which driver to install


which driver should I install for use with CUDA?
The one from the “normal” nvidia driver page (version 190.42), or the one from the nvidia cuda page (version 190.18beta)?
In the release highlights from 190.42 it says that support for CUDA 2.3 was added. But why is there then still an older version on the CUDA page?

Could anyone explain the difference between the two a little bit more in detail?

Thanks a lot.

By the way, we’re using two GTX280 on CentOS5.

Use the release driver. Often, the release driver features (from a CUDA point of view) lags behind the release of the CUDA toolkit, so NVIDA wind up having to offer beta drivers at the beginning of their CUDA version release cycle (IIRC correctly the release driver was still a 1.85 series when CUDA 2.3 was released). Unfortunately they aren’t very good at keeping the drivers they offer with their CUDA toolkits current. So if you want to run CUDA 2.3, then the current release 190.x driver is the one to use.

Be aware that if you want to run CUDA 3.0 beta, you need the beta 195.x series driver that NVIDIA offer on the CUDA tool page. Just to make things even more confusing…

I already wrote on this forum how awful the driver publishing is, it’s not easy to explain

to the other department people that when installing the servers the driver to use is not

the one proposed when downloading cuda (that is marked still in beta and then they go

ballistic about the fact to install a beta drive) but the one on the ftp nvidia site.

I hope that in the near future this driver/release cuda managment will be fixed.

Pardon me?