TCC 260 beta driver and WIndows2008 HPC Cluster Problems on inital setup of CUDA dev environment on

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi there,
I recently set up a development environment within our windows compute cluster for CUDA.
I was lucky that the NVIDIA TCC driver went out before my test so I could do them comfortably via RDP.

The basic problems where:
No mixing of NVIDIA and AMD cards in the same server (you probably know that, I didn’t).

Still after setting all up and setting the card to TCC mode via nvidia-smi the device query did not manage to communicate with the card.

Finally a missing DLL turned out to be the problem. Installing the Desktop experience feature of Win2008 adds linkinfo.dll and enables device query and development.

Read the whole story on my blog here

2

Hi there,
I recently set up a development environment within our windows compute cluster for CUDA.
I was lucky that the NVIDIA TCC driver went out before my test so I could do them comfortably via RDP.

The basic problems where:
No mixing of NVIDIA and AMD cards in the same server (you probably know that, I didn’t).

Still after setting all up and setting the card to TCC mode via nvidia-smi the device query did not manage to communicate with the card.

Finally a missing DLL turned out to be the problem. Installing the Desktop experience feature of Win2008 adds linkinfo.dll and enables device query and development.

Read the whole story on my blog here

3

Thanks for sharing!

btw, deviceQuery comes in two flavors… One of them is statically-linked one… Thats what I generally use to avoid such issues…
Not sure, if something has changed recently…

4

Thanks for sharing!

btw, deviceQuery comes in two flavors… One of them is statically-linked one… Thats what I generally use to avoid such issues…
Not sure, if something has changed recently…

5

Hi,

My environment actually got broken by the first TCC release driver.
With the newest 263.06 driver and a registry tweak Cuda works via RDP on Windows2008 like a charm.

Just add the following to the registry:

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class{4D36E968-E325-11CE-BFC1-08002BE10318}\0000]
“AdapterType”=dword:00000002

to the registry and as well in all 000X folders corresponding to a CUDA GPU.

Full entry: Blog

Source of this solution is IBM