There was 2 entries for GTX 560 in 275 drivers, and looking for lines with NVIDIA_DEV in it, there was 6 lines in all, I copied them into 266.58 and they installed normally, all seems to work...

Can you post these 6 lines?
(not only for me - this will help other people wanting to do this hack)

One strange thing though, BOINC calls my card NVIDIA D13U-15... Don't know what's up with that??

It seems you copied some line specific for DELL (?)
http://www.impactcomputers.com/wgp2g.html
NVIDIA_DEV.1081.01 = "NVIDIA D13U"
NVIDIA_DEV.1082.01 = "NVIDIA D13U "
NVIDIA_DEV.1083.01 = "NVIDIA D13U " "

(The name at the right may be anything
NVIDIA_DEV.1083.01 = "NVIDIA JohnDK SuperDuper GTX 560"

You may do this for fun (but only if you know the proper DeviceID it will have effect.)
)


I'm interested to see your real DeviceID (copy it from SIV or Device Manager)

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

The drivers will use a full CPU core anyway with actual drivers.
Claggy posted a message at beta from a dev at collatz about it.
If you dont believe me maybe you believe a developer.

CPU utilization while running GPU OpenCL applications is a problem for NVidia as well as Intel. The application uses only asynchronous kernel and memory copy functions but the driver "overrides" and decides to block anyway. One "hack" suggested by NVidia was to create allocate a large number of command queues so that the driver would think it should use the actual asynchronous calls, but that has not worked for me. If it wasn't driver related, then why would the same application work fine on AMD GPUs? So, write your local Intel developer and ask them why their driver needs CPU time to run asynchronous OpenCL kernels. You would think it would work properly with NVidia GPUs since OpenCL came from them. Nope. NVidia can't even follow their own specs properly. And it they can't, why should Intel?

Original post.

http://boinc.thesonntags.com/collatz/forum_thread.php?id=1019&postid=16833
I'm taking that with a small pinch of salt, because the part I've picked out doesn't correspond with other sources.

OpenCL is (currently) the responsibility of the Khronos Group: http://www.khronos.org/opencl/.

According to Wiki (I know, not an authoritative source), it was originated by Apple: http://en.wikipedia.org/wiki/OpenCL#History

Sorry for the delay, the screen shot below shows the CPU setting of (.5) that I previously posted. So it is possible to run 2 WUs on a single GPU using only 1/2 of a core under the conditions that I described. I noticed that JohnDK was successful in hacking the driver for the GTX560 on his i7/870, with a decrease in memory size. If someone didn't know what the GPU and its size were prior to the hack, there is no way to know what it is now, but that is beside the point. But, I am curious to see how many WUs the hacked driver is successfully processing.

---

I don't buy computers, I build them!!

Sorry for the delay, the screen shot below shows the CPU setting of (.5) that I previously posted. So it is possible to run 2 WUs on a single GPU using only 1/2 of a core under the conditions that I described. I noticed that JohnDK was successful in hacking the driver for the GTX560 on his i7/870, with a decrease in memory size. If someone didn't know what the GPU and its size were prior to the hack, there is no way to know what it is now, but that is beside the point. But, I am curious to see how many WUs the hacked driver is successfully processing.

So far the driver works fine, no errors and several validated, and I can now run 7 CPU APs instead of 5 as I did before, total CPU use is about the same.

You can check how it goes if interested here, I btw don't have the PC running 24/7.

http://setiathome.berkeley.edu/results.php?hostid=5284345&offset=0&show_names=0&state=4&appid=12

Sorry for the delay, the screen shot below shows the CPU setting of (.5) that I previously posted. So it is possible to run 2 WUs on a single GPU using only 1/2 of a core under the conditions that I described. I noticed that JohnDK was successful in hacking the driver for the GTX560 on his i7/870, with a decrease in memory size. If someone didn't know what the GPU and its size were prior to the hack, there is no way to know what it is now, but that is beside the point. But, I am curious to see how many WUs the hacked driver is successfully processing.

So far the driver works fine, no errors and several validated, and I can now run 7 CPU APs instead of 5 as I did before, total CPU use is about the same.

