Re: Is Intel exceptionally unsuccessful as an architecture designer?

In article <vcisaf$ulcv$1@dont-email.me>, ldo@nz.invalid (Lawrence
D'Oliveiro) wrote:

On Fri, 20 Sep 2024 00:58:44 +0000, MitchAlsup1 wrote:

Hint:: They can context switch every instruction.

How does that help?

All the threads are executing exactly the same instructions,on the same
code path. If any of them start taking different branches, performance
goes way down, because then they can't amortise the time for the
instruction fetch across all the threads.

The context switches don't involve any memory accesses. The GPU processor
has a set of registers for each thread, and a context switch is just a
change of which registers it's looking at. It's the same trick as the old
TI 990 architecture. I was quite amused when I figured that out in the
middle of the first presentation from Nvidia I ever sat through.

<https://en.wikipedia.org/wiki/TI-990>

John