Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

Hot Chips 36 (Die Shrink Telegrams) (Patreon)

Published:
2024-08-27 22:54:24
Imported:
2026-06
Tags:

Content

And that's a wrap on Hot Chips! Just like last year, KarbinCry and Tom will be doing an episode discussing all of the revelations divulged at Hot Chips! There were presentations involving the Zen 5 Architecture, Lunar Lake, Blackwell, and many more interesting subjects!

Please note that the presentations for the show can be viewed and downloaded at the link below, and you have ~24 hours to submit your questions! Don't be shy and don't hold back - we need your thoughts to guide the conversation!

https://onedrive.live.com/?authkey=%21ADGakoHYq6iU8xE&id=71E5E8DA21CB2FFE%21706585&cid=71E5E8DA21CB2FFE

Comments

Sakosha

Why is IBM still making CPUs? And what architecture do they use?

Sakosha

How are in-house chips of Meta (MTIA) and Microsoft (Maia) doing? Why aren't they using top HBM3? What are the drawback of these ASICs not using HBM3(e) like Hopper and MI300?

Manolis Ragkousis

Do any hardware manufacturers propose/push RISCV as an alternative to x86 and ARM for consumer electronics (laptops, phones, etc)?

Kiln God

I noticed AMD, Intel, Qualcomm gave talks on their latest cores as well as their AI accelerators. Was there anything of interest for gamers or developers for the usual suspects? Of the AI crowd Tenstorrent facinates me the most, was there any new wow moments from the AI crowd?

Cleansweep

Do you think we might see the water jet version of Frore's cooler head over to the consumer market? Even if it doesn't hit the projected maximum of 2200w of thermal dissipation, it seems like it could compete with/displace current AIO designs if it's rugged and reliable.

AngelOfCake

My first time reading one of these. I like slide 26 of the "CPU-24-AMD" one. > The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions, and typographical errors. May as well include a "warranty void at point of purchase" sticker while they're at it. Question if you have time: Any thoughts on the veracity of the "Up to 2x Perf/W improvement over prior generation (8040 Series)" claim for Strix? I've been running a x86 handheld and tablet(surface) for about two years now. Better battery life would be a compelling reason to become excited. I'm probably misunderstanding that though. Thanks for great work and keep it up!

Chris Rijk

Hot Chips isn’t really for learning much about designs that are already very public but what were your main take-aways this year from both the big players and small startups? Did it feel much different to recent years? What was the most interesting development or technology?

Chris Rijk

How would you assess the state of co-packaged optics? Do you think it could be common in new HPC systems within a few years, whether AI or traditional?

Chris Rijk

Nearly all the new architectures are using RISC-V. Since they are all high end and expensive designs, I’m sure they could have afforded an ARM licence. Do you think time to market was the main reason? It would be ironic if so since I would expect most customers of AI hardware to go with established providers for time to market reasons as well.

Brad Medlin

When do you think CXL makes the move to mainstream? it would be cool to see some of that tech come (if we got 16 lanes of it and 16 lanes of PCIe 5.0 could make for unique storage opportunities). Also do you see CDNA making a real play? RDNA is taking market share as a lower cost grab. But CDNA is verry different in my understanding than RDNA.

Crast

Last year, everything about Hot Chips seemed to center around "CXL all the things" This year, it seems to be more "new CPU arch" centric. Why do you think this conversation has shifted. Do you think the CXL rollout is not going well, or is it just old news now?

Ronigan

Would it be cost effective, or even likely for AMD to create a half-decent RDNA 4 budget GPU this time around? Maybe something with 24-28 CU's, 8-12 gigs of ram, and a 96-128 bit memory bus for around 150-170$? The last time they made a half-decent budget card was the 5500xt.

Kiln God

Appearently Tenstorrent hardware can be used as AI, Switch or CPU eliminating the need for separate components in the data center, is this true?