But, if I insert 'Use Central Processing Unit (CPU)' -> NO, the GPU-CPU-support will not stop?
[I don't think so.. but.. sometimes there are BUGs.. ;-)]
[/color]

I haven't tried it but it would not be logical since the GPU can't work at all without being primed by the CPU.

Could be a while before you can test it though since there will be no new MB work "for up to a week".

F.

---

But, if I insert 'Use Central Processing Unit (CPU)' -> NO, the GPU-CPU-support will not stop?
[I don't think so.. but.. sometimes there are BUGs.. ;-)]
[/color]

I haven't tried it but it would not be logical since the GPU can't work at all without being primed by the CPU.

Could be a while before you can test it though since there will be no new MB work "for up to a week".

F.

..yes, you are right..
Sadly..

Maybe will test then the V6.6.17 if all the time CUDA will run..


But finally I don't know what the problem was. [CUDA finished not well and because of this one GPU idle]
The BOINC version or Raistmer's V7 mod?
Maybe I will go back to V6.4.5 ..
But I read that this version have a 'security prob'. :-(

[EDIT: Or will get again V6.6.11 a try.]


BTW.
Nice for people with 10 day cache.. ;-)

---

Of course - I could/would let run stock CUDA, but not current with the 'VLAR-BUG'.

Somebody know when > CUDA V6.08 will be available?

---

You have all the time 4 x MB and 2 x CUDA?
[This will happen if all stock?
Also with opt. apps?]

No, my CPU cores run mainly optimised AP. Have you noticed that there is now an option on your website preferences that allows you to select NOT to use the CPU? Surely that would achieve what you are aiming for?

F.

OPS.. Thanks! ..something new in 'SETI@home preferences'.

Why they didn't published this? ;-)

As helpful as it would be they never publish a thing.

---

me@rescam.org

Of course - I could/would let run stock CUDA, but not current with the 'VLAR-BUG'.

Somebody know when > CUDA V6.08 will be available?

The Vlar bug is now dead with current 6.08 at a cost of time and Rac. If you run latest cuda paired with latest berkeley official cuda app you CAN run vlar apps but they run veeeery slow, that's why that vlar kill exist for now. At first when older cuda were available you got an app crash or hung cuda executable.

Kind regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Can someone plz tell me how to set up this box to run optimized apps plus cuda...I was running sse3 before I got the new video card now just cuda and the normal app. want to go to an sse3 opti app and cuda for 64bit amd that's what i need.

thanks,
ganja

@ -= Vyper =-

Yes.. I meant with VLAR-BUG the very very very slow VLAR crunching.. :-)


@ ganja

Take the latest version of Raistmer's mod. [10/11]
Follow the manual.. and everything would go well.

You can use SSE3 opt. only for CPU.
The CUDA app have no extension opt.


@ Raistmer

Which mods have free CPU-Core usage/support? [Not fixed at CPU-Core #0]
Only V7?


Maybe it's possible to reduce the time of the CPU usage at the start of the CUDA WUs?

• Maybe delete the 'CPU fall back' calculation?
  
  
• Optimize the whole CPU start?
  

---

@ Raistmer

I use your V7 mod. Only with CUDA entry.

If I change the V7 CUDA app with this following:

MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update.rar

with

MB_6.08_mod_CUDA_V10.exe


It will work?

The ..CUDA_V10.exe could run also alone without the other apps and 'number of GPUs file'?
So only this app in 'stand alone' with app_info.xml with only CUDA entry? [and V7 .dll's]

---

Maybe it's possible to reduce the time of the CPU usage at the start of the CUDA WUs?

done already - see last updates.
about affinity: for last builds affinity binding disabled.

In general, if you don't see such line "Affinity of worker thread adjusted successfully" in stderr - affinity not setted.

@ Raistmer

I use your V7 mod. Only with CUDA entry.

If I change the V7 CUDA app with this following:

MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update.rar

with

MB_6.08_mod_CUDA_V10.exe


It will work?

The ..CUDA_V10.exe could run also alone without the other apps and 'number of GPUs file'?
So only this app in 'stand alone' with app_info.xml with only CUDA entry? [and V7 .dll's]

it should work, but I didn't try.

it should work, but I didn't try.

