Building one the WW2 designs isn't really that hard today, is it? Getting the fission material on the other hand... Just ask Doc Brown, because it is not the 80s anymore were you get Plutonium at pharmacies, ah those days!
It depends. Little boy would be easiest as the speherical detonation and materials science for the Fat Man implosion would still be difficult to design and machine. I also do not beleive the activator designs have ever been declassified.
Doesn't have to be an exact copy so, but yes design might be tricky. I assume solvable so, at least in a primitive and workable way. Machining, to be honest, beside size of the parts themselves, is something you can do in some reasonably well equipped maker space. And for size, CNC machines are readily available on the used market. And those are light years ahead of anything the Manhatten project had. Or Apollo for that matter (excluding the Saturn V of course, but that is rocket science).
From my undertanding, Fat Man requires a very precise design to focus the shock wave properly, or the bomb will never detonate properly.
This design is quite non-trivial even today. It took several of the brightest minds of all time to get it just right back then (like von Neumann).
Even with modern tools, you still need a team of quite capable engineers/scientists to pull something like that off, even if you do have the plutonium.
The little boy design is easier to engineer, but bomb grade uranium is even harder to produce than plutonium.
Not even a 1000 normal engineers can replace a single von Neumann.
What makes it easier, though, is that the general principle is relatively well known, so one could always brute force the solution using computer simulations, even if unable to do the math by hand.
It wouldn't. HN lives in a SW bubble where Foss is the solution to everything , but fails to see that capital is the main barrier in the semi industry and, unlike their Node library, the secret sauce of cutting edge chips is a matter of national security at this point and will be kept as close to the chest as possible.
Actually if the IC design tools would be FOSS, the cost of new designs would decrease considerably, even up to two times.
The cost of the subscriptions to the design tools provided by Cadence, Synopsis, Siemens (Mentor) and a few others is ridiculously high and it can contribute with many millions of $ to the design cost of a state-of-the-art chip.
The design tools themselves have been improved extremely slowly and they are of a quite low quality for so expensive professional tools. In an open-source environment it would be easy to make much better alternatives.
The reason that only a few timid attempts exist to provide alternatives for only some parts of a complete toolchain, is that the semiconductor manufacturers, like TSMC, do not provide public information about the exact data that they need as input for manufacturing, about the design rules that must be implemented for valid designs and about the characteristics of their processes that must be known for simulations.
All this information is available only to big companies like Synopsis, Cadence etc. and this perpetuates their monopoly and their ability of extracting huge amounts of money with minimal software development work, from all the companies that design ICs.
While the companies that design ICs are not happy for this huge rent that they pay for the design tools, they are also extremely risk-averse and they do not want to make any change in their design workflows, no matter how small, that might increase the risks of failure in a designed chip. Because of this there are no chances of replacing the incumbents with open-source tools, at least not in the big established companies that design chips.
>Actually if the IC design tools would be FOSS, the cost of new designs would decrease considerably, even up to two times.
There are FOSS IC design tools but nothing that can match Synopsis, Mentor and Cadence and the semi industry isn't gonna start dealing with FOSS jank and destroy their productivity and competitiveness just to save a few milion in licensing costs to those 3 companies, which are peanuts in the end.
They're like the John Deer of semi tools, expensive to own and run, but necessary to be competitive and get the most out of this field.
Plus, the proprietary stuff is still the standard-cell libraries of TSMC, Samsung, Glo-Fo etc. They will never open source this.
>and this perpetuates their monopoly and their ability of extracting huge amounts of money with minimal software development work,
Oh, you mean just like Microsoft, Google, Apple, etc?
>At least Microsoft, Google, Apple etc. do not demand (yet) hundreds of thousands of $ per year for a single license.
Because they sell ads and monetize their users instead, which is a pipe of infinite money.
Why does it matter to you how much they're charging? It's not communism. It's their honest labor and they're allowed to charge as much as they want for it, same how you're allowed to ask for as much money as you want for your salary negotiations.
So please excuse traditional engineering companies for trying to make a living by monetizing their hard earned IP like every single company out there, and not selling ads like big-tech.
The world still needs traditional engineering companies to exist to progress, we can't have a society where everyone only works by selling ads to one another.
If you feel like they're price gouging, then feel free to write your own EDA tools and sell them for cheap, I'm sure the market is open to paying much less and saving some money.
Like I have already said, there is no shortage of people who could write better EDA tools, but none of them can obtain the information needed to make them compatible with e.g. TSMC, despite the fact that one would think that it is in the interest of a foundry to provide the information that would be needed to enable the appearance of new customers.
