Digital Light & Color

Hi, can anyone out there explain how CUDA is supposed to work? I just installed a new video card, the GeForce GT 710, with a GK208 GPU which supports CUDA with 192 cores. I then also uninstalled and re-installed Alpha 9-Nov-2018. But the "About" window shows "CUDA not available", and when I "Toggle CUDA support", nothing appears to happen. I'm not at all familiar with this stuff, so I don't know if there's some other step I've missed.

My old card dies, so I wanted mostly to get a card to support dual monitors, and I must say, I very much like being able to keep the transformation windows and image browser on the 2nd screen.

Any help with CUDA would be greatly appreciated.

It sounds like you need to download the latest drivers from Nvidia.com. Don't use the Help command to toggle CUDA as that will turn it off.

Of course, thanks for reminding me Jonathan, that solved it. Are there situations where one might want to toggle CUDA support "off"?

Not really -- It's there mostly for debugging and for timing the speed difference.

Oh, okay, thanks for the info. And thanks for the support of dual monitors, I'm really liking the un-cluttered view of the input and preview images.

It turns out there is another reason for turning off CUDA -- if you are using an older or low-end display adapter lacking enough free memory to hold the entire image you are operating on, you will get an error. In this case you can proceed if you turn off CUDA and process the image normally using the CPU instead of the GPU.

Good to know. so, my "new" graphics card has 1GB of memory, can you suggest what's the maximum image size that will support?

I ran into problems doing a Bilateral Sharpen on a 48-bit color image from my Sony A7RIII which has 251MB files. The card needs to hold 2 copies of the file - one for the input image and one for the output image, so that's about 0.5 GB. I was getting errors with a 2GB card. It works fine with the 8GB card in my gaming laptop and other desktop computer. It seems like the image should fit in 2GB, but I don't know how much memory is left after Windows drivers and display buffers are allocated.

Thanks for the info. 48-bit files form my D810 weigh in about 215MB, so I'd probably see the same error with only 1GB of GPU memory. Now I can plan for the desktop upgrade...

I had been assuming images would fit in GPU memory - since this is not necessarily the case I need to look into processing them in sections that do fit. I am still just getting started with CUDA and there is a serious learning curve.