O.K., but there in the 'Questions & answers : CUDA :' - area are only 3 or 4 people around which could have knowledge - ..or not.

Here in the NC are more people around which could have knowledge because of special topics or problems.

..therefore it would be better to allow CUDA related topics here in the NC.


..or make a new NC-CUDA under-forum..

It depends on the nature of the CUDA question. If we don't allow CUDA questions here, as you suggest, then why are Raistmer's threads here? The more technically advanced posts stay here in Number Crunching. The simple questions go into the CUDA forum. There is no need for an NC-CUDA forum.

CUDA Questions-PLEASE READ - sticky thread in the NC forum.

Hmm.. it's not a mod-team idea.. only from Blurf?

---

O.K., but there in the 'Questions & answers : CUDA :' - area are only 3 or 4 people around which could have knowledge - ..or not.

Here in the NC are more people around which could have knowledge because of special topics or problems.

..therefore it would be better to allow CUDA related topics here in the NC.


..or make a new NC-CUDA under-forum..

It depends on the nature of the CUDA question. If we don't allow CUDA questions here, as you suggest, then why are Raistmer's threads here? The more technically advanced posts stay here in Number Crunching. The simple questions go into the CUDA forum. There is no need for an NC-CUDA forum.

CUDA Questions-PLEASE READ - sticky thread in the NC forum.

Hmm.. it's not a mod-team idea.. only from Blurf?

Again, Raistmer doesn't like the thread hijacking, but to answer your question, no, it is not a Blurf-only idea, it was created the the intention of pointing people to the proper area in case they didn't know about it. I don't think Modding has to be so rigid and strict, but we try to do the best we can in keeping things running smoothly and tidy.

---

Sorry Raistmer.. last offtopic post..
But.. you answered also because of this.. ;-)


@ OzzFan

..but I saw some threads were 'jumped' from the NC forum to the Questions & answers : CUDA board..


So I'm afraid to open here in the NC forum CUDA topic threads..


I think here [NC] is the most frequently under-forum.. better than in the Questions & answers : CUDA board.


Like I said.. last post here because of this.. :-)

---

..but I saw some threads were 'jumped' from the NC forum to the Questions & answers : CUDA board..


So I'm afraid to open here in the NC forum CUDA topic threads..


I think here [NC] is the most frequently under-forum.. better than in the Questions & answers : CUDA board.

Like I said, yes, some posts get moved. I, myself, use my own judgement when moving posts, and if they are very technical in nature, such as using Raistmer's optimized apps, or any errors from using them, I keep them here. The simple questions get moved to the Q&A so that new crunchers who are intimidated by this forum and might not see the questions over here might have a chance of seeing it over there.

---

@ Raistmer

Thanks for your very nice work! :-)


One question..

Since BOINC V6.6.20, it would be possible to make a CUDA app with 'CUDA start opt.' and 'VLAR kill' - BUT WITHOUT priority or some other mod?

Because I saw..
With your mod the BManager is very slow.. sometimes frozen..
It depend also how big the cache is - I guess..
At some cache point, I had only download errors.. like 'expected this but got this size'.
But finally downloaded well - but needed several try (much more time needed).. It was for 1 or 2 days.. I guess no Berkeley-probs, or?
I think because the CUDA app have the same priority as boinc.exe and boincmgr.exe ('normal').


I use now all stock.. BOINC and apps..
I see in BOINC V6.6.20 that the MB CPU apps have 'low' priority.
The CUDA apps have 'lower as normal' priority.
boinc.exe and boincmgr.exe have 'normal' priority.
[TaskManager]


If you would make a 'small CUDA mod' with the upper said 2 functions.. your CUDA app (with app_info.xml) would have the same priority like the stock CUDA app?
And/or all BOINC relevated tasks in TaskManager would have the same stock priorities?


Yes, we all wouldn't need CUDA mods, if the stock CUDA would running well.. ;-)



BTW.
I use the current stock CUDA app..
..'bad' VLARs with stock CUDA app.. ~ 7 times slower as 'normal' ARs.. huhh.. buhh.. :-(
[~ 8 min. vs. ~ 55 min.]



BTW. #2
I used your V7 and V10 CUDA app [stand alone].
With both I had once a week one frozen CUDA app. So only one of my both GPUs were crunching.
Then I gave your V11 CUDA app a try - I changed my app_info.xml - restarted BOINC and after ~ 10 min. frozen BOINC I gave up and made a reboot and made a fresh installation of BOINC/SETI@home.


