Many errors w/ multiple tasks on Intel GPU

Dougga
Dougga
Joined: 27 Nov 06
Posts: 27
Credit: 24,844,941
RAC: 0
Topic 215833

I have a new Lenovo Yoga with the Intel UHD 620 GPU and for what it's worth the Intel Core i7-8550U CPU.
I've configured the GPU to run 3 tasks and it runs at about 80% utilization.
I'm finding that about half of my work units get tagged as invalid and am not sure what to try next.

Has anyone seen this behavior?

Thanks,
Doug

Zalster
Zalster
Joined: 26 Nov 13
Posts: 3,117
Credit: 4,050,672,230
RAC: 0

Try reducing your GPU

Try reducing your GPU utilization to 1 and see if they errors go away.  It's hard not to jump to multiples when you see the GPU not being utilized all the time but you have to make sure first that your system can process a single task continuously first before you try doing more than 1.   If the errors continue then it's a issue somewhere with your system. If it goes away, then you know it has to do with running more than 1 at a time. Then it's a bottleneck somewhere.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,202,024,931
RAC: 893,969

Good luck getting something

Good luck getting something useful in running Einstein on an Intel on-chip GPU.  I have at least three hosts which should be capable of that, and after multiple trials and errors many months in the past have not tried in a long while.

Some of the issues:

1. some versions of the Intel-supplied drivers gave numeric results not close enough to those from other platforms to get consistently validated.

2. some versions of the Intel drivers would not complete the tasks at all.

3. the normal system resource accounting seemed broken.  An Einstein task running on the onboard Intel GPU seemed to get in the way of other work far more than the metrics of GPU, memory, and CPU utilization I could see would have suggested.

4. aside from bad accounting, the system impact seemed pretty bad--especially considering the modest useful work output compared to any of my stable of Nvidia cards.

For my personal case I decided that the tradeoff of useful output per unit power consumption, system ease-of-use loss, and general configuration frustration seemed unfavorable, and I've not tried in some time.

None of which is meant to say I know it to be a bad idea for anyone else, just that there do seem to be some difficulties.

As very many modern Intel CPU chips come with an on-board GPU these days, I for a time expected that they would come to play a very large part at Einstein.  But as of today, the weekly credit (a good measure of actual useful work accomplished) for all Intel GPUs is less than a hundredth that for either Nvidia or AMD here at Einstein.

On a more optimistic note, the Einstein server status page asserts that there are currently 1023 validating hosts using Intel GPU, so it seems quite a few people have enjoyed success.

 

 

 

Dougga
Dougga
Joined: 27 Nov 06
Posts: 27
Credit: 24,844,941
RAC: 0

Interesting and helpful

Interesting and helpful comments.  Thanks.

I'll start tinkering with various parameters to see if I can get something working.

If the reporting from the GPU isn't reliable, that will impact my strategy.  I'll focus on reported BOINC productivity. I may start with choosing even number of concurrent GPU tasks and turning off the BOINC graphic screensaver.

 

Thanks again.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 0

archae86 wrote:Good luck

archae86 wrote:

Good luck getting something useful in running Einstein on an Intel on-chip GPU.  I have at least three hosts which should be capable of that, and after multiple trials and errors many months in the past have not tried in a long while.

Some of the issues:

1. some versions of the Intel-supplied drivers gave numeric results not close enough to those from other platforms to get consistently validated.

And for some models all driver versions have this problem.  I've got a laptop with a 6xxx series Intel CPU that I'm running SETI on because they're the only project I'm aware of to've written a special app version that uses alternate (slower, but less cumulative error generating) calculations for the IGP that are able to validate against other devices.

archae86 wrote:
On a more optimistic note, the Einstein server status page asserts that there are currently 1023 validating hosts using Intel GPU, so it seems quite a few people have enjoyed success.

I wonder how many of them are well functioning though.  I could run E@H work on my Intel IGP, but roughly 1 in 3 tasks would fail; I stopped both because the failure rate was high enough to seriously dampen the credit it generated and because I didn't want to screw any science when inevitably I got paired with another Intel GPU with the same sort of buggy driver.

 

earthbilly
earthbilly
Joined: 4 Apr 18
Posts: 59
Credit: 1,140,229,967
RAC: 0

To many background apps and

To many background apps and .exe programs running can contribute to "computation errors". They cause memory errors that can cascade into failed BOINC tasks.

I fought through this until I reloaded system mechanic, letting me turn off startup files and fix registry errors. Now all seems well.

If you are running Windows 10, open Windows administrative tools and open resource monitor to watch for how many errors your memory has and what .exe programs cause them and check if any .exe programs have multiple apps running at the same time. Before I cleaned up my registry I noticed 5 firefox identical .exe programs running at the same time, among other duplicates. No wonder BOINC tasks crash!

100% powered by solar,

 

Work runs fine on Bosons reacted into Fermions,

Sunny regards,

earthbilly

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 574,046,932
RAC: 188,762

Your UHD620 is basically just

Your UHD620 is basically just a Skylake GPU. If haven't followed things during the last 2 years, but before that we never got Skylake to reliably return valid WUs. Considering the history of Intel drivers I highly suspect it's still broken.

MrS

Scanning for our furry friends since Jan 2002

Betreger
Betreger
Joined: 25 Feb 05
Posts: 992
Credit: 1,574,081,574
RAC: 745,368

IMO IGPUs are evil.

IMO IGPUs are evil.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,202,024,931
RAC: 893,969

Betreger wrote:IMO IGPUs are

Betreger wrote:
IMO IGPUs are evil.

They put pixels on my laptop screen just fine--just don't try to use them for Einstein.  At least that's my position until I learn better.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.