You can check how it goes if interested here, I btw don't have the PC running 24/7.

http://setiathome.berkeley.edu/results.php?hostid=5284345&offset=0&show_names=0&state=4&appid=12

Congrats!! I'm glad it's working for you. But the same thing could have been accomplished by doing what I suggested without all of the h***ll and aggravation and it would have been a lot cleaner.

---

I don't buy computers, I build them!!

There was 2 entries for GTX 560 in 275 drivers, and looking for lines with NVIDIA_DEV in it, there was 6 lines in all, I copied them into 266.58 and they installed normally, all seems to work...

Can you post these 6 lines?
(not only for me - this will help other people wanting to do this hack)

One strange thing though, BOINC calls my card NVIDIA D13U-15... Don't know what's up with that??

It seems you copied some line specific for DELL (?)
http://www.impactcomputers.com/wgp2g.html
NVIDIA_DEV.1081.01 = "NVIDIA D13U"
NVIDIA_DEV.1082.01 = "NVIDIA D13U "
NVIDIA_DEV.1083.01 = "NVIDIA D13U " "

(The name at the right may be anything
NVIDIA_DEV.1083.01 = "NVIDIA JohnDK SuperDuper GTX 560"

You may do this for fun (but only if you know the proper DeviceID it will have effect.)
)


I'm interested to see your real DeviceID (copy it from SIV or Device Manager)

This is the 6 lines I copied from 275 drivers to 266 drivers, placing them under the same sections. I assume one of the 2 sets is enough, but since I don't know which is the right I just copied both. The lines ins't DELL specific, so I wonder what's missing since the PC don't show the correct card name. Maybe I can correct it with regedit...

NVIDIA_DEV.1084.01 = "NVIDIA GeForce GTX 560"
%NVIDIA_DEV.1084.01% = Section004, PCI\VEN_10DE&DEV_1084
%NVIDIA_DEV.1084.01% = Section005, PCI\VEN_10DE&DEV_1084

NVIDIA_DEV.1201.01 = "NVIDIA GeForce GTX 560"
%NVIDIA_DEV.1201.01% = Section004, PCI\VEN_10DE&DEV_1201
%NVIDIA_DEV.1201.01% = Section005, PCI\VEN_10DE&DEV_1201

Sorry for the delay, the screen shot below shows the CPU setting of (.5) that I previously posted. So it is possible to run 2 WUs on a single GPU using only 1/2 of a core under the conditions that I described. I noticed that JohnDK was successful in hacking the driver for the GTX560 on his i7/870, with a decrease in memory size. If someone didn't know what the GPU and its size were prior to the hack, there is no way to know what it is now, but that is beside the point. But, I am curious to see how many WUs the hacked driver is successfully processing.

So far the driver works fine, no errors and several validated, and I can now run 7 CPU APs instead of 5 as I did before, total CPU use is about the same.

You can check how it goes if interested here, I btw don't have the PC running 24/7.

http://setiathome.berkeley.edu/results.php?hostid=5284345&offset=0&show_names=0&state=4&appid=12

Congrats!! I'm glad it's working for you. But the same thing could have been accomplished by doing what I suggested without all of the h***ll and aggravation and it would have been a lot cleaner.

How much CPU do your APs use when running? Before mine used about 13% each, the same as CPU APs use, now they use mostly from 0-3% (more if high blanked).

The older drivers help with the CPU usage problem on AP tasks, but don't the CUDA42 and CUDA50 versions of the MB applications require later drivers?

---

Soli Deo Gloria

The older drivers help with the CPU usage problem on AP tasks, but don't the CUDA42 and CUDA50 versions of the MB applications require later drivers?

Yes I think so, for now I'm running AP only...

I know the new drivers works different the the older ones when it comes to this GPU app, I just wish I/we could used latest drivers without GPU APs suffering. I know nothing about programming, so if it's possible re-coding it to work better with the new drivers I have no idea...

