Yesterday, it became clear that my program for reading NVIDIA graphics cards is very helpful. I was able to find a problem immediately.

#NVIDIA #GraphicCard #NVML #CUDART #VideoCard #monitor #GPU

Some modifications of my tool makes it more handy. I started also to implement better evaluation of API return values.

#GPU #NVIDIA #graphiccard #NVIDIA #c++ #programming #terminal #ncurses #Linux #NVML #CUDART

The GPU tool I use all the time can be found at Copus. This is my first productive version. Open questions have been clarified as far as possible.

www.copus.io/work/9af8d41...

#GPU #GraphicCard #NVIDIA #NVML #CUDART #Monitor

I noticed an unusual effect. I can use the same GPU 1 for AI image generation and crypto mining. I didn't realize that was possible.

#GPU #GraphicCard #NVIDIA #NVML #CUDART #Monitor

Yeah, in this case I was calling functions in cudart and in OpenGL libs that are not exposed by ILGPU or OpenTK. C# can call any library. And you can do memory management on the CPU side with C# too, if you really really want to.

This is an example, where I used in principle GPU one for calculation and GPU four for the display. GPU three is considerable only used a bit. Look at the memory usage.

#GPU #NVIDIA #graphiccard #NVIDIA #c++ #programming #terminal #ncurses #Linux #NVML #CUDART

sd3.5 works with rtx instead of pytorch. maybe i should stop using pytorch & just learn wtf the deal with cudart is, might even work on my laptop that torch deprecated