I downloaded and installed the app_config.xml file and it started ripping thru the V7 GPU wus nicely at two @ a time. The it started doing APs on the GPU and suddenly the room cooled down again (humour there.. is winter in Aus so like the heat *grin*)

SO, what should I add to the app_config.xml file so that it crunches two APs at a time or is this stretching it?

(And yes, am aware that it'll be a simple answer, but it is the simple answers that sometimes lead to greater questions.)

LOL - you dug back a long way for that one - 8 pages, on my settings. Here it is en clair for reference:

<app_config>
   <app>
      <name>setiathome_v7</name>
      <gpu_versions>
          <gpu_usage>0.5</gpu_usage>
          <cpu_usage>.04</cpu_usage>
      </gpu_versions>
    </app>
</app_config>

You need to edit it (use notepad).

Keeping the outer <app_config> ... </app_config> structure intact, make a duplicate copy of the middle block (from <app> to </app>, inclusive).

In the new section you made, change 'setiathome_v7' to 'astropulse_v6'. You might well want to increase the <cpu_usage> number for AP, too. Anything 0.51 and up will free a CPU core when running two AP tasks.

6 pages on my setting.

---

Thanks for the quick reply. We Wombats are renown for out deep digging *grin*

So can I then just save the file without stopping boinc? or should I stop, save file and then restart Boinc?

(See? en clair to one is muddy waters to another *grin*)

Well, Wanted to give the Brains (ALL OF THEM) the ego rub we all need at one time or another *grin*

Oh and these last four posts are on page four for me, so wasn't that big a hole. Just had to remember which burrow to dive into LOL.

Thanks for the quick reply. We Wombats are renown for out deep digging *grin*

So can I then just save the file without stopping boinc? or should I stop, save file and then restart Boinc?

(See? en clair to one is muddy waters to another *grin*)

You can edit and save the file while BOINC is running, no problem.

When you have finished (and proofread your work), go to the 'Advanced' menu in BOINC Manager, and choose 'Read config file' - that's it, no stop/start required.

Dumb Wombat remembered reading somewhere that the file could be saved without stopping Boinc... and so now have two APs going at once, having had Boinc read the new app_config.xml file. Weirdly though, it says it (in boinc) is still using .04 CPU, yet double checked config file and it stats .04 for V7 and .51 for AP6.

Maybe should bump up the CPU usage for AP, but shall see what tomorrow.. er today brings when I check in 8 hrs.

Thanks again to all Lunatics and associates for their fine work and prompt replies.

Dumb Wombat remembered reading somewhere that the file could be saved without stopping Boinc... and so now have two APs going at once, having had Boinc read the new app_config.xml file. Weirdly though, it says it (in boinc) is still using .04 CPU, yet double checked config file and it stats .04 for V7 and .51 for AP6.

Maybe should bump up the CPU usage for AP, but shall see what tomorrow.. er today brings when I check in 8 hrs.

Thanks again to all Lunatics and associates for their fine work and prompt replies.

BOINC does not update that part of GUI until a complete restart of the client itself, but it will be using what you specify.

---

I have driver 314.07 for my Nvidia graphic card but I don't know if it's updated with a v7 aistub. How can I know or see this?

---

rOZZ
Music
Pictures

I have driver 314.07 for my Nvidia graphic card but I don't know if it's updated with a v7 aistub. How can I know or see this?

Did you run the Lunatics installer 0.41 ?
If not no of course.

---

With each crime and every kindness we birth our future.

I PM'ed you, Mike

---

rOZZ
Music
Pictures

I PM'ed you, Mike

And replied.

---

With each crime and every kindness we birth our future.

AstroPulse v6 r1843 OpenCL for NVidia.
.TXT file:
1.) High end cards (more than 12 compute units)
-unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -hp
2.) Mid range cards (less than 12 compute units)
-unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 -hp
3.) entry level GPU (less than 6 compute units)
-unroll 4 -ffa_block 2048 -ffa_block_fetch 1024 -hp

This infos are only for ATI/AMD cards?

How should be the cmdline settings for NVidia cards:
GTX260: compute units 27 - 1 - instances per device 1
GT430: compute units 2 - 3 - instances per device 2
GTX460: compute units 7 - 2 - instances per device 2
GTX580: compute units 16 - 1 - instances per device 2
GT630: compute units 2 - 3 - instances per device 2
GTX650Ti: compute units 4 - 3 - instances per device 2

- like this above?

Thanks.


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

---

I dunno...
I have been using the .txt file entries for my NV cards and they seem to be working OK.

---

"Time is simply the mechanism that keeps everything from happening all at once."

This infos are only for ATI/AMD cards?

There are five separate .TXT files distributed within the installer:

For AstroPulse applications
ReadMe_AstroPulse_Brook.txt
ReadMe_AstroPulse_OpenCL_ATI.txt
ReadMe_AstroPulse_OpenCL_NV.txt

'Brook' are old, low-performance (almost obsolete) ATI cards
The other two files are for current ATI and NVIDIA cards, respectively.

For MultiBeam applications
ReadMe_MultiBeam_OpenCL_ATI.txt
README_x41zc.txt

'x41zc' is the code number for the NVIDIA (cuda) application: the other file documents the ATI equivalent.

