Einstein speeds with cuda

[uBronan]

I thought lets posts some speed results with cuda on this project

GTS 450 (192 CC) 4950 secs p/u
GT 440 (64 CC ) 12200 secs p/u
einstein really does need more Cuda Cores to perform a bit better

[Bok]

I haven't tried the cuda app there since it was first released and it errored everything out. Can you specify only to run CUDA work ?

I'm actually running a few of the wu's on their test project - Albert@Home now seeing as they reset the credits finally, so I added them to the stats..

[uBronan]

Actually yea just use the app_info.xml
Before you switch to running this way copy the executables to a backup map then move the app_info.xml into the map so in case anything goes wrong you can copy the programs back

Copy and paste this text in a plain text editor and save as app_info.xml

<app_info>
<app>
<name>einsteinbinary_BRP4</name>
<user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
</app>
<file_info>
<name>einsteinbinary_BRP4_1.00_windows_intelx86__BRP3cuda32.exe</name>
<executable/>
</file_info>
<file_info>
<name>db.dev.win.3d35195e</name>
</file_info>
<file_info>
<name>dbhs.dev.win.3d35195e</name>
</file_info>
<file_info>
<name>cudart_xp32_32_16.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft_xp32_32_16.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>einsteinbinary_BRP4</app_name>
<version_num>100</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.330000</avg_ncpus>
<max_ncpus>1.000000</max_ncpus>
<plan_class>BRP4cuda32</plan_class>
<api_version>6.13.0</api_version>
<file_ref>
<file_name>einsteinbinary_BRP4_1.00_windows_intelx86__BRP3cuda32.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart_xp32_32_16.dll</file_name>
<open_name>cudart32_32_16.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>cufft_xp32_32_16.dll</file_name>
<open_name>cufft32_32_16.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>db.dev.win.3d35195e</file_name>
<open_name>db.dev</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>dbhs.dev.win.3d35195e</file_name>
<open_name>dbhs.dev</open_name>
<copy_file/>
</file_ref>
<coproc>
<type>CUDA</type>
<count>0.50000</count>
</coproc>
<gpu_ram>419430400.000000</gpu_ram>
</app_version>
</app_info>

This will do the trick
Just make sure allways to keep a backup elsewhere
yes albert@home is alpha and beta testing purpose if your not scared too loose credits and have crashing projects :D

[Bigred]

You can also just set it in your preferences to only use Cuda without the APP_INFO_XML. I've run a few of them but they are monster downloads (32mb).

[uBronan]

I have had 112 units on my pc and suddenly saw only 8 GB space was left on the disc :D
The most important about running by the app_info.xml is that you can change the units running at the same time, especially on fast cards (460+, 560+) with more the 1 Gb memory you can run 2 to 3 units
As i posted elsewhere its important to keep an eye on memory of the card since all units eat between 300 to 460 Mb each, so as a rule i use 2 units on 1 Gb and 3 if over 1,5 Gb
Now running more then 3 is absolute nonsense just ignore what others say about it just stay withing the 1 to 3 units a core.
Crunch3r told me that in no way any card will do more then 2 threads
Then you say why the 3th well easy if one unit is finished it immediate starts on that unit without having to load it from disc
So when you do not use ssd drives this gain a few msecs
To be honest if you don't have a huge ssd i would not advice to run these on that disc
They are all 32+ Mb as BigRed said so a normal cache full will add up quick.

When running with different cards in the same computer you must check if all can do the same or you must set it to the one who can do the least. For instance i could run 2 units with my gts 450 but it will not work well since my gt440 can not run more then 1 similar tasks since it has only 512 Mb memory