I originally started a thread over in bug reports; but with my skylake laptop and the 2015/11/21 (10.18.15.4281) driver I'm having having WUs fail any time the screen is blanked. What appears to be happening is that the WU being ran when the laptop blanks its screen (it was set to blank screen but never go to sleep on wall power) ends up failing validation and new BRP4 (Arecibo) tasks run until they hit a 9 hour timeout and fail.
I've changed my power profile to keep the screen on overnight; to see if that produces a batch of good results (and proves my GPU is capable of working right).
If there's still not a good fix for this, is the problem something that will affect many/most projects; or could I put my laptops IGP safely to work running something else?
I originally started a thread over in bug reports; but with my skylake laptop and the 2015/11/21 (10.18.15.4281) driver I'm having having WUs fail any time the screen is blanked. What appears to be happening is that the WU being ran when the laptop blanks its screen (it was set to blank screen but never go to sleep on wall power) ends up failing validation and new BRP4 (Arecibo) tasks run until they hit a 9 hour timeout and fail.
I've changed my power profile to keep the screen on overnight; to see if that produces a batch of good results (and proves my GPU is capable of working right).
If there's still not a good fix for this, is the problem something that will affect many/most projects; or could I put my laptops IGP safely to work running something else?
Ended up having a few GPU tasks go on runaway even with the screen on; so my initial theory is wrong. I'm trying soem SETI tasks now since they were the only other Intel GPU project I could find (Collatz looked like it supported it, but the server reported no tasks available unless I switched CPU on for them).
New driver 4352 for HD Graphics 530 dated the 23rd of December has appeared on downloadcenter.intel.com. The summary of changes makes no mention of OpenCL.
Currently running Seti on it and will then run some Einstein tasks. Link: Host 6274635
New driver 4352 for HD Graphics 530 dated the 23rd of December has appeared on downloadcenter.intel.com. The summary of changes makes no mention of OpenCL.
Currently running Seti on it and will then run some Einstein tasks. Link: Host 6274635
Not looking too good. Currently got 16 inconclusive, 3 valid and a bunch more pending.
New driver 4352 for HD Graphics 530 dated the 23rd of December has appeared on downloadcenter.intel.com. The summary of changes makes no mention of OpenCL.
Currently running Seti on it and will then run some Einstein tasks. Link: Host 6274635
Not looking too good. Currently got 16 inconclusive, 3 valid and a bunch more pending.
Failing on both projects or just here?
On my skylake laptop the december driver is behaving a bit better in general use on W10. It went from occasionally timing out while rendering FF - forcing the OS to restart it and requiring a window resize to get it repainted normally (instead of all gray) - to occasionally taking a broad curved divot out of the top of the tab bar until a repaint was triggered.
I'm away from home, so haven't tried crunching anything with it yet.
Not looking too good. Currently got 16 inconclusive, 3 valid and a bunch more pending.
Failing on both projects or just here?
Seti seems to be about 50/50. still got one pending over there.
I'd say it doesn't work.
I'm not sure how to log a bug with Intel about their drivers seeing as I'd have to describe the problem, just saying they don't validate against other work units doesn't help. I'd need some help from the project (result files maybe) to be able to log a bug.
Not looking too good. Currently got 16 inconclusive, 3 valid and a bunch more pending.
Failing on both projects or just here?
Seti seems to be about 50/50. still got one pending over there.
I'd say it doesn't work.
I'm not sure how to log a bug with Intel about their drivers seeing as I'd have to describe the problem, just saying they don't validate against other work units doesn't help. I'd need some help from the project (result files maybe) to be able to log a bug.
I started a batch of seti ones. SO far 10/39 validated successfully after getting the first result back, the other 29 were inconclusive. 9 of those have had a 3rd result back; 8 resulted in all 3 people getting credit and 1 invalidated my result.
That makes me suspect the issue on seti is cumulative rounding differences pushing the validation thresholds to their limits. However unless needing a 3rd result to validate is common, my GPU's only effectively contributing at 25% of its theoretical power. Which is probably not worth stressing the cooling and limiting CPU turbo levels by a few hundred MHz. I'll do a batch of E@H GPU tasks next; but I think the GPUs going to have to be shelved until another driver update is out.
Edit: Just saw a notice that seti's starting to roll out a new app. Guess I'll try that too before throwing in the towel.
Quick note: on my HD530 + Win 10 and driver x.4300 SETI is working fine. I had invalid WUs every now and then, but with an acceptable quota of <10% with my i7 6700. Since it broke (too much cooler pressure) I switched to an i3 6100 with an HD530, too, and it didn't produce a single invalid WU yet.
Regarding Einstein on Skylake GPUs: thanks for testing the beta drivers, Mark! We had been talking about that over there, but had no solution for driver 4300 and earlier. I'm briefly describing though, how I ran an offline test according to someone else's instructions. It did show massive differences between the reference result and what the Skylake produced.
I also filed a bug report at Intel (not easy) but didn't hear back yet, apart from the fact that they had escalated the issue (i.e. found no quick fix).
Quick note: on my HD530 + Win 10 and driver x.4300 SETI is working fine. I had invalid WUs every now and then, but with an acceptable quota of <10% with my i7 6700. Since it broke (too much cooler pressure) I switched to an i3 6100 with an HD530, too, and it didn't produce a single invalid WU yet.
Regarding Einstein on Skylake GPUs: thanks for testing the beta drivers, Mark! We had been talking about that over there, but had no solution for driver 4300 and earlier. I'm briefly describing though, how I ran an offline test according to someone else's instructions. It did show massive differences between the reference result and what the Skylake produced.
I also filed a bug report at Intel (not easy) but didn't hear back yet, apart from the fact that they had escalated the issue (i.e. found no quick fix).
