17
top 9 comments
sorted by: hot top new old
[-] Evilphd666@hexbear.net 15 points 5 hours ago

But it was incorrect at 15,823 tokens per second!

[-] haxboar@hexbear.net 1 points 58 minutes ago
[-] will@piefed.zip 5 points 4 hours ago

Well now what?

[-] invalidusernamelol@hexbear.net 15 points 6 hours ago* (last edited 6 hours ago)

AMD might win the "AI race" with this. Let the other ones fall, then come in with affordable purpose built hardware that's actually useful.

The general service model won't keep working forever, but a chip that's specifically designed to like translate from Spanish to English, or OCR documents, or whatever other specific LLM task you need is a product that can be sold repeatedly.

[-] Evilphd666@hexbear.net 4 points 3 hours ago

On the specific chip, I've seen another format called compute-in-memory from Anker.

So it will be interesting to see what comes out of the AI tech race.

[-] invalidusernamelol@hexbear.net 1 points 1 hour ago

I think the GPU model of "a thing you jam data into and get a thing out" is gonna become more common. Baking a network into a chip is more efficient than fpgas too.

[-] HexReplyBot@hexbear.net 1 points 3 hours ago

I found a YouTube link in your comment. Here are links to the same video on alternative frontends that protect your privacy:

[-] HexReplyBot@hexbear.net 1 points 6 hours ago

I found a YouTube link in your post. Here are links to the same video on alternative frontends that protect your privacy:

this post was submitted on 07 Aug 2026
17 points (100.0% liked)

technology

24439 readers
287 users here now

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

Rules:

founded 6 years ago
MODERATORS