hobel (do.delete@this.not.reply) on February 25, 2019 1:10 am wrote:
> end of an era (eternity0-9.delete@this.hotmail.com) on February 24, 2019 2:18 pm wrote:
>
> > Listen Linus, Nvidia's CUDA architecture is pretty successful on many different applications, either
> > you accept it or just stop talking, cpus will follow the same step, take a look at threadripper, it is
> > a really successful cpu. my advice to you; take a longer vacation, retire maybe. your time is over.
>
>
> show me that gcc running on CUDA with speedup.
>
> No one doubts that special hardware can do special things well,
> bit we are talking 128 server cores here for general workloads.
>
>

gcc is actually damn interesting case.
I have a strong feeling (not my field at all, so just feeling) that the same deep learning techniques that in last couple of years produced excellent results in go and chess* have big potential in compiler optimization.
Unfortunately, gcc code base is probably too convoluted to incorparate something that different quickly.

-------
* - BTW, yesterday Leela Zero narrowly lost to Stockfish in TCEC Season 14 Superfinal (!). Who would believe two years ago that such quick progress is possible in such short time? May be, Gian-Carlo Pascutto would, but certainly not me.