EDIT:
BTW.
With all stock, the CUDA apps are not disturbed from the CPU apps.
Simultaneously CPU and GPU no disturbing - and same CUDA performance as only CUDA.

---

My CUDA MB builds have THREAD_PRIORITY_BELOW_NORMAL worker thread priority and IDLE_PRIORITY_CLASS process priority (older BOINC rised CUDA process priority to normal that lead to user-noticeable delays in UI).
It was established in another BOINC-related thread already that big cache can effectively kill BOINC.
If you want try to get rid of CUDA freezing you could try build from thread on Lunatics site dedicated to this problem.

Like I posted upper under 'BTW. #2'.. I used your V7, V10 and V11 [freezing/restart] CUDA apps.

But after ~ 10 min. frozen BOINC I gave up.. :-(
[installed V11]

---

x64 SSSE3 & SSE3_Intel V10b updates added.
AMD users with x64 Windows should use SSE3_AMD x86 binary.
http://lunatics.kwsn.net/12-gpu-crunching/v10-of-modified-seti-mb-cuda-opt-ap-package-for-full-multi-gpucpu-use.msg15445.html#msg15445

Raistmer, I tried using one of Your packages and since I normally have done optimized MB Seti, I can't seem to get Cuda working on My GTX295, Temps are fine and the card is at stock speed, I just get compute errors when I try Cuda. Help, Where do I download a 6.6.21 compatible file that can do both cpu and gpu seti crunching? My video driver is 182.08 and the OS is XP x64 and I don't do AP and I won't do AP.

My cpu is a Q9300 running at 3.51GHz and crunches Seti perfectly and SLI is Disabled.

---

Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

Never mind It seems I got It to work, I had to download Your V10b, I'd tried the earlier V10a and got errors, V10b fixed that, Now how and where do I find V11? And It seems I have 2 cores, But when I modified the file "number_of_GPUs" from 1 to 2 I got a computation error again and yep slow downs too, Yet when I reduced "number_of_GPUs" back to 1 everything runs fine, I may need another 295 and a bigger psu to do more.

---

Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

Congrats to your new GTX295 ! :-)

With BOINC V6.6.20 you don't need longer the complete mod from Raistmer.

If needed, only the CUDA app.
V10/11 you can find here:
http://lunatics.kwsn.net/12-gpu-crunching/index.0.html

You can use this app_info.xml
http://setiathome.berkeley.edu/forum_thread.php?id=52589

---

I just get compute errors when I try Cuda. Help, Where do I download a 6.6.21 compatible file that can do both cpu and gpu seti crunching? My video driver is 182.08 and the OS is XP x64 and I don't do AP and I won't do AP.

My cpu is a Q9300 running at 3.51GHz and crunches Seti perfectly and SLI is Disabled.

With BOINC 6.6.20 you have choice to use V10 pack as before (no additional changes needed) or to use BOINC's own scheduling ability for CUDA MB tasks. This way you should replace "teamed" CPU MB apps from pack by ordinary AK_v8 counterparts (can be downloaded from Lunatics site) and replace app_info with properly edited app_info.xml (more info about what this app_info should contain and how it should look search in this http://setiathome.berkeley.edu/forum_thread.php?id=52589 thread).

There is no non-teamed AMD x64 SSE3 AK_v8b build available for now, will rebuild with "team" mod off and post on Lunatics in V10 pack thread. Right now you could use SSE2 x86 builds for AMD as replacement.

BTW, there is third way should be possible - you need to replace app_info.xml, leave all binaries untouched and edit Number_of_GPUs file (set zero instead of actual GPU number there). This "hack" could effectively disable re-scheduling by app itself. Didn't try that but don't see reasons why it should no work.

And don't forget to delete cc_config.xml (or edit it by removing ncpus part) file. You don't need to fool BOINC about number of CPUs in system if you want it to do correct scheduling by itself.

ADDON:
link to "non-teamed" build for AMD CPUs.
http://lunatics.kwsn.net/12-gpu-crunching/v10-of-modified-seti-mb-cuda-opt-ap-package-for-full-multi-gpucpu-use.msg16649.html#msg16649

V11 app is found in this post:

Re: V10 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use

Claggy

I just get compute errors when I try Cuda. Help, Where do I download a 6.6.21 compatible file that can do both cpu and gpu seti crunching? My video driver is 182.08 and the OS is XP x64 and I don't do AP and I won't do AP.

