it seems to be that they've thrown out a lot of what the GPU does and went back to some old school shit, treating the gpu cores are just "hey lol we have 800 CPUs from the year 2000" and going from there.
they're literally doing software rendering code on the GPU and just ignoring the triangle fill hardware
🇳🇱
@icedquinn Actually you cant treat it like a 800 CPU's due to one very vital difference between a GPU and a CPU that is far more restrictive than that, its called SIMP/SIMT/SIMD (Single Instruction Multiple Data). In short when using a GPU as a processing cluster all the processors in the GPU must be running the same instruction set at the same time but each one can process different data.
In addition you cant have excessive branching in your GPU instructions or else the GPUs will be uselessly slow. So you cant just take each code block and split it up with a bunch of "if" either.
@icedquinn yea im not trying to criticize the presentation. Mostly just talking about why you cant (and probably never will be able to) think of a GPU as just 800 cpu cores from 2000. I really wish you could since I do OpenCL stuff all the time and it can be rather limiting.
doesn't usually get fielded in production tho.