Now the curious thing is I saw are also "integer units" which also have lanes, but there was no mention of AVX there. I wonder if there's parallel instructions supported there too?
Learned the other day how AVX-512 instructions are implemented at the chip level. Chips have "floating point units", which have lanes going to and from them. Each lane handles 1 instruction/cycle, and there's 512 lanes! So 512 ops in parallel = AVX. Pretty cool!
-
-
Show this thread
-
Found the article I was reading. Has some really neat schematics in it! https://fuse.wikichip.org/news/1815/amd-discloses-initial-zen-2-details/ …pic.twitter.com/StKJgE5VD7
Show this thread
End of conversation
New conversation -
-
-
My understanding is that AVX handles 512 bits per operation, which means that there is less parallel floating point operations: 512/32 = 16 with single precision.
-
Ugh, my understanding might be totally wrong, so never mind.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.