However most foundries, like too many other businesses today, are not interested in new customers, unless they already are very big companies, which is one of the factors that has resulted in the extraordinary reduction in the number of companies that compete in every field, in comparison with the situation from a couple of decades ago.
This is exactly the kind of thing that will never be open source. Any entity with the extremely high level of resources to fabricate chips at this node have zero incentive to open source anything, and no number of amateurs with 3D printers can change that.
Read Chip Wars [1]. The cost to stand up an advanced chip fab has reached tens of billions of dollars. It’s so capital intensive that even the US Department of Defense has exited the game.
And it's not just needing to buy the ASML machine. There's a ton of proprietary knowhow to actually get it integrated into a fab so that yields are reasonable.
Semiconductor manufacturing requires a lot of capital expense just to open the doors, it also requires a staff of all levels to keep the machinery working and multiple suppliers for raw materials that require multi-thousand dollar minimum orders. The process these facilities implement are fairly well understood, it’s just cost prohibitive.
If you’re trying to draw parallels between this is and, say 3D printing, you will not find many similarities. PCB fabrication is also a a different beast but a lot of people think it’s as easy as finding a fab house that takes gerber files and creates a board for them.
If we want something that more closely resembles how we develop software, we probably will need to develop processes that enables us to flash FPGAs like MCUs, make them fully read only, and allow the customer to etch their own marking onto the chip. There facilities that produce ASICs like this already, but I can see something more standardized becoming more successful for general purpose. This does however require that we have a foundry create the original chips, the difference here is that they are producing the same standardized chip for everyone which would allow any foundry to make compatible alternatives.
Note 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 alternatives from other packaging vendors that chip designers could use if they had known TSMC’s packaging would be a bottleneck. Demand spiked a lot all of a sudden, and it just takes time to build factories – there are no obvious constraints.
Given that the nanometer rating is purely marketing/branding and doesn’t translate to actual sizes of anything nor translate between manufacturers, this really doesn’t mean anything.
Chinese 7nm is on regular ultra violet 193nm lithography, anything lower needs ASML or clone and it's 13.5nm lithography. ASML and it's suppliers were working on 13.5nm lithography for 20 years.
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.
I wonder how this will impact consumer gpus? If nvidia can print money making h100's you have to wonder if they will continue to invest on the retail side or if they'll funnel most of their efforts to server and let desktop parts languish.
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 were really "shipping a ton of units" of the 40 series and didn't have excess inventory i doubt they'd have done that.
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.
Yes but it’s hard to disentangle the effect of the crypto crisis ending last year, the pandemic boom in consumer hardware sales coming to a close, and the pricing of Ada from one another. So many GPUs were sold to miners and made their way to the secondary market that gamers have lots and lots of options that aren’t new sales.
Looking at Steam concurrent player count it is continuing to ride a steady escalator upwards to I don't think PC gaming as a hobby is gonna shrink anytime soon. My guess is that in a year or two you'll see a big rebound in sales as all of the pandemic hardware starts to get old.
The consumer GPU margins aren't even close to what Nvidia rakes in on an H100 sale.
Rarely does Nvidia sell a GPU for 2000 dollars or more, and when they do its relatively low volume.
An H100 sells for tens of thousands of dollars and is sold out to 2024 - companies are literally fighting over them - I can absolutely see conditions might arise Nvidia might want to use some fab capacity for their much more profitable H100 parts vs consumer GPUs.
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.
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.
Nah, when you ask around on AI/ML subs, the default advice is to use google collab. Even shelling out for their premium offerings, you could pay for all the compute you would need through a PHD for the price of a 4090.
That's not their moat at all. Most students don't have a capable gpu in their laptop. And these days everyone who is learning doesn't care about running on Nvidia, AMD, a cpu, etc. It's easy to move from one to the other (nominally).
The moat is their high quality libraries for speeding up common operations on Nvidia gpus.
AMD refuses to invest in ML and build the same libraries. Their stack is horribly buggy to the point of being unusable. It takes forever for their ML stack to get support for their new GPUs, even the crappy version they always produce.
The students who are really interested in AI will buy desktop PC anyway. And the fact that they can use the same GPU for model training and for playing Diablo IV makes the whole investment easier to justify.
I dunno man, stuff like google collab is ubiquitous for students these days. I asked about building a pc on the AI/ML subs on reddit and people basically and the default advice was that few did that these days, and they're right, compute prices have been really low until recently even if you decided to go with premium offerings.
