Funny. It does NOT support CUDA apps based on the anonymous platform, so if you are running any of Raistmer's GPU modifications, do not update to 6.6.12

---

Need some clarification from Raistmer. I am not brave enough to try as don't want to trash my cache. My understanding of the team work mod is all cuda handling is done in app and boinc is not even aware cuda is being used...So, I think the teamwork mod should work? So far I have 6.6.11 on my cuda hosts and been running flawlessly for what a couple days now. Also scheduling seems better as I have all full caches now and it asks for work when needed.

---

"The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov

The below is in response to Geek@Play on the BOINC Alpha email list:
David Anderson wrote:

Your app_info.xml file has an entry for a CUDA app.
This confuses BOINC, which expects that all entries in app_info.xml are for CPU apps.

I'll change the client to detect this situation and ignore the CUDA entry in app_info.xml (eventually we'll support anonymous CUDA apps, but not in this release).

I asked for what he meant with the last line and got this answer back:
David Anderson wrote:

Supporting anonymous CUDA apps will require medium development on both client and server. I'll look at it, but probably not in time for 6.6.

---

Funny. It does NOT support CUDA apps based on the anonymous platform, so if you are running any of Raistmer's GPU modifications, do not update to 6.6.12

Hmm... Installed 6.6.12 at March, 4 (morning). Didn't see a difference in working of Raistmer's team mod V10a. Probably due to old setting <no_gpus>1?

I have just replied to David Anderson, and I don't care who knows it.

Well, excuse me.

First, could somebody remind me what the whole development plan for BOINC v6.6 was all about? Why did we need one, and in such a hurry, too? Something to do with a nVidia/CUDA marketing drive, wasn't it?

CUDA works well - well, well-ish - under BOINC v6.4.5 with an app_info: so you say you've TAKEN OUT anonymous platform support in v6.6? Why, exactly? Or did you just forget to maintain full project compatibility?

Jorden has already mentioned that SETI relied on nVidia corporate support to develop the SETI CUDA application to its present half-baked state. Having discovered, I'm told, that programming proper science on a GPU is too hard for them, and having got their pound of marketing flesh in time for the Christmas market, they've scarpered. Leaning Eric Korpela to fix the bugs they left behind on his own, or to enlist the help of skilled, experienced, willing volunteers to develop a better application. Those volunteers are the human beings you've just dumped on.

Here's another reason why app_info matters. SETI, in case you hadn't noticed, has a second application, called Astropulse. The present Astropulse code is desparately, desparately inefficient. Even the current, half-hearted (pending finalisation of the radar-rejection algorithm) optimised application increases the efficiency by a factor of some 4.5 times, or if you prefer reduces runtimes to 22% of stock. Yes, TO 22%, not BY 22%. And you can't run optimised AP alongside even stock CUDA without an app_info.xml file. So you've dumped on Josh and Astropulse as well.

I suggest that, given the choice of an early v6.6 to suit some artificial timetable, but without essential CUDA support in app_info, or a later v6.6 that is fully functional, function trumps timetable every time.

Delay the release, but make it complete.

They are not all that interested in adding support for ATI GPUs either even though we already have one project that has a "functional" application that runs on the ATI cards.

Were they generalizing the CPU/GPU scheduler in BOINC Manager adding detection code (I provided some, and someone else indicated that so did they) and scheduling would not be all that hard ... and what better time than now?

And before you tell me how busy the developers are, I will remind you that that is a hole of their own making ... if you really pay close attention, most help is refused ... if you don't believe me, try offering your own time and see how far you get ...

But.. BOINC V6.4.x is stable? Hmm..?

For example if I click from Messages to Tasks (or something other) BOINC V6.4.5 and V6.4.7 looks like frozen for ~ 5 - 10 sec.
Sometimes the BOINC Manager window is looking like: Hey.. sorry.. I'll crash soon..

In BOINC V6.6.5 and V6.6.11 the views changing immediately.
But need sometimes few sec. to update.

---

Been running on my cuda hosts ahwile now. Working great; reporting work and getting new tasks as needed.

