I just tested a few weak Intel GPUs (mostly an Intel HD Graphics 4000), and they did not work but seemed to be working. I'd like to summarize this situation (described in parts throughout this thread) for anyone else who encounters the same.

For weak Intel GPUs, the task seems to be making progress because...
- its Progress percent keeps going up, though the increase slows over time (mine reached 98% after 23 hours after an initial 6-hour-total pace)
- it fully runs a CPU core
- it uses a lot of RAM (around a GB for me depending on device) even though my AMD GPU uses only a little RAM, though I would predict that Intel GPUs that do not stall would not use as much RAM

However, there are many ways to know that a NumberFields task never really started...
- checkpoints are not being done (Progress resets after task is suspended)
- in its BOINC/slots/ folder, boinc_task_state.xml is never created
- in its BOINC/slots/ folder, the stderr.txt file only has a couple lines (for me, I only saw the usual "GPU not found in table" 6 lines)

Yes, we only use integer ops here.

The good news is the results are correct.

Since OpenCL makes arithmetic errors on AMD and Intel GPUs (for 64-bit integers at least), how do you know if the results are correct? I am curious since this project has single-task validation.

Yes, we only use integer ops here.

The good news is the results are correct.

Since OpenCL makes arithmetic errors on AMD and Intel GPUs (for 64-bit integers at least), how do you know if the results are correct? I am curious since this project has single-task validation.

What do you mean by "OpenCL makes arithmetic errors... for 64-bit integers".

That should never happen. I could possibly see 64-bit floats having some round off errors, but not integers.

Anyways, during development, I ran hundreds of GPU cases and compared against the CPU versions and they all agreed after computing billions of polynomial discriminants. So I am pretty confident when I say the OpenCL code produces correct results.

You have more months of experience to consider now. Are you leaving the App as BETA because of the issues that the weaker IGPU systems may have with inexperienced users or other reason's ?

Thanks
Bill F

You have more months of experience to consider now. Are you leaving the App as BETA because of the issues that the weaker IGPU systems may have with inexperienced users or other reason's ?

Thanks
Bill F

Yes, that is the primary reason. And it's easy to enable beta apps for those that want to try it out.

I have added two systems to the project in the last 24 hours that have Intel GPU's. After reviewing this tread and seeing that some Intel GPU's have been identified as successful and some not. Can the successful models be added one or more at a time, to the look up table, for optimization ?

From my first successful task

GPU Summary String = [INTEL|Intel(R)UHDGraphics620|1|3227MB||300].
Loading GPU lookup table from file.
GPU was not found in the lookup table. Using default values:
numBlocks = 1024.
threadsPerBlock = 32.
polyBufferSize = 32768.


System information
GenuineIntel
Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz [Family 6 Model 142 Stepping 11]
(8 processors) INTEL Intel(R) UHD Graphics 620 (3227MB) OpenCL: 3.0

Thank you
Bill F

I have added two systems to the project in the last 24 hours that have Intel GPU's. After reviewing this tread and seeing that some Intel GPU's have been identified as successful and some not. Can the successful models be added one or more at a time, to the look up table, for optimization ?

From my first successful task

GPU Summary String = [INTEL|Intel(R)UHDGraphics620|1|3227MB||300].
Loading GPU lookup table from file.
GPU was not found in the lookup table. Using default values:
numBlocks = 1024.
threadsPerBlock = 32.
polyBufferSize = 32768.


System information
GenuineIntel
Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz [Family 6 Model 142 Stepping 11]
(8 processors) INTEL Intel(R) UHD Graphics 620 (3227MB) OpenCL: 3.0

Thank you
Bill F

Yes, a line could be added to the lookup table. But there is no reason to do that until someone has done the dirty work of determining what the optimal settings should be. The default settings were chosen conservatively so they would work well for most cards.

If you were interested in trying to get optimal values, one thing you could do is edit the gpuLookupTable.txt file in the project directory to add a line for your gpu. You would have to run enough WUs to get a good average in order to compare your new settings with the default settings. And it would definitely help to understand the underlying architecture of the gpu to give a good starting point for the settings. For example, Nvidia uses a "warp size" of 32, which is the number of threads to keep in lockstep, and it turns out the optimal value for threadsPerBlock was 32.

Can someone create a script that would run test workunit and tune worker?

Can someone create a script that would run test workunit and tune worker?

This has actually been done. It took some searching but I found where this was discussed in the past:
https://numberfields.asu.edu/NumberFields/forum_thread.php?id=488&postid=3063#3063

In a nutshell, there is a public github project with all the source code and testing scripts. For the GPU testing, see the README file in the test directory. And don't hesitate to ask if you have any questions.

