Hi Folks,

I just dropped a 3850 (AGP) into a 2.8G P4 running WinXP32 and am hoping that I can get the GPU app running. I'm having trouble with Catalyst and I don't understand what's going on. After installing the driver and CCC that came with the card, I see that I have CCC version 2008.0109.2141.38743 (according to the CCC info display). What's the shorthand version (e.g., 8.12, 9.1, etc.)? Regardless of what version it is, though, I can find no atical*.dll files in the windows\system32 directory.

Also, if I try to install the 8.12 version (or any of the 9.x versions at guru3d), the install stops mid-stream and I get a message saying that it's unable to find a driver compatible with my software. The only option is to ack the message and exit the install.

Any ideas what I'm doing wrong?

Thanks,

MarkR

---

Hi Folks,

I just dropped a 3850 (AGP) into a 2.8G P4 running WinXP32 and am hoping that I can get the GPU app running. I'm having trouble with Catalyst and I don't understand what's going on. After installing the driver and CCC that came with the card, I see that I have CCC version 2008.0109.2141.38743 (according to the CCC info display). What's the shorthand version (e.g., 8.12, 9.1, etc.)? Regardless of what version it is, though, I can find no atical*.dll files in the windows\system32 directory.

Also, if I try to install the 8.12 version (or any of the 9.x versions at guru3d), the install stops mid-stream and I get a message saying that it's unable to find a driver compatible with my software. The only option is to ack the message and exit the install.

Any ideas what I'm doing wrong?

Thanks,

MarkR

Mark, I had a similar problem with my Powercolor 3850. 8.12 would not recognise the card, but 8.10 did (disc which came with the card). I then just copied the required files from the directory with 8.12 in it across to system32 and I was able to crunch the MW GPU app.

Mark, I had a similar problem with my Powercolor 3850. 8.12 would not recognise the card, but 8.10 did (disc which came with the card). I then just copied the required files from the directory with 8.12 in it across to system32 and I was able to crunch the MW GPU app.

Thanks for the tip! I'll give this a try. Sorry, but I'm not sure I understand what you mean by "...copied the required files from the directory with 8.12 in it across to system32..." If you installed 8.10, how is it that you copied files from the 8.12 directory? Can you elaborate a bit?

---

GPU-Z (google it) is a tool which will tell you what driver you have.

You may have to go to your vendors website (Sapphire, Gigabyte, Asus or w/e) to get 8.12 drivers for your card.

There's also an ATi AGP hotfix you have to apply - I did see a link to it once. I think it's in the forums somewhere.

---

Also, if I try to install the 8.12 version (or any of the 9.x versions at guru3d), the install stops mid-stream and I get a message saying that it's unable to find a driver compatible with my software. The only option is to ack the message and exit the install.

Hi Mark

I had this problem with my AGP 3850
The workaround is to let it fail (as it has) then manually add new hardware/Display Adaptor, select Have Disk and point it to the .inf file in the newly created ATI/SUPPORT/driver_version/XP_INF directory.
I selected the first 3850 driver and its worked fine

Its a while ago that I got mine running and the above is from memory, so it may be wrong in places but in general it should work

GPU-Z (google it) is a tool which will tell you what driver you have.

You may have to go to your vendors website (Sapphire, Gigabyte, Asus or w/e) to get 8.12 drivers for your card.

There's also an ATi AGP hotfix you have to apply - I did see a link to it once. I think it's in the forums somewhere.

Actually, if you go to the ATI website under legacy versions and scroll to 8.12 (at least for win 32-bit), the hotfix should be there if I remember correctly.

---

GPU-Z (google it) is a tool which will tell you what driver you have.

You may have to go to your vendors website (Sapphire, Gigabyte, Asus or w/e) to get 8.12 drivers for your card.

There's also an ATi AGP hotfix you have to apply - I did see a link to it once. I think it's in the forums somewhere.

That's it! I went to the Sapphire website and downloaded the hotfix for CCC 9.6. After the copy/rename workaround, I'm in business! Thanks a lot for all the feedback, you guys! :-)

---

After getting the GPU app running in general, I have at least one problem. How do I keep the BOINC manager from running more than one WU at a time. I've added "n1 w1.1" to the <cmdline> parameters in the app_info file, but it will still try to run two or sometimes more at once. Shortly after that I get what I think is a VPU recovery, although my display never comes back until I re-boot (VPU recovery is enabled in the CCC). I was running BOINC 6.6.20 but downgraded to 6.4.7 and I still have the same problem.

This is a P4/WinXP/3850 host running one (non-MW) CPU project in addition to the MW GPU app.

---

@ritterm:

That's the big question, how to stop the phantom running WUs. I'm sure there are more than one way to do it, but I do it by keeping the CPU thinking it is very busy.

It's probably easier to let MW try to run 2 or 3 WUs at the same time, up to 3 doesn't really seem to affect the run times and might possibly be a little more efficient.

I've run MW ATI with SETI, Spinhenge, Enigma, and Aqua (not all at the same time) and it has been fairly effective in keeping the Phantoms away. By keeping my cache settings at "connect every 1 day" and "keep an additional 1 day" of WUs and a MW Resource Share at 500 and the other projects at the default 100.

This works for me, I'm sure others have different methods, it just takes some experimenting to find out what works for your machine.

---

Calm Chaos Forum...Join Calm Chaos Now

Thanks, Labbie. I think, though, that I misunderstand what you're saying, you misunderstand my post, or I didn't write it very clearly. :D

You wrote...

It's probably easier to let MW try to run 2 or 3 WUs at the same time, up to 3 doesn't really seem to affect the run times and might possibly be a little more efficient.

