Hi everyone,

We are proud to announce our new OpenCL applications for AMD GPUs!
Development process took some time but we finally managed to build it.
There are still many rough edges to be polished but thanks to you guys and to your feedback we'll continue to improve the code.

Linux application requires 64 bit OS with GLIBC v2.31 or higher and will run on most AMD Radeon GPU cards and some AMD Integrated Graphics with support of OpenCL 1.1 or higher.

Windows application requres 64bit Windows 10 or 11 and will run on most AMD Radeon cards and some AMD Integrated Graphics with support of OpenCL 1.2 or higher.

Happy crunching and thank you for your support!

Asteroids@home team

This is exciting or awesome, i am participating in/for boinc/asteroids since 2012 , never really managing to get parallel tasks on some personal computer, as if some "technical curse" weighed heavy on me or my system. Having spotted how your work starts to run since about september 14th is like a (demon rearranging) spellbreak, i will observe how(e) the crunching points will develop. A profond thank you, and a chapeau for my impression of some coolest progress occuring since quite a while.

What do you mean you can't get parallel tasks? Your Ryzen computer has plenty RAM and a decent graphics card. Are you unable to get more than one Asteroids running at a time on the CPU?

As far as i know there is no project where you can run more then one WU per core on CPU. If you want to run more then one WU on GPU, you can edit in config file.

This is awesome! thank you for the work!

As far as i know there is no project where you can run more then one WU per core on CPU. If you want to run more then one WU on GPU, you can edit in config file.

I assumed he meant more than one work unit per CPU, I wasn't sure what he meant by "never really managing to get parallel tasks on some personal computer".

And of course you can run more than one task per core. Not usually advantageous to do so, but I do it with Theory tasks on LHC, because if they all happen to download masses of data (they do this internally inside the Virtualbox task) at the same time, several cores end up idle. So I run say 30 on a 24 core machine.

    <app_version>
        <app_name>Theory</app_name>
        <plan_class>vbox64_theory</plan_class>
        <avg_ncpus>0.8</avg_ncpus>
    </app_version>

I do the opposite on another machine where Theory tasks, although inly using one core each, tend to hog some kind of resource in the CPU/RAM/something, which slows other stuff down, like me using the web:

    <app_version>
        <app_name>Theory</app_name>
        <plan_class>vbox64_theory</plan_class>
        <avg_ncpus>1.6</avg_ncpus>
    </app_version>

For Nbody tasks on Milkyway, I change both how many cores they're told to run on, and how many they average running on (an instruction to Boinc's scheduler as to how many to run at once). Here I've got a 24 core machine and their default is 16 cores. 16 doesn't fit into 24 too well, so I run three at once, but let them try to take 16, so while one is doing the single core setup at the start, the others can take over the spare cores:

    <app_version>
        <app_name>milkyway_nbody</app_name>
        <plan_class>mt</plan_class>
        <cmdline>--nthreads 16</cmdline>
        <avg_ncpus>8</avg_ncpus>
    </app_version>

Hi all,

Thank you for this new OpenCL version, really appreciated :)

I must be doing something wrong on my end, because it's not working yet. I have enabled AMD GPU tasks on both my account on asteroidsathome.net/boinc and my account manager (BAM).

I have synchronized my client with BAM and updated the project configuration with boinctui, but still no GPU tasks for my RX 7900 XTX. Actually all references to OpenCL are gone from sched_reply_asteroidsathome.net_boinc.xml, I don't know why...

I'm running Ubuntu 22.04 x86_64 (glibc 2.35).

Could you please give me a hand ?

Best regards,

Samuel

what does the event log say? is it asking for GPU work? in general projects only give you what the client asks for. so if the client isnt asking, it wont get.

my hunch is that it's something up with the account manager. remove it and see if you get work. if you do then you know BAM is the issue.

---

Shooting in the dark, have you tried "Reset project"?
As a final resort you may try and reinstall boinc.

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson

Hello all,

Must have been a transient issue, the latest version of the GPU application just got downloaded and it's currently running :)

I'll keep you all posted.

EDIT: some logs below about one of the first task to complete successfully on my RX 7900 XTX

https://asteroidsathome.net/boinc/result.php?resultid=401627544

Best regards,

Samuel

what's the power draw for those 6 mins?

---

there are 3 phases, but the main, most power hungry part draws
140W 1,5Ghz 10m
160W 2,0Ghz 8m
220W 2,5Ghz 7m
300W 3,0Ghz 6m

it's fast for a single unit, but for the overall throughput, it's 3 times less power efficient compared to a modern CPU. With slower cards, the ratio is even worse. Anyway, it's great we can finally test it on AMD cards after all these years :)

Why go by power? The card is gonna draw power for any project. Put it on the project you want to contribute to.

Why go by power? The card is gonna draw power for any project. Put it on the project you want to contribute to.

it depends on what your goals are. For instance, Einstein tasks are 200x faster on a GPU than on a CPU, it's the best-case scenario because their algo is very bandwidth limited (960GB/s on a GPU (7900 xtx) vs ~50GB/s on a CPU (ddr4 dc)), but not all algorithms could be parallelized to utilize a GPU efficiently, it's a different architecture with pros and cons.

also one of the reasons why Milkyway gave up https://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4748 it's not an easy task for a complex app.

Dear Sirs.
I kindly ask you to remove completely from the Asteroids@Home project a non-optimized application for 32 bit systems: https://asteroidsathome.net/boinc/apps.php
(version 102.10 28 Sep 2013, 13:14:03 UTC).
Look at the page: https://asteroidsathome.net/boinc/apps.php where an unoptimized 32 bit application produces 14 GigaFLOPS.
That is, in fact, that turns out that much users of 32 bit systems are crunching for 6 hours with an unoptimized application. In case you remove it, all users will begin to crunch at sse2/sse3-optimized applications in just 2 hours.
In other words, huge opportunities are being lost doing it!!!
There are no processors without sse2 support for a long time, since 2003 year!!! And if they are where, then no one runs BOINC on them.

The server automatically gives the best task to your computer. Clearly it is better than the sse apps for some processors, or nobody would be getting them. Later is not always better, I've got computers which do sse2 faster than sse3, and computers which do fma faster than avx.

Any plans for NVIDIA?

Any plans for NVIDIA?

Lppk at app page there is already there
Microsoft Windows running on an AMD x86_64 or Intel EM64T CPU 102.17 (cuda118_win10)

Any plans for NVIDIA?

Let Me Bing That For You :)

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson

Any plans for NVIDIA?

Let Me Bing That For You :)

Oh I was thinking of another project that dropped GPU support and thought you were that project bringing GPU support back.

BTW I get a 404 error on that link you sent me