Thank you Georgi Vidinski!!! The AMD WUs are running like clockwork on my rigs. I jumped to 41k/day lol. Im looking forward to getting some work done for this project. Thank you again!

On my RX6800 the first WU with the new OpenCL app seemed to work, and finished OK.

But it didn't validate againtst a CUDA card.

There are many more WU's in que, I'll keep an eye on them to see if they validate.

On my R9 280X (7 of them) and my R9 Nano (1 of them), they're all validating, against Cuda and CPU. Seems to work differently on different cards, the previous version crashed in 2 seconds on my 280X cards.

On the previous version, the one WU that I ran, ran for 22 hours! or something like that (78.777 seconds), with 0.01% progress, where I aborted :)

So this is better now, with the new app.

And I'm sure they'll fix whatever error there is.

On the previous version, the one WU that I ran, ran for 22 hours! or something like that (78.777 seconds)

78.777 seconds is just over a minute.

since the app is more serialized, latest gen cards with high clock speeds seem to excel. that being the AMD 7000 series and nvidia 4000 series with clocks exceeding 2.5GHz. with more of an edge to the nvidia/CUDA cards still.

CPU remains overall more productive though (especially the high core count models), maybe this can change in the future. but nice to have more available devices to crunch if they wish to.

---

That is 78 thousand and 777 seconds.

I don't know if you can see it, but this is the output.

https://asteroidsathome.net/boinc/result.php?resultid=400155934

But it seems other computers had problems with that WU too, so it might be a fluke.

https://asteroidsathome.net/boinc/workunit.php?wuid=175662331

And as a followup, the few Asteroids OpenCL that my GPU has completed until now, have all failed to validate.

since the app is more serialized, latest gen cards with high clock speeds seem to excel. that being the AMD 7000 series and nvidia 4000 series with clocks exceeding 2.5GHz. with more of an edge to the nvidia/CUDA cards still.

CPU remains overall more productive though (especially the high core count models), maybe this can change in the future. but nice to have more available devices to crunch if they wish to.

Georgi said they would be much faster when he increases the grid size. If he can make it 3.5 times faster, I can do the same amount of work on my GPUs as my CPUs. At the moment I'd just use CPUs and not add another 1/3.5th to the mix when they could be doing something else faster.

That is 78 thousand and 777 seconds.

Oh, you write in English words but European maths :-P We call it a decimal point not a decimal comma. Do European compilers use commas? Big errors could happen!

I don't know if you can see it, but this is the output.

Yes, they're public.

https://asteroidsathome.net/boinc/result.php?resultid=400155934

That's an aborted one. Can you link to one which failed to validate?

And as a followup, the few Asteroids OpenCL my GPU has completed until now, have all failed to validate.

I think all mine are working, but I do a lot of CPU tasks too, and I can't work out how to list only the GPU ones in my account on here.

I can link the 4 that have failed until now.

https://asteroidsathome.net/boinc/result.php?resultid=401416451

https://asteroidsathome.net/boinc/result.php?resultid=401416309

https://asteroidsathome.net/boinc/result.php?resultid=401416396

https://asteroidsathome.net/boinc/result.php?resultid=401416402

I have a few more with validation pending, but I'm not hopefull they'll validate.

I feel I have to say that my GPU participates in many other projects, and they all validate fine.
So I think the GPU is OK.

I can link the 4 that have failed until now.

https://asteroidsathome.net/boinc/result.php?resultid=401416451

https://asteroidsathome.net/boinc/result.php?resultid=401416309

https://asteroidsathome.net/boinc/result.php?resultid=401416396

https://asteroidsathome.net/boinc/result.php?resultid=401416402

I have a few more with validation pending, but I'm not hopefull they'll validate.

I feel I have to say that my GPU participates in many other projects, and they all validate fine.
So I think the GPU is OK.

Those outputs look like my successfull ones, except your card seems to only have 60 compute units instead of 64, strange for a better card, but the app has reduced the grid size accordingly. I guess it made a mistake with your compute unit count.

Also strange, the first link shows yours was a validate error, before the workunit has been finished, nobody else has checked if you or the cuda card was wrong. I guess there's a primary check before comparison, so it's not just slightly wrong but very wrong. I guess you have to wait for the next version of the program, perhaps leaving them running will give them data to work out what is wrong.

I will leave it to go through the 60 WU's it has downloaded.

And as a positive note, I now have two GPU WU's that have validated (and one more invalid).

Valid:

https://asteroidsathome.net/boinc/result.php?resultid=401416242

https://asteroidsathome.net/boinc/result.php?resultid=401416429

Invalid:

https://asteroidsathome.net/boinc/result.php?resultid=401416391

But not all is bad.

And as a positive note, I now have two GPU WU's that have validated (and one more invalid).

Odd some work and some don't. I can understand human inconsistency, but not computers!

There must be some small errors in the calculations.

I hope they are able to fix it.

Example of task failing due to Pop!_OS having disabled rusticl https://asteroidsathome.net/boinc/result.php?resultid=401473651

thanks for the AMD app, nice job!

it takes about 6 minutes/task on my 7900 xtx https://asteroidsathome.net/boinc/result.php?resultid=401460520

and here's a profile


enqueueing small kernels waiting on each other seems to be a significant bottleneck. Perhaps they could be inlined as functions?

i think your opencl version is too old, 1.1 is ancient. Mesa is generally not well suited for real compute loads. Many people have issues using Mesa with BOINC

try installing the real AMD drivers. you might have to back up to an older OS/kernel though. or try the ROCm drivers, but i think ROCm 5.6? was the last to support Vega

Works now with an extracten OpenCL stack from an older AMD-Driver (20.40). :)