Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
crest
on March 5, 2025
|
parent
|
context
|
favorite
| on:
Apple M3 Ultra
Too bad it lacks even the streaming mode SVE2 found in M4 cores. If only Apple would provide a full SVE2 implementation to put pressure on ARM to make it non-optional so AArch64 isn't effectively restricted to NEON for SIMD.
vlovich123
on March 5, 2025
|
next
[–]
This is for AI which is going to benefit more from use of metal / NPU than SIMD.
bigyabai
on March 5, 2025
|
parent
|
next
[–]
Sure, but larger models that fit in that 512gb memory are going to take a long time to tokenize/detokenize without hardware-accelerated BLAS.
microtonal
on March 5, 2025
|
root
|
parent
|
next
[–]
Why would you need BLAS for tokenization/detokenization? Pretty much everyone still uses BBPE which amounts to iteratively applying merges.
(Maybe I'm missing something here.)
ryao
on March 5, 2025
|
root
|
parent
|
prev
|
next
[–]
Tokenization/detokenization does not use BLAS.
stouset
on March 5, 2025
|
prev
[–]
Hell I’m just sitting here hoping the future M5 adopts SVE. Not even SVE2.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
