Comments by "Gilad Barlev" (@GSBarlev) on "Analyzing Deepseek's "undefined" NVIDIA PTX optimizations (with benchmarks!)" video.

This takes me back. I wrote the numerical simulations for my PhD in CUDA, back in 2010-13. I don't even remember the API back then giving you per-thread / non-global memory aside from the thread ID. And there were so few GPGPUs out in the wild that the guides and forums for writing well-optimized CUDA code were simply not there. I haven't touched an Nvidia GPU since then, but videos like this definitely make me want to pick one up just to dust off those brain gears.
47
@Iswimandrun Ooh, do you have an FPGA at your disposal? Technically I do with two 7840-based systems, but AFAIK there aren't any Xilinx / XDNA drivers that let you directly manipulate them yet.
2