Last time I did the math, by the time I spent 4090 money on compute, I could have been done with a PHD. Now, if all you need is a 3060? Yeah you could probably make that work, but you wouldn't be doing LLM inference even quantized
Either you only use the GPU sporadically or your math is very different from Tim Dettmers` [1]:
"The break-even point for a desktop vs a cloud instance at 15% utilization (you use the cloud instance 15% of time during the day), would be about 300 days ($2,311 vs $2,270):"
And that was written before the current GPU crunch and assumes availability in the cloud, which currently is not a given at all.
I expect nvidia to sell or close down their consumer division. Their whole business reality has changed in a few short months. Investors want undivided attention from the ceo to the new b2b cash cow.
This would be extremely foolish, considering the rate of ML research. There is no evidence that LLMs & supersized models are the end-state of ML research. In 3 years time, there might be completely new SoA technique that doesn't demand high memory bandwidth, which will completely flip the economics of pricey high-end cards.
It's good they are selling shovels while there's a gold rush, but discarding the rest of your business to focus on a specific shovel that's currently popular may not end well.
You may be right. However, Nvidia seems to develop their stack at a pace that it's difficult for the competition to keep up with.
They may end up in a position similar to US Steel or Standard Oil at the start of the 20th century if they continue to meet market demand as they do now, and AI is goes on to have the economic impact that steel and oil had in the last century.
I'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.
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.
There isn't one, that's what's so annoying. There are a lot of Pi-likes, but nothing with the same level of support and enthusiasm the Pi used to have. The community has just withered.
It's incredibly sad. I was in the hobbyist community as a younger teen since the beginning of pi till 2016-17. When I stopped at that time ESP8266 had just so much potential as the cheap microcontroller with wifi and RPi Zero W as the cheap microcontroller with wifi as well. The development environment had improved so much over the years and things seemed great. The alternative pi clones also were building their own identity and figuring out their software.
Over the years I felt guilty for not experimenting given that Rpi4 had 8 gigs of ram! and surely the competition would be better. I bought a bunch of Zeros, ESPs and 4s when I couldn't find other alternatives much but surely that was because I was out of the community and didn't know what to buy. And surely there would have been ESP alternative in all these years with lower power alternatives now that IOT is going to be a bigger thing finally.
Alas, last year I came back to the hobbyist scene finally! Excited! And it was all a major disappointment. All my favourite hobbyist stores in my country had shut down or increased prices for basically everything by a huge margin. It all seemed more polished but so much more void of the vibrancy of the earlier scene. All the experimental boards sort of failing. And I can't even buy the boards anymore. Thankfully I could retrieve all the boards I had bought in past so didn't have to chase boards in flash sales as everyone else I knew had to.
But it's so sad.
There isn't a successor board, not really. But people have started to buy ultra small form factor (USFF) x86 computers and thin clients that got cycled out by corporations and leasing companies, with great success I might add.
The price/performance (+size) is about the same as for the Pi, the power draw is not as bad as with older Intel/AMD machines, and your I/O options and x86 software compatibility is vastly better.
The only drawback is that you're buying used instead of new, and that your starting price for a small home lab (albeit one that starts at 2-3x the power of a single Pi) is larger, which harms students and other makers on a shoestring budget.
>There is no way in hell all these expensive fabs will pay for themselves if chips are too cheap.
While this isn't exactly the case here since it is Hi-End packaging that is the problem. But I am still happy to read the statement on HN as I have been ranting about 99% of comments giving little to zero credit to current state of art chip manufacturing and cost.
> 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.
Far left ones don't go and shout "Ausländer raus" while marching in the thousands, which was a common occurrence during the high days of the weekly "Pegida" nazi marches in Dresden. Nor do they randomly beat up random people (not to mention protesters) that are just standing around [1].
While far left violence is currently a smaller problem in Germany than far right violence, it certainly exists.
This article [1] claims that there were 399 reported assaults associated with left-wing extremists, compared to 1013 associated with right-wing extremists.
Edit: and when it comes to property damage and arson, the far left is worse than the far right by about 3:1 [2]. This may be particularily relevant for companies who want to start new businesses.
There are some concerns that the internationally recognized image of Saxony as an extremely xenophobic region will deter international talent.
The main problem isn’t staffing the plant with human assembly and packaging drones but having engineers and IT join permanently after production starts.
