GpuGrid acemdlong 6.16 (cuda42) i686/win 1WU

[Christopher Herr]

Hi everyone,

since there seems to be a 'desire'/need for a working GpuGrid app_info with CUDA 4.2 support I will give one here:

Download http://www.gpugrid.net/download/acemd.2562.cuda42
and http://www.gpugrid.net/download/cudart32_42_9.dll
and http://www.gpugrid.net/download/cufft32_42_9.dll
and http://www.gpugrid.net/download/tcl85.dll

and put all those files into the GpuGrid project folder. ;)

For Windows 7 open System settings, Advanced system settings, Environment variables and enter 'SWAN_SYNC' as new variable for the system with value '0' (exactly written that way without the quotation marks, of course).
Rumor had it, this was no longer necessary or of any use, but I respectfully disagree, it will shave off a couple/dozens of minutes of the runtimes in average, though it makes my systems at least a little unresponsive at times! Well, it is "good" ole CUDA... ;)

Another good idea is to reserve the thread/core via ProLasso ( http://bitsum.com/prolasso.php ), say thread/core 0 to 'acemd*' and give this wildcard string also high CPU priority and restrict the other say CPU applications run by BOINC to the remaining threads/cores. ;)

Content of the app_info, running one at a time with one CPU thread/core support, to be saved into a text file named 'app_info.xml' (again without quotation marks) in the GpuGrid project folder:


<app_info>
<app>
<name>acemdlong</name>
<user_friendly_name>ACEMDLONG: GPU Molecular Dynamics</user_friendly_name>
</app>
<file_info>
<name>acemd.2562.cuda42</name>
<executable/>
</file_info>
<file_info>
<name>cudart32_42_9.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft32_42_9.dll</name>
<executable/>
</file_info>
<file_info>
<name>tcl85.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>acemdlong</app_name>
<version_num>616</version_num>
<platform>windows_intelx86</platform>
<plan_class>cuda42</plan_class>
<api_version>6.7.0</api_version>
<avg_ncpus>1.000000</avg_ncpus>
<max_ncpus>1.000000</max_ncpus>
<file_ref>
<file_name>acemd.2562.cuda42</file_name>
<main_program/>
<copy_file/>
</file_ref>
<file_ref>
<file_name>cudart32_42_9.dll</file_name>
<open_name>cudart32_42_9.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>cufft32_42_9.dll</file_name>
<open_name>cufft32_42_9.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>tcl85.dll</file_name>
<open_name>tcl85.dll</open_name>
<copy_file/>
</file_ref>
<coproc>
<type>CUDA</type>
<count>1.000000</count>
</coproc>
<gpu_ram>402653184.000000</gpu_ram>
</app_version>
</app_info>

Hope this works out for you!

Note that older cards (with Compute Capability beneath 2.0) can not run the CUDA 4.2 application!
--> For those I will post the CUDA 3.1 application and app_info as well. :)
Note as well that Fermi generation cards will need 25 mV more voltage on the core with the same otherwise stable OC to run this CUDA 4.2 application reliably!

[Christopher Herr]

Just to add another thing; this application and app_info also works well on Windows 64bit! There is not a 64bit application for Windows available anyway... ;)

[Mike029]

Sweet!! Thx CH...

[Bikermatt]

This is awesome!:D The project still gives out the occasional cuda 31 task which run much slower so this will correct that problem. I do all of my GPU grid crunching on linux because it seems to crunch these tasks much faster, or at least it did in the past. I will have to run a few on windows with the new app and check my run times. I will adapt this app info to linux and try it out.

[Christopher Herr]

Sweet!! Thx CH...


This is awesome!:D The project still gives out the occasional cuda 31 task which run much slower so this will correct that problem. I do all of my GPU grid crunching on linux because it seems to crunch these tasks much faster, or at least it did in the past. I will have to run a few on windows with the new app and check my run times. I will adapt this app info to linux and try it out.

Well, your responses are awesome and thank you for the feedback! :) :-h
Yes, the CUDA 4.2 application shaves off a lot compared to CUDA 3.1.

Unfortunately I have never been a real 'crack' at Linux app_infos, but I remember trigggl having said that it works just the same way. If I had one, I would have posted that already here...

So please make sure to post that info in a Linux thread here, once you were successful! \m/ :-bd
Speaking of it, I just took a quick peek on the GpuGrid download server and I do not seem to be able to find a CUDA 4.2 application for Linux (which would be only x86_64, IIRC)... :confused:
So do you know if there really is one?

[Bikermatt]

Well, your responses are awesome and thank you for the feedback! :) :-h
Yes, the CUDA 4.2 application shaves off a lot compared to CUDA 3.1.

Unfortunately I have never been a real 'crack' at Linux app_infos, but I remember trigggl having said that it works just the same way. If I had one, I would have posted that already here...

So please make sure to post that info in a Linux thread here, once you were successful! \m/ :-bd
Speaking of it, I just took a quick peek on the GpuGrid download server and I do not seem to be able to find a CUDA 4.2 application for Linux (which would be only x86_64, IIRC)... :confused:
So do you know if there really is one?

Hmm.. Going to work on the app info tonight after work.

I'm looking in the projects directory and I have:
acemd.2562.x64.cuda42
acemd.linux64.2352
libcudart.so.4.2.9
libcufft.so.4.2.9

Which should be what I need for the app info.

Also the Windows system with a GTX670 beat the linux system with a 660ti by 1000 sec. I know it had a better card but with the old app linux would have still dominated. Very good gains in windows with the new app!:cool:

Edit: had some wrong file names, I need glasses:o