The official CUDA2.0 is out. We are going to update the compilers when the driver installed by Linux installers (as yum for Fedora) are higher than 177.67.

gdf

The 177.35 driver for win is CUDA 2.0 beta2, so this should be only a minor update? Is 177.35 still the recommended driver for win?

MrS
____________
Scanning for our furry friends since Jan 2002

Things looking good:

- I couldn't install 177.35 on my 9800GTX+, it said "no compatible hardware found"
- the .inf-modified driver did install and can be found here:
http://forums.hardwarelogic.com/f13/nvidia-forceware-177-35-a-14176.html

- now it works for me! Well, too early to shout "success", but it did get past the first minute, which was not possible with 177.79
(see http://www.gpugrid.net/forum_thread.php?id=304 )

MrS
____________
Scanning for our furry friends since Jan 2002

So, we have found out several issues with the drivers and anomalies between cuda1.1 and recent drivers.

We were using driver 173.x thanks to the nvidia-installer --update which does not seem to update to anything recent. Our machines were all on old drivers which worked quite well with the cuda1.1 toolkit which was used to compile. All of you were on newer drivers. Our fault, we did not realize that nvidia-installer does not work as expected.

Although, they had told us that from 1.1 to 2.0 there would be back compatibility this is not the case. This was the problem with the Windows 64 boinc client.

Upgrading our machines to the latest drivers we could finally reproduce the errors, which means that we can now solve them.

In a way it seems that driver-toolkit version need to match to work properly, but there is not specification of which driver work with which compiler.

NEXT

I am expecting that everything will be much better with cuda2.0.
Finally cuda2.0 is out, so we will upgrade as soon as possible to the 2.0 toolkit and will test on the newest drivers.

gdf

That's not exactly what he's saying. I'd imagine it's more a matter of how the different compilers tolerate inaccuracies in the syntax etc. And of course bugs in the compilers :p

MrS
____________
Scanning for our furry friends since Jan 2002

That's not exactly what he's saying. I'd imagine it's more a matter of how the different compilers tolerate inaccuracies in the syntax etc. And of course bugs in the compilers :p

MrS

here with cuda 2.0 177.84 drivers and boinc 6.3.8 on WinXP 32 bits

PS3grid cuda 6.39 app crashes for all WUs :(

For me the 177.35 work fine with app 6.30, BOINC 6.3.9 and Win XP 32 - maybe you should give them a try instead of the 177.84?

MrS
____________
Scanning for our furry friends since Jan 2002

If all the "older" 88xxer GPUs who only have CUDA 1.0 support cannot crunch any more,is it possible to send two different applications? one for the older cards and the other for the rest?
It schould be easily done with an appl.info.xml file in the ps3grid directory.
There are a lot of the 88xxer cards around who could not participate otherwise.

If all the "older" 88xxer GPUs who only have CUDA 1.0 support cannot crunch any more,is it possible to send two different applications? one for the older cards and the other for the rest?
It schould be easily done with an appl.info.xml file in the ps3grid directory.
There are a lot of the 88xxer cards around who could not participate otherwise.

The problem is rather more serious. The code for 1.0 does not work (it crashes) when compiled with CUDA2. So at the moment, we do not have an application 1.0 compatible.

gdf

ok sorry, i missed that.
Does this means that if my cards that are running fine at the moment they will not have troubles with coming/new applikations?

Hi GDF

When do you think to release the new application based on CUDA 2.0 for windows?

Thanks

Thanks for the 6.40 application

Which driver should we use? the unit crashed after one minute with 177.83 and 177.84 (the latest driver available on http://www.nvidia.com/object/cuda_get.html )

My host http://www.ps3grid.net/results.php?hostid=6854

Just noticed that I have a 6.40 app workunit waiting to run.
Should I just abort it if it's causing problems?

Just noticed that I have a 6.40 app workunit waiting to run.
Should I just abort it if it's causing problems?

yes

Umm I can't abort it in Boinc 6.3.10.
Everytime I select the task and click abort, Boinc asks me if I'm sure if I want to abort a PrimeGrid workunit!

Infact no matter what task I pick, it gives the name of another task....