Earlier quoted context omitted.
[flagged]
What's ChiComs?
Chinese Communists, i.e. "Red Chinese". The people in control of the mainland after the nationalists got kicked out to Taiwan.
TSMC warns AI chip crunch will last another 18 months
51–60 of 107 posts
Re: TSMC warns AI chip crunch will last another 18 months
#52I'd bet $50 in 18 months "the crunch" will still be in full swing (and Raspberry Pi will still be "bringing huge volumes to the market any week now"). There is no way in hell all these expensive fabs will pay for themselves if chips are too cheap. So no one (especially not TSMC) will risk oversupply by building too many fabs.
> So no one (especially not TSMC) will risk oversupply by building too many fabs.
Everyone and their dog is investing like crazy. Intel and TSMC are both setting up fabs in Silicon Saxony (although the question remains on how productive these will be, given the issues with the far-right in Saxony), TSMC is also expanding capacity in Arizona, and Samsung is building one in Texas.
There will be no oversupply, not for the next few years. Demand from automotive, hyperscalers, AI and pent-up consumer demand is ludicrous.
Re: TSMC warns AI chip crunch will last another 18 months
#53Earlier quoted context omitted.
They're selling consumer GPUs with huge margins too. The markup on the 40xx series is insane but they're still shipping a ton of units.
is that the case? They sold a bunch early on but theres been lots of press about the lousy 40 series sales volume this year and their struggle to get people buy 4080/70's. In Feb they artifically constrained 4090 volume to try and push 4080s no one wanted. The 4070 was dead on arrival and had production cut by April. They've near stopped production now - theoretically because the AI stuff is so lucrative, but if they…
I blame it on Nvidia's product segmentation falling apart. Only a few years ago, the 1060, 1070, 1080, and Titan Xp had price differences that matched their performance differences. In their current lineup, buying anything that isn't a 4090 is getting ripped off. The 4060 and 4070 are basically e-waste. The 4090 had a 100% performance improvement in a lot of tasks over its predecessor, and the 4060 managed 10%. Sometimes. I don't know why they bothered making them--it's no wonder they cut production.
Re: TSMC warns AI chip crunch will last another 18 months
#54Earlier quoted context omitted.
The Raspberry Pi wouldn't have issues if not for the fact that they are explicitly an outlet for Broadcom's excess chips, and they have never paid full price for their CPUs. If they doubled the price of the boards, they would be able to get the parts.
Why don't they? The current market price is 3 or 4 times the MSRP and has been for years--the Raspberry Pi has lost its place as the affordable hobbyist board. I suppose it would be admitting defeat, but if they'd actually pay for parts and make new Pis again, I'm sure they could at least bring the market prices down.
The price in my market is back to normal for the versions I want. Has been for a month or so now. I don't see why the other versions won't follow soon enough, exactly as was promised at the start of the year.
Re: TSMC warns AI chip crunch will last another 18 months
#55Earlier quoted context omitted.
I'm not saying it's not useful, but much like the internet, claims have been overstated.
LLMs/transformers are a breakthrough on the level of convolutional neural networks — significant, and one that opens up lots of new interesting applications as well as invites lots more R&D — but like the neural net it’s not gonna fundamentally transform society or industry. (And as far as I can tell this current craze is entirely predicated on LLMs/transformers.)
llm/transformers are 5 years old already.
llm companies raise lots of money on crazy evaluation, the question is if they will be able to build enough revenue stream to justify further money injections.
Re: TSMC warns AI chip crunch will last another 18 months
#56I'd bet $50 in 18 months "the crunch" will still be in full swing (and Raspberry Pi will still be "bringing huge volumes to the market any week now"). There is no way in hell all these expensive fabs will pay for themselves if chips are too cheap. So no one (especially not TSMC) will risk oversupply by building too many fabs.
Increasing supply is not an issue if there is demand. As long as building more fabs brings more profit(even if at smaller margin), I can't see any reason why they won't be building it. It's economics 101. Also as moore's law is coming to end, the number of fabs has no way to go but up.
Also I don't think manufacturing cost is the biggest of issue, because the margins are significantly higher than consumer GPUs which are in oversupply now for Nvidia. It's that Nvidia has to commit years in advance for the capacity.
Re: TSMC warns AI chip crunch will last another 18 months
#57Note that the shortage now is different than the past few years: it is not in fabricating any of the primary chips, it is entirely attributed to insufficient capacity in TSMC’s CoWoS packaging. This is only used for the highest-end accelerator products that use HBM (and a few other things that use full silicon interposers like Apple’s M-Ultras), they are scaling capacity, and there are increasing numbers of alternati…
Exactly. This is a different stage of the pipeline.
I can't wait for copackaged optics, which will have a very similar manufacturing pipeline issue, to turn the NPU manufacturing pipelines into hell, that's surely coming and will stall performance networking for a solid year or two if CPO takes off.
Re: TSMC warns AI chip crunch will last another 18 months
#58Earlier quoted context omitted.
That would be foolish. A large part of their moat is that lots of people already have Nvidia cards in their computers, making development really smooth and allowing hobbyists to easily get started and contribute to the ecosystem. If they were to give up their consumer gpu lead and let AMD overtake them that will lead to more AMD support in ML, enabling AMD to eventually compete in the server gpu space too.
> allowing hobbyists to easily get started and contribute to the ecosystem I think many, maybe even most, would claim this is now best done on cloud services, like huggingface.co.
I think that's still a minority. Everyone has heard of cloud platforms giving people huge surprise bills and other such limitations. Another factor is availability and potential T&S concerns when you're doing something they don't want you to do.
Stable Diffusion blew up when people are able to run it at home.
Re: TSMC warns AI chip crunch will last another 18 months
#59Earlier quoted context omitted.
I'm not saying it's not useful, but much like the internet, claims have been overstated.
LLMs/transformers are a breakthrough on the level of convolutional neural networks — significant, and one that opens up lots of new interesting applications as well as invites lots more R&D — but like the neural net it’s not gonna fundamentally transform society or industry. (And as far as I can tell this current craze is entirely predicated on LLMs/transformers.)
Autonomous cars alone will be transformative. And LLMs are not in their end-state, we went from GPT-1 to 4 in 5 years over that timespan it gained lots of new capabilities. Extrapolate, don't just look at what they can do today.
Atoms are more difficult than bits, so physical applications (e.g. in factories, robotics) will lag. I also expect specialist models like AlphaFold slowly finding their niches over time.
> significant, and one that opens up lots of new interesting applications as well as invites lots more R&D — but like the neural net it’s not gonna fundamentally transform society or industry.
Yeah, and semiconductor electronics that grad-students built from of germanium by painstakingly pressing them together are very interesting and worth additional R&D but won't transform society either.
Re: TSMC warns AI chip crunch will last another 18 months
#60I'd bet $50 in 18 months "the crunch" will still be in full swing (and Raspberry Pi will still be "bringing huge volumes to the market any week now"). There is no way in hell all these expensive fabs will pay for themselves if chips are too cheap. So no one (especially not TSMC) will risk oversupply by building too many fabs.
Pi availability has been improving steadily since the beginning of the year.