Thus far, they are BOINC 7, only. I'm deciding whether it's safe to continue releasing BOINC 6 revisions of ATI apps. X64 Linux versions should be out shortly.

There are now two versions. One for HD5000+ cards only and that works on all cards including pre-HD5000 cards. These versions should get delivered to Catalyst version 12.10 and earlier, and Catalyst 13.2 and later. If you've got one of those driver revisions and aren't getting work, let me know.

---

Why we still need these ones?

Windows/x86 7.02 (ati_opencl_pre129) 8 Apr 2013, 16:34:53 UTC
Windows/x86 7.02 (ati_opencl_sah) 8 Apr 2013, 16:34:53 UTC

Thanks for reminding me. I'll deprecate them.

---

Few observations:

1) I detached BOINC7 host and attached it again - recived many MB7 tasks, not just 1 - should it be so?

2) Did the same with BOINC6 host... and recived 1 task for ati_opencl_pre129 plan class. I did this few mins ago, MB 7.02 ATi app not listed on web page already. How it can be then? BOINC sends deprecated apps ? It could be good in my particular case cause w/o it BOINC6 could not participate at all currently but it's bug still.

Now that MB 7.03 is here what is the lowest card for ATI.

After updating to cat 13.4 I've managed to reduce downloads from ati5 + ati to just ati.


Michael

---

A computer program will always do what you tell it to do, but rarely what you want it to do.

The ATI version should work with Abstraction Layer 7 hardware (which I believe starts in the HD5000 series). The ATI5 version should go down to Abstraction Layer 5, which I believe is everything that supports OpenCL.

---

1) I detached BOINC7 host and attached it again - recived many MB7 tasks, not just 1 - should it be so?

Yes, if the work estimate indicates you can do many tasks before deadline or full queue.

2) Did the same with BOINC6 host... and recived 1 task for ati_opencl_pre129 plan class. I did this few mins ago, MB 7.02 ATi app not listed on web page already. How it can be then? BOINC sends deprecated apps ? It could be good in my particular case cause w/o it BOINC6 could not participate at all currently but it's bug still.

Apparently the server will send deprecated versions if the only tasks your GPU or CPU can use are for deprecated versions. I'll have to see if there is a way to turn off this "feature."

---

Thanks Eric.

I'll give it another try I'm still using AMD Radeon HD4600.

Michael

The ATI version should work with Abstraction Layer 7 hardware (which I believe starts in the HD5000 series). The ATI5 version should go down to Abstraction Layer 5, which I believe is everything that supports OpenCL.

Hm, not sure what you mean by AL5 or AL7, but HD5 will work with HD5xxx and up, non-HD5 with HD4xxx and up. Both are OpenCL 1.0 compatible, but HD5 make use of local memory that emulated via global one on HD4xxx cards hance such difference.

Thanks Eric.

I'll give it another try I'm still using AMD Radeon HD4600.

Michael

Did your GPU work with prev OpenCL MB releases? If not than don't waste time to try again, nothing changed in low-level HD4xxx GPU support.

Abstraction layer level was how ATI was going to tell you what functions were supported on a GPU. Like everything else ATI ever did, it's now deprecated. It's even been scrubbed from the wikipedia pages for ATI/AMD GPUs. Hopefully the API will continue to return 7 or higher for the level for future cards.

The reason it was necessary is that in the past there was no clear relationship between model number and capabilities.

---

I'll look at the logs to see whats up if you can't get work.

Thanks Eric.

I'll give it another try I'm still using AMD Radeon HD4600.

Michael

---

Hopefully the API will continue to return 7 or higher for the level for future cards.

In what API that call exists?

calGetDeviceAttribs() It's the target field of the attributes. I guess I should be calling it target rather than level. Last I checked they are up to 23, but I think your HD5 version runs on everything above 7.

---

Linux MBv7_7.03 GPU app does not work, because of wrong .cl filename.

It is expecting "MultiBeam_Kernels_r1831.cl" in the "projects/setiweb.ssl.berkeley.edu_beta/" dir and NOT "MultiBeam_Kernels_r1831_linux_ati.cl".

Examples for error outcome are on this host.

---

_\|/_
U r s

Linux MBv7_7.03 GPU app does not work, because of wrong .cl filename.

It is expecting "MultiBeam_Kernels_r1831.cl" in the "projects/setiweb.ssl.berkeley.edu_beta/" dir and NOT "MultiBeam_Kernels_r1831_linux_ati.cl".

Examples for error outcome are on this host.

The windows apps had changes so it would follow the <open_name> as command, did that not make it to the Linux apps?:

    <file_ref>
        <file_name>MultiBeam_Kernels_r1831_windows_ati.cl</file_name>
        <open_name>MultiBeam_Kernels_r1831.cl</open_name>
    </file_ref>

http://lunatics.kwsn.net/gpu-crunching/mb-ati-stock-r1807.0.html

Claggy

Linux MBv7_7.03 GPU app does not work, because of wrong .cl filename.

It is expecting "MultiBeam_Kernels_r1831.cl" in the "projects/setiweb.ssl.berkeley.edu_beta/" dir and NOT "MultiBeam_Kernels_r1831_linux_ati.cl".

Examples for error outcome are on this host.

The windows apps had changes so it would follow the <open_name> as command, did that not make it to the Linux apps?:

    <file_ref>
        <file_name>MultiBeam_Kernels_r1831_windows_ati.cl</file_name>
        <open_name>MultiBeam_Kernels_r1831.cl</open_name>
    </file_ref>

Claggy

Guess not. Otherwise it would work. There is no MS file handling on Linux, so has to be implemented separately. Where and when was that communicated ? Must have missed that!

---

_\|/_
U r s

Oops. We'll have to get a new one out ASAP.

---

Oops. We'll have to get a new one out ASAP.

You'll have it today (your time).

---

_\|/_
U r s

They've been updated. Fingers crossed....

---