Kirin 9050 apparently has 1.5 micron pitch HBI die-to-wafer.
I spoke to a US packaging company several weeks ago with standard high performance HBI tools, saying they're beyond 0.75 micron pitch. It's just a case of manufacturing at scale. They're also seeing scaling better on wafer-to-wafer with die-to-wafer having hard limits.
I reiterate, Tau Scaling isn't new. It's a repackaging of what the industry has always done. If you want to get sucked up into the marketing hype, go ahead, but it's still not new. Too many people get sucked up in the marketing hype. It's about the physics.
I reported on @Broadcom's 2028 XPU a couple of weeks back, but I'd missed one of their key presentations on how to get the packaging to work. If they're going to double revenue each year for the next two years, it's going to have to work.
Big day today. Our interview with Richard Ho (@rh00), the VP Hardware at OpenAI, is now live!
We discuss the new chip, chip design, how OpenAI views the ecosystem, and what's coming down the pipe. Very insightful, thanks Richard!
Here at the TSMC OIP conference and keynote. No pictures allowed, but I'll try and tweet the stuff :) #thread#
First up: the semicon market is $1.7t, years earlier than expected.
I sat down with Cat a week before Hot Chips for a couple of hours to talk about life, liberty, and the pursuit of Dylan Patel. The article came out really well, congrats!
I sat down with @dylan522p and spoke with over a dozen people about his information machine.
"I don't think anyone who has a traditional equity research background would even try to figure out where $3T in [the AI labs' combined] ARR comes from."
Slack private messaging limits, finding CoWoS volume, and the industry they're looking at next - I wrote about SemiAnalysis here:
Well here's a 1 in 10000 output from my Claude Code. Apparently the character is used to signify 'genuine' and when I asked Claude about it, it said that it likely happened due to
(a) close proximity language embedding
(b) a low precision model
(c) speculative decoding.
I've not been tweeting much from AI Infra Summit as I've had my head down with the keynote and panels and such, but here's my walkabout on the show floor that we're now calling The Silicon Notebook from day 2.
Hey @CTATech@CES, I know you don't want randoms signing up, but YouTube is a valid media outlet alongside written media. It's not a social media profile in the sense you're using the phrase. Something might be up with your regex.
"Hi techtechpotato,
I caught your recent breakdown on why AMD acquired Taalas and the technical nuances of that fixed-function inference approach. Your analysis on the token-per-second metrics compared to the H200 was incredibly detailed.
I'm with [company]. They make the M6 Ultra 2.0, an ergonomic chair with cooling and massage features designed to support those who spend long hours in deep technical workflows and editing."
lol wut M6 Ultra
Also can I say that the @AnthropicAI@claudeai chat search function is complete arse. Sometimes I want to sort my search by date, or disable the fact you're assuming what I want to search for rather than actually what I typed. Or I want to search titles only.