The RAM Crunch Could Kill Products and Even Entire Companies, Memory Exec Admits (theverge.com)
- Reference: 0180827004
- News link: https://it.slashdot.org/story/26/02/19/1514236/the-ram-crunch-could-kill-products-and-even-entire-companies-memory-exec-admits
- Source link: https://www.theverge.com/tech/881062/ram-shortage-kill-products-companies-phison-ceo-interview
The interview, conducted in Chinese by Ningguan Chen of Taiwanese broadcaster Next TV, drew an important distinction: it was the interviewer who raised the possibility of shutdowns and product discontinuations, and Khein-Seng largely agreed rather than volunteering the prediction himself. The shortage stems from AI data centers consuming the vast majority of the world's memory supply, a buildout that has sent RAM prices up by three to six times over the past several months. Only three companies control 93% of the global DRAM market, and all three have chosen to prioritize profits over rapid capacity expansion. Even Nvidia may skip shipping a gaming GPU for the first time in 30 years, and Apple could struggle to secure enough chips. Khein-Seng also expects consumers will increasingly repair broken products rather than replace them.
[1] https://www.theverge.com/tech/881062/ram-shortage-kill-products-companies-phison-ceo-interview
We never learn. (Score:5, Interesting)
Consolidation is not an intelligent path long term. Yet, somehow, we're allowing a singular industry trend to capture not just the speculative market, but entire giant segments of the manufacturing sphere, as well as starting to make preparations to allow it to allow it to capture other resources, like electrical production, access to fresh water, and, of course, the all important tax dollar subsidies that all big business actually runs on.
When you pull back from this, and look at it from afar, what it looks like is an attempt to clamp down and maintain a hold of an entire society via technological means. We're putting all our eggs in one basket, and potentially limitless profit generating basket for a very small number of people, at the expense of all the rest of us. It's already consolidating the data of people, of books, of music, of movies, all data. And it seems determined to consolidate the rest of humanity's available resources. And when that consolidation is complete, will there be anything left for the rest of us? And even if there is, what happens if/when that one, singular entity that we have given all power, all resources, all data, and all focus suddenly breaks, or loses momentum? Do we just shrug and standby watching as our world falls into the technologically driven blackhole we've created?
We, the collective we, are being absolute idiots about this whole AI/LLM thing. We've allowed it to subsume too much already, and it seems all world leaders are determined to keep throwing resources at it. "We must or someone else will." It seems stupid to continue down this path, but no one with the power to top it or even slow it a bit and consider the consequences, has any interest in doing anything other than continuing to accelerate the consolidation. It's like the greed of the elites manifested in a completely carcinogenic and caustic manner, and it will not be stopped until it has metastasized and subsumed the entirety of human society.
Re: (Score:2)
Butlerian Jihad?
Re: (Score:1)
It wasn't the "we" that is causing the problem. The issue is every billionaire out there wants their own AI, so they are willing to outspend every other billionaire in building out their data centers and training their own LLMs. In the end, we will end up with a very small handful of successful ones and the rest, just wasted billions of dollars and eWaste because their egos told them they are smarter than the rest of the billionaires and they are the only ones capable of succeeding.
Re: (Score:2)
> We, the collective we, are being absolute idiots about this whole AI/LLM thing.
Indeed. It is both fascinating and really embarrassing to watch. At the same time, the best figures on "efficiency gain" are currently at something like 4%. That is likely way below the measurement error, i.e. there may not be any gains at all and the whole thing is a loss.
This means prices go up even more (Score:2)
Because we will lose some of the few competitors we have left.
If you want to be able to afford things in the future you need to start giving serious thought to politicians that enforce antitrust law.
windows 11 (Score:3)
terrible time to require a new pc for windows 11 for perfectly fine working PCs that run windows 10
Re: (Score:2)
And on top of that, Win11 runs much, much worse than Win10 with bizarre problems and unreliabilities on hardware that was perfectly fine with Win10.
Re: (Score:1)
If only a free alternative existed that ran better on that "obsolete hardware" than the new devices nobody can afford.
Microsoft could exit the way they arrived: on the availability of the hardware they require.
Great chance for new business (Score:3)
Start a business making RAM
Re: (Score:3)
Yes, it's that simple.
Re: Great chance for new business (Score:4, Insightful)
Sounds good, until the AI hyperscale market collapses in a year or two. Then RAM (and storage) will sell for ten cents on the dollar. Just as your new DRAM foundry rolls it's first wafers off the assembly line.
Re: Great chance for new business (Score:2)
Two years to roll out a fab for ddr5?
That is unbelievable optimistic.
Re: (Score:2)
Someone will probably say "something something AI" to say we could build out faster.
Re: Great chance for new business (Score:2)
About 40 months in the US and EU, more like 20 months in China and Taiwan. Somewhere inbetween for S. Korea.
Successfully producing top tier DRAM with high enough yield to be profitable is a little harder to pin down. If you hired a bunch of people that walked out of Samsung and SK Hynix maybe 5 months. Or if you do things from the ground up, 5 years. Doing a previous generation and then catching up on your next foundry build out is perhaps safer.
Re: (Score:2)
No. Not even with retooling an existing fab that already makes something else with the same structure density.
Re: (Score:2)
Even more unbelievable that somebody will lend you the money for it, because FABs on that level are extremely expensive.
Re: (Score:2)
Seems the good mid-term play is to develop a motherboard that can take something obscene like 512GB of DRAM to roll out about the time that happens.
Re: (Score:2)
Consumer DDR5 platforms have a hard time using more than a pair of DDR5 modules at any but the slowest timing and currently don't support DIMMs larger than 64GB. Workstation and Server Platforms can already support more RAM than that, but if you're buying a new enough Threadripper, Epyc, Xeon or Ampere platform to handle DDR5, you're almost certainly buying it with rDIMMs in the first place.
Re: (Score:2)
Further, even if the market doesn't "collapse", it would still transition to something more like a "sustaining" market rather than ever expanding buildout.
Very slow and expensive to try to solve what will, one way or another, be a temporary problem.
Re: (Score:2)
I agree that it is slow, expensive, and ultimately, a temporary problem.
My question is, how long will it last? When will demand for RAM decrease or when will RAM production capabilities catch up with demand?
I don't have a clue. One could make an argument for a crash in demand this year or five years from now.. Too many unknowns.
Re: (Score:2)
And that is why the RAM manufacturers will do no or only very careful extension of production volumes.