I am using Raistmer's V10 and see 3 cuda tasks running (1/gpu) and 4 cpu tasks (1/core) q6600+c1060,260,9500

---

"The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov

Been running on my cuda hosts ahwile now. Working great; reporting work and getting new tasks as needed.

I am using Raistmer's V10 and see 3 cuda tasks running (1/gpu) and 4 cpu tasks (1/core) q6600+c1060,260,9500

I tried installing Raistmer's team mod v10a under 6.6.12 and it just won't run any cuda. Its quite happy to run AP on the cpu. Had to trash all the wu and go back to ak_v8/AP and gpugrid.

I copied all the files into projects\setiathome while BOINC was down. started it up, it found the apps and gpu. It just refused to do any cuda using the team mod. I managed to get the seti cuda_608 to run on it, so it looks like something with the team mod and 6.6.12. I had gpugrid set to NNW and there wasn't anything using the gpu at the time. Did this on 2 machines.

---

BOINC blog

Maybe someone could send my experiences to the BOINC devs?

I used BOINC V6.6.5 and V6.6.11 and had probs.

I use Raistmer's V7 mod.* [With self mod of the app_info.xml without AP]
Only GPU crunching. Nothing on CPU.

[* With the new V9 multi GPU mod the whole GPU support is fixed at Core #0 of the CPU.
So evertime if a new CUDA task start, the other GPU idle and don't crunch because of the ~ 25 sec. 100 % Core support of the other GPU. So for my usage not well.]
[EDIT: V10 it's the same, or?]

2* GTX260 Core216 feat. AMD Quad.

This mean 2 idle CPU Core.

With cc_config and <.._debug> entries I saw:

BOINC detected this and thought this two idle Core need the whole cache with WUs.
So ~ every minute BOINC asked for new work for the CPU.

I changed the 'computing preferences' to 4, 2 and finally 0 CPUs.
But BOINC thought all the time I have 2 idle CPUs..
And don't need new work for the GPUs. [But didn't reached the max. cache settings]
But I don't want to crunch current on the CPU.. ;-)

So it's a BUG?

Now with BOINC V6.4.7 it's well.. and after ~ one month I reached nearly (filled up) the cache settings.

---

I saw this little gem in the boinc_alpha mailing list this afternoon, which might explain why I can't get v10a of the team mod to work. Richard has already replied to this message on the mailing list. As mentioned above.

Your app_info.xml file has an entry for a CUDA app. This confuses BOINC, which expects that all entries in app_info.xml are for CPU apps.

I'll change the client to detect this situation and ignore the CUDA entry in app_info.xml (eventually we'll support anonymous CUDA apps, but not in this release).

In the meantime, remove the entry (i.e. the following lines) and let me know what happens.

-- David

<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>MB_6.08_mod_CUDA_V9.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft.dll</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3-1-1a_upx.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<plan_class>cuda</plan_class>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.040000</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V9.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
</app_version>

---

BOINC blog

Fires extinguished, all back to normal.

David Anderson wrote:

It turns out that the changes needed to support coprocessor (e.g., CUDA) apps in the anonymous platform mechanism are entirely server-side.

With these changes, you can safely specify GPU apps in app_info.xml;
the scheduler will take this into account, so if your client is asking for CPU work, the scheduler won't send you jobs that will run on the GPU (this caused the infinite-work-fetch scenario).

In fact, your app_info.xml can specify both GPU and CPU app versions for the same app, and everything will work as designed.

I'm currently testing these scheduler changes, and hopefully will have them deployed on SETI@home later today.

-- David

PS: no progress on ATI GPU detection; I'm waiting for ATI to send me a board.

---

Now If I can solve the other problem, the one that makes the sonic booms and shakes My house...

Surface to Air Missile sale?

---

Its been superceeded by 6.6.14 now.

---

BOINC blog

For example if I click from Messages to Tasks (or something other) BOINC V6.4.5 and V6.4.7 looks like frozen for ~ 5 - 10 sec.
Sometimes the BOINC Manager window is looking like: Hey.. sorry.. I'll crash soon..

I get the same with 6.4.6

---

me@rescam.org