My problem is that if more than 1 MW GPU WU runs my display crashes and it doesn't recover. Everything runs fine when only one MW GPU WU is underway. I can be running a CPU WU from any other project as well, or not, and there's no problem.

---

Yep, I misunderstood.

Don't have a ready answer for your real problem tho', Sorry.

What version of the ATI drivers are you using?

---

Calm Chaos Forum...Join Calm Chaos Now

Looks like he is running 9.6, which crashed almost everyones computers.

I have tried both 9.6 & 9.7 and have went back to 9.5 as the most stable of the recent drivers.

---

Yep, I misunderstood.
Don't have a ready answer for your real problem tho', Sorry.

No problem at all. :-)

What version of the ATI drivers are you using?

This is from the CCC:

Driver Packaging Version 8.62-090520m1-081751E-ATI
Catalyst™ Version 09.6

These were installed from a hotfix provided by Sapphire. CCC 8.12 would not install and they didn't have a fix for some of the older versions.

---

CCC 8.12 would not install and they didn't have a fix for some of the older versions.

Of course there is ;)
Catalyst 8.12 AGP hotfix driver package, WinXP32

CCC 8.12 would not install and they didn't have a fix for some of the older versions.

Of course there is ;)
Catalyst 8.12 AGP hotfix driver package, WinXP32

Well, yes, of course! :-) I'm a little hesitant to try this because I've never been able to install 8.12 -- the install stops mid-stream saying that no driver can be found that's compatible with my OS/hardware. Also, the description doesn't sound like it addresses anything that might have to do with a 3850 card.

I may still, but I don't want to lose sight of the main issue: Why is the BOINC manager allowing more than one GPU WU to run when I have "n1" in the <cmdline> parameters? I've tried several different versions of BOINC software, 6.4.7, 6.5.0, 6.6.20, and 6.6.36, and they all have the same issue.

Here's my app_info file. Is it possible something is corrupted?

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>astronomy_0.19_ATI_SSE2f.exe</name>
<executable/>
</file_info>
<file_info>
<name>brook.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>19</version_num>
<flops>1.0e11</flops>
<avg_ncpus>0.1</avg_ncpus>
<max_ncpus>1</max_ncpus>
<cmdline>n1 w1.1</cmdline>
<file_ref>
<file_name>astronomy_0.19_ATI_SSE2f.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>brook.dll</file_name>
</file_ref>
</app_version>
</app_info>

---

I may still, but I don't want to lose sight of the main issue: Why is the BOINC manager allowing more than one GPU WU to run when I have "n1" in the <cmdline> parameters?

Because the BOINC client simply doesn't know anything of your ATI card and also not what this n1 parameter means. It is just used for the coordination of several instances of the ATI application launched by the BOINC client. That parameter has no influence whatsoever how many instances are launched. You have to control that with the resource share and the ncpu settings.

And that behaviour won't go away as long the BOINC client does not recognize ATI cards (like it is doing with CUDA capable nvidia cards). I guess we can wait another year until the Berkeley crowd finally starts something, or we hope that Crunch3r is a bit more determined to get it done.

From what I'm hearing it's going quite well forward. I wouldn't be surprised if within the next month some project suddenly supports that even from the server side, i.e. that a modified client reports a CAL capable ATI card, the server sends the appropriate GPU application and the ATI GPU is even shown on the project website. So it would look something like that:


And yes, that's a real screenshot ;)

Because the BOINC client simply doesn't know anything of your ATI card and also not what this n1 parameter means. It is just used for the coordination of several instances of the ATI application launched by the BOINC client. That parameter has no influence whatsoever how many instances are launched. You have to control that with the resource share and the ncpu settings.

Thank you. Thank you, very much. :-) I have a only a basic concept of how all this works and clearly misunderstand the discussion of the "n" parameter at the beginning of this thread and the readme file that comes with the 19f app.

I'll try tweaking some of the other parameters to get something that works and makes it worthwile. What I'd like to do is have one WU in progress all the time, with some in the queue.

Thanks, again!

---

And that behaviour won't go away as long the BOINC client does not recognize ATI cards (like it is doing with CUDA capable nvidia cards). I guess we can wait another year until the Berkeley crowd finally starts something, or we hope that Crunch3r is a bit more determined to get it done.

Boinc client with ATI/CAL support

I'm guessing Cluster Physik refers to that one :p

---

Join Support science! Joinc Team BOINC United now!

Because the BOINC client simply doesn't know anything of your ATI card and also not what this n1 parameter means. It is just used for the coordination of several instances of the ATI application launched by the BOINC client. That parameter has no influence whatsoever how many instances are launched. You have to control that with the resource share and the ncpu settings.

Thank you. Thank you, very much. :-) I have a only a basic concept of how all this works and clearly misunderstand the discussion of the "n" parameter at the beginning of this thread and the readme file that comes with the 19f app.

I'll try tweaking some of the other parameters to get something that works and makes it worthwile. What I'd like to do is have one WU in progress all the time, with some in the queue.

Thanks, again!

There is a way around it, but it does mean that your CPU won't get any work.

Change to <avg_ncpus>1</avg_ncpus> or however many threads your system thinks it has (so if it's a P4 with hyperthreading, 2). That means only one WU will say it is working at a time.

The AGP cards are a bit flaky - there's nothing we can do.

To CP: Hi CP, any plans to rebuild the ATi code on the latest version of catalyst? Or is the Stream/CAL SDK technically still not compatible with it? I've been put on the CUDA side with work - getting a 1000x increase (in floats rather than doubles) in my finite difference algorithms thanks to rigorous abuse of texture memory and millions of threads :)

---

I will be testing it on my Quad once I complete the work that is on the machine already. Don't want to trash anything if something goes wrong.

---