My cpu is a Q9300 running at 3.51GHz and crunches Seti perfectly and SLI is Disabled.

With BOINC 6.6.20 you have choice to use V10 pack as before (no additional changes needed) or to use BOINC's own scheduling ability for CUDA MB tasks. This way you should replace "teamed" CPU MB apps from pack by ordinary AK_v8 counterparts (can be downloaded from Lunatics site) and replace app_info with properly edited app_info.xml (more info about what this app_info should contain and how it should look search in this http://setiathome.berkeley.edu/forum_thread.php?id=52589 thread).

There is no non-teamed AMD x64 SSE3 AK_v8b build available for now, will rebuild with "team" mod off and post on Lunatics in V10 pack thread. Right now you could use SSE2 x86 builds for AMD as replacement.

BTW, there is third way should be possible - you need to replace app_info.xml, leave all binaries untouched and edit Number_of_GPUs file (set zero instead of actual GPU number there). This "hack" could effectively disable re-scheduling by app itself. Didn't try that but don't see reasons why it should no work.

And don't forget to delete cc_config.xml (or edit it by removing ncpus part) file. You don't need to fool BOINC about number of CPUs in system if you want it to do correct scheduling by itself.

ADDON:
link to "non-teamed" build for AMD CPUs.
http://lunatics.kwsn.net/12-gpu-crunching/v10-of-modified-seti-mb-cuda-opt-ap-package-for-full-multi-gpucpu-use.msg16649.html#msg16649

My system is an Intel Q9300 M1(A Quad Core cpu) system, But then I have no AMD PCs anymore, Not for at least a week and a half that is. Oh and I run SSSE3x.

Update:

4/11/2009 7:20:22 AM

CUDA devices:
GeForce GTX 295 (driver version 18208, CUDA version 1.3, 896MB, est. 106GFLOPS),
GeForce GTX 295 (driver version 18208, CUDA version 1.3, 896MB, est. 106GFLOPS)

This is My app info file:


<app_info>
<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.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>
<file_info>
<name>MB_6.08_mod_CUDA_V10.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>528</version_num>
<file_ref>
<file_name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.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>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V10.exe</file_name>
</file_ref>
</app_version>

<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>603</version_num>
<file_ref>
<file_name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.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>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V10.exe</file_name>
</file_ref>
</app_version>

<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<file_ref>
<file_name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.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>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V10.exe</file_name>
</file_ref>
</app_version>

</app_info>

---

Savoir-Faire is everywhere!
The T1 Trust, T1 Class 4-4-4-4 #5550, America's First HST

Then just leave AMD-related build for AMD users ;)

F.A.O. Raistmer
When I run Task Manager the A/pulse opp app shows up as
ap_5.03r112_sse3.exe*32 and cuda shows up as setiathome_6.08_windows_intelx86__cuda.exe*32
does that mean that they are 32bit apps and if so are
there 64bit apps and should I be running them?
I d/loaded what I thought were for 64bit o/s's.
I am running Vista o/s 64bit.
Big learning curve this for a Silver Surfer.
Dave

---

F.A.O. Raistmer
When I run Task Manager the A/pulse opp app shows up as
ap_5.03r112_sse3.exe*32 and cuda shows up as setiathome_6.08_windows_intelx86__cuda.exe*32
does that mean that they are 32bit apps and if so are
there 64bit apps and should I be running them?
I d/loaded what I thought were for 64bit o/s's.
I am running Vista o/s 64bit.
Big learning curve this for a Silver Surfer.
Dave

That looks like the original stock app name, so that suggests that you did not install the optimized app properly.

---

Your CUDA part is stock indeed, and opt AP 5.03 is available only in 32-bit version for Windows. There were some precision improvements that allowed x64 build with version 5.04 and up. Current beta version 5.05 will be available in x64 variant too (some speed improvement is achieved).

Thank's both for that the box is working ok as it is and doing 8 AP's in 20hrs.
Is that about right?

---

Thank's both for that the box is working ok as it is and doing 8 AP's in 20hrs.
Is that about right?

My Q9450 does AP in 10-11hours approx.

Thank's both for that the box is working ok as it is and doing 8 AP's in 20hrs.
Is that about right?

Looking at your results it seems you are using optimized AP and standard CUDA. For the optimized AP times, yes, that is normal.

---