This is extremely different, apple was targeting end consumers that just want their app to run. The performance between apple rosetta and native cpu were still multiple times different.
People writing CUDA apps don't just want stuff to run, performance is an extremely important factor else they would target CPUs which are easier to program for.
From their readme:
> On Server GPUs, ZLUDA can compile CUDA GPU code to run in one of two modes:
> Fast mode, which is faster, but can make exotic (but correct) GPU code hang.
> Slow mode, which should make GPU code more stable, but can prevent some applications from running on ZLUDA.
The too chatty protocol is bitswap. Because the merkle-tree is not coupled to the protocol, they developped their own custom protocol on top of UDP for communication.
I went through my history, for one code example I copy pasted the old prompt into new GPT4.
It was about writing a CGO wrapper given an hpp header for a lib I have.
Back then it used to give me almost correct code, it understood is had to write a C++ C ffi using extern "C" first because Golang's CGO FFI only support C not C++. And then it generated me correct CGO wrapper with a Go looking type that made sense.
The only wrong thing is that it didn't understood it had to call the initialization function of my C++ lib at init time, instead it called it in the New factory function (which would segfault when you build more than one object) trivial fix for the human in the loop move it to func init().
TL;DR back then almost perfect
Now with the exact same prompt it doesn't even generate me code, it just list me a list of tasks I could do to achieve this task and give me vague statements. If I change the prompt insisting for code, it instead give me a very dumb 1 to 1 mapping of the C++ into Go trying to create constructor and destructor functions, oh and it's tries to use CGO ffi to call C++ (even tho again, only C is supported by golang).