Yes Collatz is set up differently and identifies computers that have a recognised ATI card and so requests and receives tasks as GPU tasks, even with an ATI app_info.xml. I have no knowledge of how this is set up, perhaps it uses the BOINC plan class somehow.

Using the app_info.xml ATI version with the current MilkyWay setup you now get the same BOINC debt problems that used to happen back when there was no ATI aware BOINC versions. Overcoming it means various combinations of fiddling with resource share, cache size, setting No new tasks on CPU projects, using an update script, using methods of clearing BOINC debt levels such as <zero_debts>1</zero_debts> in a cc_config.xml file, detaching and reattaching to projects, etc.

It would be much easier for me to just use the default version that is automatically downloaded, but it is very hot here and some days I need to adjust the w parameter to prevent my GPU overheating. The 5870 is fine but the 4890 runs super hot on default parameters even on cooler days. If this stays the way it is I will go back to using an earlier non ATI aware BOINC version which will enable me to use one GPU on MikyWay and the other one on Collatz.

.....Milkyway is still only giving out work when being asked for CPU - which only happens once in a while.....

If you untick "Use ATI GPU if present" in MilkyWay preferences, BOINC will stop asking for GPU tasks and just ask for CPU tasks.

I use the command line setting to keep the system more responsive and to be able to view videos without the need to stop Milkyway. So I don't want to use the automatic download either. However I don't really want to go back to all that hassle before 6.10 was released. I'm still hoping Travis will be able to change the current behavior.
Right now I've to go through these 85 Collatz tasks I downloaded to test their scheduler response to the ATI work request. So I'll wait and see what happens.

---

Starfire

Website is really slow today. No page errors yet.

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

If you are using the ATI app_info.xml version the upgraded server now treats the anonymous platform as a CPU. You need to tick "Use CPU" in MilkyWay preferences. Otherwise BOINC will continue to ask for GPU tasks only and not receive any.

Thanks for pointing this out.
However with several other CPU projects running BOIC now will only ask for more Milkyway-CPU-work after it has run dry - and probably only as long as I continuously raise it's resource share because of increasing STD and LTD.

Limiting anonymous platform to CPU doesn't make sense to me. Is that also done that way at Collatz? I'm using anonymous platform there too and never had to play with my resource share to get work.

Edit: I just tried Collatz again. Using anonymous platform BOINC requested 259200 secs of ATI work and got 85 tasks. Milkyway is still only giving out work when being asked for CPU - which only happens once in a while. Both projects appear to use server version 611 according to the scheduler reply.

I use an app_info.xml (anonymous platform) on all 4 of my ATI machines and have no problem getting WUs. "Use CPU" is set to no. My preferences:

Use CPU (enforced by 6.10+ clients) no
Use NVIDIA GPU if present (enforced by 6.10+ clients) yes
Use ATI GPU if present (enforced by 6.10+ clients) yes

I'm using BOINC v6.10.25 on all of them. That could quite possibly be a difference.

Nope, I just set mine to the same thing after installing .25 and all I get is Message from server: No work sent.

Something is still screwed up with the server, I even added platform and planclass tags to my app_info to see if those would do anything.

---

Nope, I just set mine to the same thing after installing .25 and all I get is Message from server: No work sent.

Something is still screwed up with the server, I even added platform and planclass tags to my app_info to see if those would do anything.

My cc_config.xml also contains the following:

<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
<use_all_gpus>1</use_all_gpus>
<zero_debts>1</zero_debts>
<ncpus>-1</ncpus>
</options>
</cc_config>

Maybe the use_all_gpus is having an effect. It's odd that it works on all my boxes and not on others.

What makes it work for you are propably these two options

<zero_debts>1</zero_debts>
<report_results_immediately>1</report_results_immediately>

Boinc keeps track of how much CPU time projects have used in comparison with their resource share by maintaining debt values for each project. By these values it determines which CPU projects will run next and which project will download CPU workunits next.
Since the server change Milkyway only gives out work to users using app_info.xml when the client requests CPU work. Direct requests for GPU(ATI) work are always answered with "No work sent". Before the change ATI work requests where answered with new WUs if the cache was not full.
If you are using multiple projects you run into the problem that the BOINC client may only request CPU WUs for Milkyway in small doses or none at all if its CPU debts are too high even though your ATI card is idle.
<zero_debts> will clear these debts across all projects every time the client is restarted. So after every restart each projects has the same priority in getting CPU work which makes it more likely for BOINC to ask for CPU WUs from Milkyway.

This seems to be a configuration problem on the server since Collatz is using the same server version (611) and doesn't suffer from this. Lets hope it will get fixed soon.

---

Starfire