Also see this for a discussion of the lookup table:
https://numberfields.asu.edu/NumberFields/forum_thread.php?id=501&postid=3176#3176

Adding an Intel ARC A750 to the pool, watch these spaces => https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=2835341 and https://numberfields.asu.edu/NumberFields/results.php?hostid=2835341

Adding an Intel ARC A750 to the pool, watch these spaces => https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=2835341 and https://numberfields.asu.edu/NumberFields/results.php?hostid=2835341

It's been several hours and nothing returned yet, I hope that's not a bad sign.

I would expect the Arc A750 to do well, assuming Intel puts out a decent openCL driver.

Adding an Intel ARC A750 to the pool, watch these spaces => https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=2835341 and https://numberfields.asu.edu/NumberFields/results.php?hostid=2835341

It's been several hours and nothing returned yet, I hope that's not a bad sign.

I would expect the Arc A750 to do well, assuming Intel puts out a decent openCL driver.

The machine is busy with Einstein@home tasks :D
Arc A750 doesn't have FP64 though

I'll pause Einstein and let some tasks run

Adding an Intel ARC A750 to the pool, watch these spaces => https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=2835341 and https://numberfields.asu.edu/NumberFields/results.php?hostid=2835341

It's been several hours and nothing returned yet, I hope that's not a bad sign.

I would expect the Arc A750 to do well, assuming Intel puts out a decent openCL driver.

The machine is busy with Einstein@home tasks :D
Arc A750 doesn't have FP64 though

I'll pause Einstein and let some tasks run

I saw some cpu tasks go through. Any problems with tasks on the Arc A750?

Adding an Intel ARC A750 to the pool, watch these spaces => https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=2835341 and https://numberfields.asu.edu/NumberFields/results.php?hostid=2835341

It's been several hours and nothing returned yet, I hope that's not a bad sign.

I would expect the Arc A750 to do well, assuming Intel puts out a decent openCL driver.

The machine is busy with Einstein@home tasks :D
Arc A750 doesn't have FP64 though

I'll pause Einstein and let some tasks run

I saw some cpu tasks go through. Any problems with tasks on the Arc A750?

I don't see any errors, I've been away for a few days and only Einstein tasks were processed even though numberfield was enabled and on the same ressource share ...

Set Einstein to no new tasks then.

bunch of WU processed :)

https://numberfields.asu.edu/NumberFields/result.php?resultid=161210884
https://numberfields.asu.edu/NumberFields/result.php?resultid=161211661

and plenty others

bunch of WU processed :)

https://numberfields.asu.edu/NumberFields/result.php?resultid=161210884
https://numberfields.asu.edu/NumberFields/result.php?resultid=161211661

and plenty others

Thx! Good to see it's working, and that run times are reasonable on Intel's newest GPUs.

bunch of WU processed :)

https://numberfields.asu.edu/NumberFields/result.php?resultid=161210884
https://numberfields.asu.edu/NumberFields/result.php?resultid=161211661

and plenty others

Thx! Good to see it's working, and that run times are reasonable on Intel's newest GPUs.

Run times are longer for the Intel GPU tasks. Are they doing larger amounts of science and should the credit amount be higher ?

Bill F

---

In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

bunch of WU processed :)

https://numberfields.asu.edu/NumberFields/result.php?resultid=161210884
https://numberfields.asu.edu/NumberFields/result.php?resultid=161211661

and plenty others

Thx! Good to see it's working, and that run times are reasonable on Intel's newest GPUs.

Run times are longer for the Intel GPU tasks. Are they doing larger amounts of science and should the credit amount be higher ?

Bill F

Run times are longer compared to what? Compared to the previous dataset, run times are longer and credits have gone up accordingly.

If comparing to a different gpu then average times will probably be different. Each WU is a fixed amount of work and therefore has a fixed credit. Just because they run longer doesn't mean they deserve more credit. If it's running longer than expected then it could be an inefficient driver or bad OpenCL implementation.

bunch of WU processed :)

https://numberfields.asu.edu/NumberFields/result.php?resultid=161210884
https://numberfields.asu.edu/NumberFields/result.php?resultid=161211661

and plenty others

Thx! Good to see it's working, and that run times are reasonable on Intel's newest GPUs.

Run times are longer for the Intel GPU tasks. Are they doing larger amounts of science and should the credit amount be higher ?

Bill F

Run times are longer compared to what? Compared to the previous dataset, run times are longer and credits have gone up accordingly.

If comparing to a different gpu then average times will probably be different. Each WU is a fixed amount of work and therefore has a fixed credit. Just because they run longer doesn't mean they deserve more credit. If it's running longer than expected then it could be an inefficient driver or bad OpenCL implementation.

I was comparing to a default Windows Get Decic Fields v4.00 (default)windows_x86_64

You may have a point about Driver or implementation

Bill F