Working with my new RTX 3090 has been one of the biggest headaches in my life. I have never seen such a janky collection of rubbish foisted on us as ‘drivers’ and as ‘solutions’. I do not feel like I purchased a premium card at $2500. Instead I feel like I am struggling with a nickel and dime operation. I have debian bullseye. Sorry, I had debian bullseye. In the endeavor of installing the nvidia super-duper ‘democratizing’ cuda core drivers I am now sitting on debian bookworm-sid unstable … … … … … . . . … and tensorrt still does not work. I have juggled the main driver. cuda. cudnn. python. tensorflow. pycuda. cudapy. cudpy. pycunnndd. duncpul. jenpycud. cudpyjen. sonpycunjenn. what is going on … ? These blogs are also chock full of a million topics with all the appearance of activity and just running around in circles for all intent and purpose. My card meanwhile is toasting its vram on the backplate, where i might as well be frying eggs and bacon for breakfast with jensen in his kitchen. How can nvidia do such an awful job of driver version management, while their standalone gpu drivers are so seamless on windows. i am … in total disbelief. I had read a lot of comments online about the horror of the new nvidia drivers, but oh, my god. What I am experiencing is far worse than anything I could have imagined. I have never before in my life wanted to pull out a gpu and smash it against a wall, but i do right now. How can you charge premium pricing for such trash support? nothing is organized, just like this post, everything is back and forth like a panicked crazed lunatic. i am so confused. i don’t know, honestly, what to do about this card, and if it is worth keeping, or to get the new macbook max and just support apple’s more level-headed, coordinated and premium method. I am so disappointed with nvidia, i have been a loyal supporter for two decades. But this. oh me oh my. I have a lot to think about right now. and no, docker is not a solution, it is a pathetic band aid of a solution with a massive dab of desperation. a lot to think about. a lot.
Hello, this forum is dedicated to discussions related to using the cuda-memcheck tools.
Questions related to CUDA can be raised at CUDA - NVIDIA Developer Forums