Thanks a lot!

Later I'll make the update and will see if I'll lose some WUs.. ;-)

---

it should work, but I didn't try.

Thanks a lot!

Later I'll make the update and will see if I'll lose some WUs.. ;-)

You should lose no tasks while upgrading at all. Just follow rules for upgrading...

@ Raistmer

Hmm.. I told you since your V9 mod, that it would be better not to fix the GPU/s at CPU-Core #0 ! ;-D



Yes, nothing lost. It was little kidding with ';-)' marked.. ;-D


I use now your:
MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update.rar
with
MB_6.08_mod_CUDA_V10.exe

..and selfmade app_info.xml mod.. ;-)


[I recommend this app for all 'only' GPU cruncher! ;-)]


Hey! Well done job! *thumb up*

The 'GPU WU- CPU prepare time' is from ~ 25 sec. down to ~ 12 sec.!

~ 50 % faster! This is opt. which I like! :-D

[EDIT:
How you done it?
What you had done?
Your are 'David Copperfield'?? ;-)
Is there more space for to opt. the CUDA app?]


Hope the app will continue without freezing like other saw..



Something what is to say from time to time.. ;-)

Thanx a lot for your work! *thumb up*
[And of course thanx a lot also to the whole opt. crew!]

I hope we see you (all) here endless for to support the project!

Your work is much appreciated from the whole SETI@home community!

---

...
MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update.rar
...
The 'GPU WU- CPU prepare time' is from ~ 25 sec. down to ~ 12 sec.!
...

The speedup is mainly from FFTW_ESTIMATE, which is the fastest method FFTW has for planning what code it is going to use. Stock and earlier Raistmer builds use FFTW_MEASURE which actually times execution of the possible plans. The estimate doesn't get the fastest FFTs, but that would only affect CPU fallback mode (perhaps as much as 20% slower in that mode). Another minor speedup came from not testing the 'optimal functions' for the CPU to use in fallback; only the default functions are available so testing isn't needed.

                                                                 Joe

Bump! I saw on the lunatics site you have the V11 out for special Cuda problems would you like to start a new thread for that

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Bump! I saw on the lunatics site you have the V11 out for special Cuda problems would you like to start a new thread for that

No, it's special build and not ready for general audience.

Now I found a - for me - new error..
Something to worry about?

[V10 mod]
The new error:
http://setiathome.berkeley.edu/result.php?resultid=1195160891

Exception detected inside cudaAcc_find_triplets, dumping client state
icfft=86665, PoT_activity=0, PoT_freq_bin=-1SETI@home error -12 Unknown error
cudaAcc_find_triplets erroneously found a triplet twice in find_triplets_kernel
File: ..\analyzePoT.cpp
Line: 348

[V7 mod]
The last errors were:

SETI@home error -12 Unknown error
cudaAcc_find_triplets erroneously found a triplet twice in find_triplets_kernel
File: d:/BTR/seticuda/seti_boinc/client/cuda/cudaAcc_pulsefind.cu
Line: 236

BTW.
I guess it's not an error because of your mod Raistmer.. but I think you [or other profis] could know/guess what does it is?

---

It's a well-known error - here's one I had earlier today with the stock application: task 1196503081

SETI@home error -12 Unknown error
cudaAcc_find_triplets erroneously found a triplet twice in find_triplets_kernel
File: c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcc_pulsefind.cu
Line: 236

This, and many other, rough edges await the next round of development - which has remained fundamentally unchanged (save the odd kludge and tweak) since 21 January 2009.

The mentioned 'last errors' are known..
[Line: 236]

But the new one look little bit different.
[Line: 348]
It's also known, or new?

---

The mentioned 'last errors' are known..
[Line: 236]

But the new one look little bit different.
[Line: 348]
It's also known, or new?

It's the same error. I just added exception handling in attempt to get more info about this bug.
It didn't work as expected in standalone run, but seems to work online :)
Not bad. Now we can have some more info about this bug.
Please, keep posting stderr output about "-12 triplet error", it contains some useful info along with error message.