrote:Under V1.16
)
Yes, actually on my EVGA NVIDIA GTX-760, on Win XP Pro x64, I get into the high 30%-age area, (roughly 34 to 35 Min. into crunching), and then it shoots to 100% and begins Uploading. All have validated so far...
As mentioned prior, I crunch 2 Units at a time.
[EDIT:]
Scratch that, I was thinking of my MAC timings on the two GTX-750TI SC cards... The 1.12, 1.13, and 11.4 Units are the ones that hit the high 30%-age area, (roughly 34 to 35 Min. into crunching), then shoot up to 100% and begin Uploading.
The 1.16 Units on the XP and 760 system run for 1 Hr 2 Min, to 1 Hr 3 Min, and then finish crunching at 100% So, actually, for me - the 1.16 Units act more "normal" under XP x64 than MAC and the 1.12, 1.13, and 1.14 Units.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
The reason the percentage
)
The reason the percentage done isn't working as it should is because the app doesn't update it's progress to Boinc so instead of showing zero progress for the duration of the task Boinc makes a guess and starts showing progress. Obviously Boinc isn't that good at guessing progress so when the task finishes it jumps to 100%.
There's a file in the slot folder of a running task called boinc_task_state.xml inside that file is a tag called <fraction_done> that should be updated by either the task or Boinc but it always shows zero for the FGRP GPU app.
I just issued app vesrion
)
I just issued app vesrion 1.17 (currently OSX only, Beta test) that, amog other things, should have the progress counting fixed.
BM
It currently looks like we
)
It currently looks like we will make a CUDA port of the app, but not this year anymore. If the 32Bit build of this works, we'll have a 32Bit app at least for NVidia as well, but probably never for AMD.
BM
Pete_28 wrote:I also read
)
I switched my browser Mozilla Firefox to Google Chrome and it works much better. I can play videos in HD it's running fine and scrolling is smooth. But don't know why!? I experience a little bit of lag in some apps but I can tolerate it, my first and biggest problem was the browser thing. It seems resolved.
Another "problem":
2016. dec. 19., hétfő, 12:28:43 CET | Einstein@Home | No work is available for Gamma-ray pulsar binary search #1 on GPUs
2016. dec. 19., hétfő, 12:28:43 CET | Einstein@Home | (reached daily quota of 24 tasks)
I think it's a little bit low for this type of GPU.
(reached daily quota of 24
)
The currently configured maximum daily quota is 32*8=128 per core. You probably reported a lot of computing errors lowering your daily quota (this is to prevent misconfigured hosts to trash arbitrary numbers of tasks). Every successful result returned should double your quota.
I just doubled the quota for GPU apps (32*16) for the remaining time of the current "short" tasks (LATeah2003L), will set it back when we start issuing tasks that do run (roughly) as long as the BRP4G do.
BM