Rural saxony is not a place where you want to live if even your last name isn’t deemed ‘white’ or ‘German’ enough. And after years of cultivating this image it started to work against the best interests of the more rural parts of the region: brain drain galore and no sane person seeking work in a future proof place to build a life would come.
Let me know when companies actually start pulling out of Saxony (not blog posts from randos - most people don't care as much about politics as the people who wrote those blog posts do).
Translation: there are a lot of talented Asians, and we’re worried they won’t want to live there. Is that accurate?
It seems like most of the US has this problem. And yet rural IT always seems to find a way. I’m in Lake Saint Louis with gigabit symmetric fiber, which is a dream. We were backwater for awhile, but post COVID a lot of talent can be sourced remotely, at least for software.
Is a fab really so regional? I could imagine them having a team in Taiwan who they work with remotely. But I know very little about fab plants, and it sounds like you have a better idea of their engineering and IT requirements.
I used to work at Groq, a hardware company, and we were fully remote. So I’ve seen this work in practice. (It’s a tossup whether Groq will succeed, but that’s a separate conversation.)
The US still benefits from latent good marketing/rep in the 80s/90s, and projection of military power in decades past. For example, until relatively recently (the Trump era and especially how it ended disillusioned many), South Koreans still looked to the US as their model of the developed world. If you'd send your children anywhere to get educated or marry, it would be there. Inner benchmarks were done against the US. The rose-tinted glasses were enough to convince many to move without checking too carefully what the place they'd be moving to would actually be like.
This is changing, and more recently more attention is being paid to Europe, partially because the national conversation has shifted more toward quality of life, birthrate problems, aging population, etc. and there's more interest in social systems, healthcare, family planning support and so on.
Still, Koreans have a lot of memes about this or that place in Europe being racist in this or that way (and of course, quite lot of it is true), so it remains a big impediment.
That seems like a 'slight' overstatement. Saxony is also full of crypto-communists (third highest polling state for Linke)...
> The main problem isn’t staffing the plant with human assembly and packaging
So why on earth are these companies building they are fabs there instead of some other place in Germany? I mean they can't be that stupid to not understand all of this?
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.
The problem is the bull whip effect. Yes there’s loads of demand now, but are we really saying that the AI boom is a permanent uptick in demand for silicon, or is it more likely that the hype dies down? By the time capacity comes online demand may well gave peaked and dropped. Suddenly these massive sunk cost fabs will be taking any business they can to amortise the costs.
It seems more likely that efficiency gains decrease aggregate compute requirements than people stop using AI tools altogether.
At minimum, LLMs work as a better human-software interface medium, which means they're going to be infused in almost everything non-technical consumer-facing.
Even absent any other breakthroughs, that seems like a huge consumer of AI hardware capacity.
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 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.
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.
I guess where both you and the GP are correct is that you can't make an autonomous car by applying transformers exclusively.
LLMs are "language I/O interfaces", and as such there isn't a lot of value they can create alone (even most of the currently proposed uses are bullshit). But it's quite likely that they'll be there on a lot of advanced technology helping it do what you meant.
Still even though both will be there in the transformative technology, and even though the technology may not even be useful without them, they are not central pieces on most of it.
Everyone is buying up H100's (and A100's if they can't find H100's). Sure, in 18 month people may lose interest in buying H100's but then there will already be the next Nvidia model everyone wants to buy to get ahead of the competition in terms of compute capabilities.
106 comments
[ 3.2 ms ] story [ 171 ms ] threadThis whole fabrication thing is a biiig chokepoint.
This design is quite non-trivial even today. It took several of the brightest minds of all time to get it just right back then (like von Neumann).
Even with modern tools, you still need a team of quite capable engineers/scientists to pull something like that off, even if you do have the plutonium.
The little boy design is easier to engineer, but bomb grade uranium is even harder to produce than plutonium.
The machining bit so, uaing proper machinery, is easy. And the necessary engineers and technicians readily available.
What makes it easier, though, is that the general principle is relatively well known, so one could always brute force the solution using computer simulations, even if unable to do the math by hand.
The cost of the subscriptions to the design tools provided by Cadence, Synopsis, Siemens (Mentor) and a few others is ridiculously high and it can contribute with many millions of $ to the design cost of a state-of-the-art chip.
The design tools themselves have been improved extremely slowly and they are of a quite low quality for so expensive professional tools. In an open-source environment it would be easy to make much better alternatives.