Sorry for the delay, the screen shot below shows the CPU setting of (.5) that I previously posted. So it is possible to run 2 WUs on a single GPU using only 1/2 of a core under the conditions that I described. I noticed that JohnDK was successful in hacking the driver for the GTX560 on his i7/870, with a decrease in memory size. If someone didn't know what the GPU and its size were prior to the hack, there is no way to know what it is now, but that is beside the point. But, I am curious to see how many WUs the hacked driver is successfully processing.

So far the driver works fine, no errors and several validated, and I can now run 7 CPU APs instead of 5 as I did before, total CPU use is about the same.

You can check how it goes if interested here, I btw don't have the PC running 24/7.

http://setiathome.berkeley.edu/results.php?hostid=5284345&offset=0&show_names=0&state=4&appid=12

Congrats!! I'm glad it's working for you. But the same thing could have been accomplished by doing what I suggested without all of the h***ll and aggravation and it would have been a lot cleaner.

How much CPU do your APs use when running? Before mine used about 13% each, the same as CPU APs use, now they use mostly from 0-3% (more if high blanked).

I use Outertech's Cachman 7.5.0 to monitor what's running on my machines. All AP tasks average about 12.1 -> 12.3 and max out at 12.5. I've seen the GPU tasks go down to 9.1. This is with 6 cores (80%) reserved for BOINC/Seti, even with this the remaining 20% is taken up running other things on the box; so I'm running 100% 24/7. Btw, as of this post only 2.1Gb out of 6 is being used and no 0 for the page file.

I wish I'd have access to a GTX560 2GB.

---

I don't buy computers, I build them!!

I know nothing about programming, so if it's possible re-coding it to work better with the new drivers I have no idea...

Unfortunately, the problem is on the driver's side, not the application side. At least, that's how I understand it.

---

Soli Deo Gloria

I know nothing about programming, so if it's possible re-coding it to work better with the new drivers I have no idea...

Unfortunately, the problem is on the driver's side, not the application side. At least, that's how I understand it.

That depends on your point of view, if you're Raistmer it's a driver synchronisation problem/Bug, if it's Jason it's an app problem where the app should be using a callback instead:

http://setiathome.berkeley.edu/forum_thread.php?id=70723&postid=1340324

If so, could somebody put together a recommended toolset and programming guide for study?

Like the nVidia OpenCL Ocean Demo (from the GPU compute SDK) ?

I'll see if I can find a pull quote from there (I should have it somewhere, else my download link is going to get some more exercise) that I can pass on to Raistmer.

Will have to find it again myself, though I remember that using an OpenGL callback in a renderloop for OpenCL 1.0, since OpenCL didn't get them until 1.1.

Pity the OpenCL code samples aren't broken out into a separate download - looks like I need the whole GPU Computing [sic] SDK download (that's the same thing, I take it?)

None of

CUDA C/C++ Code Samples
DirectCompute Code Samples
CUDA Library Samples

look hopeful.

Things have moved around. The oclSimpleGL sample implements the Callbacks using OpenGL, and the ocean simulation has been moved to DirectCompute Using DirectX render callbacks. In either case (both workably demonstrating callbacks & very low CPU usage, with high GPU usage), the basic computation changes from the old style:

Setup device & Kernels etc.
main processing loop:
    Do some computation on GPU
    Hard Synchronise (implicit or explicit)
    Transfer back to CPU for postprocessing

to:
Setup device & Kernels etc.  Context with Vsync off where applicable.
Setup Callbacks (OpenGL, DirectX, OpenCL1.1+ or CUPTi)
Sleep loop 
       Check for exit requests etc
       
with callback ('render') function:
  Do Some computation on GPU
  transfer back to CPU & postprocess where needed  
  [Optionally] draw something

The core changes being only that processing comes out of the main polled/blocking loop, and into a callback called by either a graphics context or a callback interface as supplied by OpenCL 1.1 or Cupti without the need for a graphics context.

Claggy

I know nothing about programming, so if it's possible re-coding it to work better with the new drivers I have no idea...

Unfortunately, the problem is on the driver's side, not the application side. At least, that's how I understand it.