Make sure you refer to the documentation for the right application.

(but now you ask, I do note that the examples for 'high end' and 'mid range' cards in the OpenCL_NV file seem to look like ATI model numbers. I'll refer that one back to the OpenCL team.)

AstroPulse v6 r1843 OpenCL for NVidia.
.TXT file:
1.) High end cards (more than 12 compute units)
-unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -hp
2.) Mid range cards (less than 12 compute units)
-unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 -hp
3.) entry level GPU (less than 6 compute units)
-unroll 4 -ffa_block 2048 -ffa_block_fetch 1024 -hp

This infos are only for ATI/AMD cards?

How should be the cmdline settings for NVidia cards:
GTX260: compute units 27 - 1 - instances per device 1
GT430: compute units 2 - 3 - instances per device 2
GTX460: compute units 7 - 2 - instances per device 2
GTX580: compute units 16 - 1 - instances per device 2
GT630: compute units 2 - 3 - instances per device 2
GTX650Ti: compute units 4 - 3 - instances per device 2

- like this above?

Thanks.


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

Philip, the above should help others

If you wish, you could also add
GTX295: compute units 30 - 1 - instances per device 1

I have tried 2 instances with criteria set as 1 above however there were to many video issues and the system was unusable. I have not tried 2 instances at 2 above yet.

There is also an app that you can download, GPU Caps Viewer, that should help people with seeing how many compute units their GPU cards have.

cheers

Lionel

---

I did a lot of benchmarking on a 6670 (6 cu's) and I found out that -unroll 9 -ffa_block 2048 -ffa_block_fetch 1024 worked best with <count>.5</count> (ie running 2-up)

Diff architectures ymmv.

Richard Haselgrove wrote:

(...)
(but now you ask, I do note that the examples for 'high end' and 'mid range' cards in the OpenCL_NV file seem to look like ATI model numbers. I'll refer that one back to the OpenCL team.)

Any news?
I understood this like this, that you ask there .. (or I'm wrong?) - if the recommendations are also meant for NV cards.
If not, how are the recommendations for NV cards?

Lionel wrote:

Philip, the above should help others

If you wish, you could also add
GTX295: compute units 30 - 1 - instances per device 1

I have tried 2 instances with criteria set as 1 above however there were to many video issues and the system was unusable. I have not tried 2 instances at 2 above yet.

There is also an app that you can download, GPU Caps Viewer, that should help people with seeing how many compute units their GPU cards have.

cheers

Lionel

I took the infos about the compute units from the 'stderr output'.
SAHv7 & APv6 's <stderr_txt> show the number of CUs.

Pre-Fermi cards like 8xxx, 9xxx, 2xx series, can run only 1 WU/GPU.
I tested on my GTX260 2 (APv6) WUs/GPU, the calculation times increased for each WU.
So just 1 (SAHv7 & APv6) WU/pre-Fermi GPU.

AFAIK, the WUs are not calculated really simultaneously on GPU cards.
The hardware jump from one WU to the other WU, this feature is very slow on pre-Fermi cards.


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

---

I found no infos about CUDA versions here: http://www.nvidia.com/object/win7-winvista-64bit-266.58-whql-driver.html
http://us.download.nvidia.com/Windows/266.58/266.58_Win7_WinVista_Desktop_Release_Notes.pdf - .PDF file

- and here: http://www.nvidia.com/object/win7-winvista-64bit-275.33-whql-driver.html
- but:
http://us.download.nvidia.com/Windows/275.33/275.33-Win7-WinVista-Desktop-Release-Notes.pdf - .PDF file - here they say added CUDA v4.1.

So with this both drivers (266.58 & 275.33) you can let run max cuda32 app - cuda42 will not work.


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

---

Just so the question doesn`t reamain unanswered i repost it here as well.
The other thread might get lost.

Your NV 260 is a mid range card.

So normal settings would be -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536.
Given the fact pre Fermi`s can only handle one instance you can increase the settings.

Your settings are already improved -unroll 12 -ffa_block 8192 -ffa_block_fetch 4096.

You can use -unroll 12 -ffa_block 12288 -ffa_block_fetch 6144.
Evenso you could try -unroll 14 to 16 but use at your own risk.

Beware if you experience overflow tasks reduce the settings again until you find the sweet spot.

As always your milage might vary.

I will rework the NV readme until the next release.

BTW: You dont need to post in 2 threads.

Mike

---

With each crime and every kindness we birth our future.

Thanks.

Just so the question doesn`t reamain unanswered i repost it here as well.
The other thread might get lost.
(...)
BTW: You dont need to post in 2 threads.

Mike

This two messages you mentioned are not the same.

BTW, it looks like not all NV card user know that they can set cmdline settings ..


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

---

Which SAHv7 CUDA app should work on a:

Quadro FX 570, 256 MiB, regsPerBlock 8192
computeCap 1.1, multiProcs 2

&

Quadro NVS 420, 256 MiB, regsPerBlock 8192
computeCap 1.1, multiProcs 1


- because of the very low RAM just the cuda22 app?


Maybe the old 185.85 driver could use less RAM?
OS: Win8 x64


Thanks.


* Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *

---