The reason that only a few timid attempts exist to provide alternatives for only some parts of a complete toolchain, is that the semiconductor manufacturers, like TSMC, do not provide public information about the exact data that they need as input for manufacturing, about the design rules that must be implemented for valid designs and about the characteristics of their processes that must be known for simulations.
All this information is available only to big companies like Synopsis, Cadence etc. and this perpetuates their monopoly and their ability of extracting huge amounts of money with minimal software development work, from all the companies that design ICs.
While the companies that design ICs are not happy for this huge rent that they pay for the design tools, they are also extremely risk-averse and they do not want to make any change in their design workflows, no matter how small, that might increase the risks of failure in a designed chip. Because of this there are no chances of replacing the incumbents with open-source tools, at least not in the big established companies that design chips.
There are FOSS IC design tools but nothing that can match Synopsis, Mentor and Cadence and the semi industry isn't gonna start dealing with FOSS jank and destroy their productivity and competitiveness just to save a few milion in licensing costs to those 3 companies, which are peanuts in the end.
They're like the John Deer of semi tools, expensive to own and run, but necessary to be competitive and get the most out of this field.
Plus, the proprietary stuff is still the standard-cell libraries of TSMC, Samsung, Glo-Fo etc. They will never open source this.
>and this perpetuates their monopoly and their ability of extracting huge amounts of money with minimal software development work,
Oh, you mean just like Microsoft, Google, Apple, etc?
Because they sell ads and monetize their users instead, which is a pipe of infinite money.
Why does it matter to you how much they're charging? It's not communism. It's their honest labor and they're allowed to charge as much as they want for it, same how you're allowed to ask for as much money as you want for your salary negotiations.
So please excuse traditional engineering companies for trying to make a living by monetizing their hard earned IP like every single company out there, and not selling ads like big-tech.
The world still needs traditional engineering companies to exist to progress, we can't have a society where everyone only works by selling ads to one another.
If you feel like they're price gouging, then feel free to write your own EDA tools and sell them for cheap, I'm sure the market is open to paying much less and saving some money.
However most foundries, like too many other businesses today, are not interested in new customers, unless they already are very big companies, which is one of the factors that has resulted in the extraordinary reduction in the number of companies that compete in every field, in comparison with the situation from a couple of decades ago.
[1] https://en.m.wikipedia.org/wiki/Chip_War:_The_Fight_for_the_...
Open sourcing software works because some many people already have the hardware (a computer)
If you’re trying to draw parallels between this is and, say 3D printing, you will not find many similarities. PCB fabrication is also a a different beast but a lot of people think it’s as easy as finding a fab house that takes gerber files and creates a board for them.
If we want something that more closely resembles how we develop software, we probably will need to develop processes that enables us to flash FPGAs like MCUs, make them fully read only, and allow the customer to etch their own marking onto the chip. There facilities that produce ASICs like this already, but I can see something more standardized becoming more successful for general purpose. This does however require that we have a foundry create the original chips, the difference here is that they are producing the same standardized chip for everyone which would allow any foundry to make compatible alternatives.
I'll bet 3 years. Using mostly stolen IP, because they aren't allowed to buy it.
Given that the nanometer rating is purely marketing/branding and doesn’t translate to actual sizes of anything nor translate between manufacturers, this really doesn’t mean anything.
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.
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 were really "shipping a ton of units" of the 40 series and didn't have excess inventory i doubt they'd have done that.
- https://www.dexerto.com/tech/nvidia-rtx-4070-production-repo... - https://www.dexerto.com/tech/nvidia-rtx-40-series-production...
Rarely does Nvidia sell a GPU for 2000 dollars or more, and when they do its relatively low volume.
An H100 sells for tens of thousands of dollars and is sold out to 2024 - companies are literally fighting over them - I can absolutely see conditions might arise Nvidia might want to use some fab capacity for their much more profitable H100 parts vs consumer GPUs.
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.
I think many, maybe even most, would claim this is now best done on cloud services, like huggingface.co.
Stable Diffusion blew up when people are able to run it at home.
The moat is their high quality libraries for speeding up common operations on Nvidia gpus.
AMD refuses to invest in ML and build the same libraries. Their stack is horribly buggy to the point of being unusable. It takes forever for their ML stack to get support for their new GPUs, even the crappy version they always produce.
If you start having physical space guarantees, then the calculus changes...
"The break-even point for a desktop vs a cloud instance at 15% utilization (you use the cloud instance 15% of time during the day), would be about 300 days ($2,311 vs $2,270):"
And that was written before the current GPU crunch and assumes availability in the cloud, which currently is not a given at all.