Is it only the GTX560 that has had this problem? The reason why I asked is that prior to installing the 2nd GTX660SC to the i7/950 machine, the 2nd GPU was a GTX460SE v1 1Gb that used the same settings and driver and processed the same type of WUs with no problems. Even though the GTX460SE was a Fermi, it still ran Cuda50, because of the GTX660SC, again with no problems.

---

I don't buy computers, I build them!!

This is the 6 lines I copied from 275 drivers to 266 drivers, placing them under the same sections. I assume one of the 2 sets is enough, but since I don't know which is the right I just copied both. The lines ins't DELL specific, so I wonder what's missing since the PC don't show the correct card name. Maybe I can correct it with regedit...

NVIDIA_DEV.1084.01 = "NVIDIA GeForce GTX 560"
%NVIDIA_DEV.1084.01% = Section004, PCI\VEN_10DE&DEV_1084
%NVIDIA_DEV.1084.01% = Section005, PCI\VEN_10DE&DEV_1084

NVIDIA_DEV.1201.01 = "NVIDIA GeForce GTX 560"
%NVIDIA_DEV.1201.01% = Section004, PCI\VEN_10DE&DEV_1201
%NVIDIA_DEV.1201.01% = Section005, PCI\VEN_10DE&DEV_1201

You may search the .inf file for D13U-15

For 'correct it with regedit' it is maybe possible (unless some checksum exists, IIRC months ago I came across a checksum but for ATI to protect some settings in registry from manual edit )
But I found ~20 places in registry which contain the string 'NVIDIA GeForce 6150SE nForce 430' (this is my disabled in the BIOS integrated GPU)

Instead of using RegEdit directly to search I use RegScanner (it shows all what is found in a single list, use F8 or Right-click to open the same place by RegEdit)
http://www.nirsoft.net/utils/regscanner.html


Edit:
I have the file 266.58_desktop_winxp_32bit_english_whql.exe and in it NV4_DISP.INF
Search for D13U do not find anything.

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

You may search the .inf file for D13U-15

For 'correct it with regedit' it is maybe possible (unless some checksum exists, IIRC months ago I came across a checksum but for ATI to protect some settings in registry from manual edit )
But I found ~20 places in registry which contain the string 'NVIDIA GeForce 6150SE nForce 430' (this is my disabled in the BIOS integrated GPU)

Instead of using RegEdit directly to search I use RegScanner (it shows all what is found in a single list, use F8 or Right-click to open the same place by RegEdit)
http://www.nirsoft.net/utils/regscanner.html


Edit:
I have the file 266.58_desktop_winxp_32bit_english_whql.exe and in it NV4_DISP.INF
Search for D13U do not find anything.

Used regscanner and found 14 entries with D13U-15, I saved it as a .reg and changed D13U-15 into geforce gtx 560... I saw regedit changing a key when I imported the .reg file, but after a reboot it was all changed back to D13U-15. Damn.

That depends on your point of view, if you're Raistmer it's a driver synchronisation problem/Bug, if it's Jason it's an app problem where the app should be using a callback instead:

http://setiathome.berkeley.edu/forum_thread.php?id=70723&postid=1340324

I remember that thread. Are both Raistmer and Jason referring to the AP NV application though? I thought Jason focuses on MB CUDA application development, though I can't remember if he was lending his expertise for the OpenCL side of things.

---

Soli Deo Gloria

Do a search for text 'D13U' in all files found in:
C:\WINDOWS\inf\

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

That depends on your point of view, if you're Raistmer it's a driver synchronisation problem/Bug, if it's Jason it's an app problem where the app should be using a callback instead:

http://setiathome.berkeley.edu/forum_thread.php?id=70723&postid=1340324

I remember that thread. Are both Raistmer and Jason referring to the AP NV application though? I thought Jason focuses on MB CUDA application development, though I can't remember if he was lending his expertise for the OpenCL side of things.

The general advance of computer science and computational techniques over the decades isn't restricted to one type of search over the other.

Do a search for text 'D13U' in all files found in:
C:\WINDOWS\inf\

Total commander didn't find any :(