Since when was cooperative launch now supported in Windows non TCC mode?

Accelerated Computing CUDA CUDA Programming and Performance
1

I was surprised that when I query device props that suddenly devprops.cooperativeLaunch is now seemingly supported on my Windows 10 PC that has an RTX4090 as its sole GPU that is connected to the displays. I thought this was only supported in Windows if your device was not connected to a display and in TCC mode?

When did this change and under what circumstances? Does this mean cooperative launches are safe to use on all consumer Window machines now? Is it limited to specific GPU’s? Does it work on RTX 3060, RTX 4060 or what configurations? Are there limitations to be aware of? It seems a strange change to have quietly happened???

2

Qualifying conditions are outlined at the end of 8.7.

3

Ok your link says:

Only devices with compute capability of 6.0 and higher are supported. In addition, you need to be running on either of these:

  • The Linux platform without MPS
  • The Linux platform with MPS and on a device with compute capability 7.0 or higher
  • The latest Windows platform

So the Windows TCC requirement was dropped without any obvious announcement! And that means a lot of information at places like this are now incorrect:

Would have been nice to have this communicated better! As that’s a useful feature to have working across a wider range of RTX3000/RTX4000 series consumer GPU’s.

4

If it helps anyone, we tested across our office PC’s and this seems to now work across laptop RTX3070, and desktop RTX4070, RTX4080, RTX 4090 GPUs. I guess that was a sneaky stealth update by nVidia the past year!