[1] https://timdettmers.com/2023/01/30/which-gpu-for-deep-learni...
It's good they are selling shovels while there's a gold rush, but discarding the rest of your business to focus on a specific shovel that's currently popular may not end well.
I mean, we can just go ahead say "Intel" instead of beating around the bush.
They may end up in a position similar to US Steel or Standard Oil at the start of the 20th century if they continue to meet market demand as they do now, and AI is goes on to have the economic impact that steel and oil had in the last century.
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.
I keep hearing this - but I've seen no evidence of any successor board. Which board would you say is now the go-to board for hobbyists?
The price/performance (+size) is about the same as for the Pi, the power draw is not as bad as with older Intel/AMD machines, and your I/O options and x86 software compatibility is vastly better.
The only drawback is that you're buying used instead of new, and that your starting price for a small home lab (albeit one that starts at 2-3x the power of a single Pi) is larger, which harms students and other makers on a shoestring budget.
While this isn't exactly the case here since it is Hi-End packaging that is the problem. But I am still happy to read the statement on HN as I have been ranting about 99% of comments giving little to zero credit to current state of art chip manufacturing and cost.
https://rpilocator.com/
CM4s are still being sold for hundreds of dollars on eBay.
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.
What does that have to do with anything?
California has plenty of neonazis itself. The western world in general has no lack of far right extremists.
[1] https://www.zeit.de/gesellschaft/zeitgeschehen/2018-10/recht...
This article [1] claims that there were 399 reported assaults associated with left-wing extremists, compared to 1013 associated with right-wing extremists.
Edit: and when it comes to property damage and arson, the far left is worse than the far right by about 3:1 [2]. This may be particularily relevant for companies who want to start new businesses.
[1] https://www.dw.com/en/left-wing-extremism-in-germany-how-muc...
[2] https://www.statista.com/statistics/963069/right-wing-left-w...
It seems like most of the US has this problem. And yet rural IT always seems to find a way. I’m in Lake Saint Louis with gigabit symmetric fiber, which is a dream. We were backwater for awhile, but post COVID a lot of talent can be sourced remotely, at least for software.
Is a fab really so regional? I could imagine them having a team in Taiwan who they work with remotely. But I know very little about fab plants, and it sounds like you have a better idea of their engineering and IT requirements.
I used to work at Groq, a hardware company, and we were fully remote. So I’ve seen this work in practice. (It’s a tossup whether Groq will succeed, but that’s a separate conversation.)
The US still benefits from latent good marketing/rep in the 80s/90s, and projection of military power in decades past. For example, until relatively recently (the Trump era and especially how it ended disillusioned many), South Koreans still looked to the US as their model of the developed world. If you'd send your children anywhere to get educated or marry, it would be there. Inner benchmarks were done against the US. The rose-tinted glasses were enough to convince many to move without checking too carefully what the place they'd be moving to would actually be like.
This is changing, and more recently more attention is being paid to Europe, partially because the national conversation has shifted more toward quality of life, birthrate problems, aging population, etc. and there's more interest in social systems, healthcare, family planning support and so on.
Still, Koreans have a lot of memes about this or that place in Europe being racist in this or that way (and of course, quite lot of it is true), so it remains a big impediment.
That seems like a 'slight' overstatement. Saxony is also full of crypto-communists (third highest polling state for Linke)...
> The main problem isn’t staffing the plant with human assembly and packaging
So why on earth are these companies building they are fabs there instead of some other place in Germany? I mean they can't be that stupid to not understand all of this?
[1] https://www.spiegel.de/wirtschaft/fachkraeftemangel-bloss-ni...
Do they have the same issues?
I assume the open border facilitates travel somewhat.
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.
At minimum, LLMs work as a better human-software interface medium, which means they're going to be infused in almost everything non-technical consumer-facing.
Even absent any other breakthroughs, that seems like a huge consumer of AI hardware capacity.
——
Being a bit more fair, yes a bubble did burst then, but it was a “too much, too quick” type of bursting, not a “this whole thing is a scam” type.
(And as far as I can tell this current craze is entirely predicated on LLMs/transformers.)
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.
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.
LLMs are "language I/O interfaces", and as such there isn't a lot of value they can create alone (even most of the currently proposed uses are bullshit). But it's quite likely that they'll be there on a lot of advanced technology helping it do what you meant.
Still even though both will be there in the transformative technology, and even though the technology may not even be useful without them, they are not central pieces on most of it.