MrS
Did you have any performance issues while running SETI? While running that on the GPU the FGRP tasks on my laptops CPU were taking ~2x as long. ~15% of that could be attributed to the CPU clock running at 2300mhz instead of 2700 (presumably SETI was loading the GPU harder and eating more of the TDP) but the FRGP tasks spiked from 35-40k seconds to 80k.
I'm seeing "massive" main memory bandwidth consumption by SETI: 1.5 W for Collatz, 3.5 W for Einstein anf 4.5 W for SETI. I know Einstein loves memory bandwidth, hence my classification of SETI as "massive". I can easily imagine this to impact other processes, especially ones that themselves long for bandwidth. However, I can't quantify the impact because my CPU cores are crunching WCG and the runtimes & credits vary significantly over there.
I'm running 2 x DDR4 3200 at 1.36 V, so my power consumption numbers (read out with HWiNFO64) are probably higher than what most others will be seeing. My point is the relative difference between those numbers, though, which should apply to anyone.
I originally started a thread
)
I originally started a thread over in bug reports; but with my skylake laptop and the 2015/11/21 (10.18.15.4281) driver I'm having having WUs fail any time the screen is blanked. What appears to be happening is that the WU being ran when the laptop blanks its screen (it was set to blank screen but never go to sleep on wall power) ends up failing validation and new BRP4 (Arecibo) tasks run until they hit a 9 hour timeout and fail.
I've changed my power profile to keep the screen on overnight; to see if that produces a batch of good results (and proves my GPU is capable of working right).
If there's still not a good fix for this, is the problem something that will affect many/most projects; or could I put my laptops IGP safely to work running something else?
RE: I originally started a
)
Ended up having a few GPU tasks go on runaway even with the screen on; so my initial theory is wrong. I'm trying soem SETI tasks now since they were the only other Intel GPU project I could find (Collatz looked like it supported it, but the server reported no tasks available unless I switched CPU on for them).
New driver 4352 for HD
)
New driver 4352 for HD Graphics 530 dated the 23rd of December has appeared on downloadcenter.intel.com. The summary of changes makes no mention of OpenCL.
Currently running Seti on it and will then run some Einstein tasks. Link: Host 6274635
BOINC blog
RE: New driver 4352 for HD
)
Not looking too good. Currently got 16 inconclusive, 3 valid and a bunch more pending.
BOINC blog
RE: RE: New driver 4352
)
Failing on both projects or just here?
On my skylake laptop the december driver is behaving a bit better in general use on W10. It went from occasionally timing out while rendering FF - forcing the OS to restart it and requiring a window resize to get it repainted normally (instead of all gray) - to occasionally taking a broad curved divot out of the top of the tab bar until a repaint was triggered.
I'm away from home, so haven't tried crunching anything with it yet.
RE: RE: Not looking too
)
Seti seems to be about 50/50. still got one pending over there.
I'd say it doesn't work.
I'm not sure how to log a bug with Intel about their drivers seeing as I'd have to describe the problem, just saying they don't validate against other work units doesn't help. I'd need some help from the project (result files maybe) to be able to log a bug.
BOINC blog
RE: RE: RE: Not looking
)
I started a batch of seti ones. SO far 10/39 validated successfully after getting the first result back, the other 29 were inconclusive. 9 of those have had a 3rd result back; 8 resulted in all 3 people getting credit and 1 invalidated my result.
That makes me suspect the issue on seti is cumulative rounding differences pushing the validation thresholds to their limits. However unless needing a 3rd result to validate is common, my GPU's only effectively contributing at 25% of its theoretical power. Which is probably not worth stressing the cooling and limiting CPU turbo levels by a few hundred MHz. I'll do a batch of E@H GPU tasks next; but I think the GPUs going to have to be shelved until another driver update is out.
Edit: Just saw a notice that seti's starting to roll out a new app. Guess I'll try that too before throwing in the towel.
Quick note: on my HD530 + Win
)
Quick note: on my HD530 + Win 10 and driver x.4300 SETI is working fine. I had invalid WUs every now and then, but with an acceptable quota of <10% with my i7 6700. Since it broke (too much cooler pressure) I switched to an i3 6100 with an HD530, too, and it didn't produce a single invalid WU yet.
Regarding Einstein on Skylake GPUs: thanks for testing the beta drivers, Mark! We had been talking about that over there, but had no solution for driver 4300 and earlier. I'm briefly describing though, how I ran an offline test according to someone else's instructions. It did show massive differences between the reference result and what the Skylake produced.
I also filed a bug report at Intel (not easy) but didn't hear back yet, apart from the fact that they had escalated the issue (i.e. found no quick fix).
MrS
Scanning for our furry friends since Jan 2002
RE: Quick note: on my HD530
)
Did you have any performance issues while running SETI? While running that on the GPU the FGRP tasks on my laptops CPU were taking ~2x as long. ~15% of that could be attributed to the CPU clock running at 2300mhz instead of 2700 (presumably SETI was loading the GPU harder and eating more of the TDP) but the FRGP tasks spiked from 35-40k seconds to 80k.
I'm seeing "massive" main
)
I'm seeing "massive" main memory bandwidth consumption by SETI: 1.5 W for Collatz, 3.5 W for Einstein anf 4.5 W for SETI. I know Einstein loves memory bandwidth, hence my classification of SETI as "massive". I can easily imagine this to impact other processes, especially ones that themselves long for bandwidth. However, I can't quantify the impact because my CPU cores are crunching WCG and the runtimes & credits vary significantly over there.
I'm running 2 x DDR4 3200 at 1.36 V, so my power consumption numbers (read out with HWiNFO64) are probably higher than what most others will be seeing. My point is the relative difference between those numbers, though, which should apply to anyone.
MrS
Scanning for our furry friends since Jan 2002