I don't think $1 billion dollars is actually enough in this kind of space.
Probably the most important factor in Intel's success is its (relatively) new CEO's ability to fix cultural problems and internal dysfunction that inhibit any technical improvement.
At least this is a sign that Intel recognizes this and is reaching out to startups - only time can tell if it works.
Next TMSC foundry is x1.7 transistors from current process (the 'obsolete' process of the foundries intel is currently building in the USA). Samsung will probably follow, if not done already.
EUV core tools of state of the art foundries are ASML (Netherland with bits from Germany). One of this tool was 100 millions last time I did check the price. Hence 1 billion is too small to be relevant in the foundry business, for large scale mass production.
If you read the article, the $1B is not to go the fab itself (there's another, much larger pot of money for that); instead, it's for ancillary developments of "disruptive technologies for the foundry ecosystem."
They way I read it it's about enabling fabless semi companies to use Intel fabs, and providing Intel tech for designing high end SoC's like TSV's and whatnot.
What I could gather from the net (much on wikipedia):
Samsung seems not to have finished its n3 process, pushing it back to 2024 (bad news for AMD since they are fed up with apple using much of TSMC production capacity).
TSMC should start risk production this year. If they don't increase too much on-die static ram, GPUs based on this process ""should"" dwarf current GPUs. How much more transistors are they going to allocate to ray-tracing :) ?
I don’t think this is necessarily true. The large investment required to set up a fab does benefit them in certain ways. It reduces competition. If the machines were 50 million instead of 150 million, I’m sure they’d like that. If the machines were 1000 dollars, I doubt they’d be happy about it.
Right, but it's so easy to comment on the internet "if only they could reduce the price," as if that's not something TSMC and Intel literally hire engineers for to do whatever they can. (Yes, ASML builds the machines in the end, but they have a very tight relationship with their customers and their needs, to the degree it is almost a co-engineering job.)
Intel even pre-ordered the next gen EUV (called High-NA EUV) machine from ASML that is still in development and isn’t expected to ship before 2025. Each one will cost $320m
Historically, the semiconductor business has been highly cyclical. High prices make firms overbuild capacity. Then overcapacity crashes chip prices. There is a good chance we are at the top of a cycle with this chip shortage thing. When you are at the top there is only one direction you can go...
They are just not able to ship faster and I think there was a fire in their factory a while ago which may contribute to even slower fulfillment...not sure though.
Because they aren't able to increase the production, and because they aren't able to sell to China. Which means no clients except TSMC, Samsung, and maybe Intel.
Heard the US admnistration under Trump "pressured" ASML (~netherlands) not to sell the state of the art EUV tools to China. True? how much true or how much false?
Why? Because China would flood the world with dirt cheap open source based super performant (mobile/desktop/servers) royalty-free RISC-V cpus (and gpus...) using a top notch node process (that with tons of backdoors ofc)?
ASML EUV tool: 20 millions of $, then 100 millions of $, and the next one would be 320 millions of $?
The valuations on ASML are super high (approx 40 P/E ratio). ASML is a good company and the word got out, but it was overbought, seeing approximately 4x in price from the beginning of 2019 or so to now.
The better question is why is the stock up several hundred percent in the past few years, even after falling a bit recently? The business is growing, but not that much. Good company, bad price.
Can anyone in the space explain why it would be hard for Intel to be competitive in the GPU space? Is GPU development so far away from CPU development that they couldn't succeed?
One reason I can think of is that GPUs are a much newer invention than CPUs, and as such there is less history behind how they work and there are far more active patents on their technology.
I'm not "in the space" but there are many reasons a detractor might give for being doubtful.
- Intel is not already experienced in designing high-end graphics cards, so there's less institutional knowledge in the company
- Intel iGPUs have a mediocre reputation
- Intel has a history of failed attempts to enter the GPU/accelerator space (e.g. i740 and Larabee)
- DG2 has been delayed from the timeline that was expected in the techspace when we first heard about it
- The person leading Intel Xe is Raja Koduri who led AMD graphics during AMD Vega (though it's unclear how responsible he is for its shortcomings) and arguably has a history of overpromising.
That said I'm pretty bullish on Intel GPUs because supply still hasn't caught up with demand, so as long as they are reasonably functional, I think they will succeed.
> - The person leading Intel Xe is Raja Koduri who led AMD graphics during AMD Vega (though it's unclear how responsible he is for its shortcomings) and arguably has a history of overpromising.
Raja was the technical leader of ATI in 2001+. His hands were involved in every single GPU from that time until Vega.
Intel has been pretty bad with bringing new products to market and sticking with them.
Intel Optane, Itanium, Xeon Phi, Cell-phone Atom (yes, this was a thing), Intel Edison, etc. etc.
Intel pretty much makes good Ethernet adapters, Wifi adapters, and CPUs these days. Everything else they've tried seems to die off a few years later.
----------
Its like seeing "Google" try to make a new free webservice. We all know Google has the expertise to make it work, but for some reason, it doesn't.
Similarly, Intel has a wide variety of experts in a number of fields related to chip-making, along with historically successful products and a few chips that are highly important in the field. But something is clearly "off" with Intel's current culture.
---------
Intel has some GPU-engineers working for them. The question is if the organization as a whole can actually release a product (step 1), and then support it for more than 5 years.
I think its a bit of the "curse of the successful company". No matter what product Intel makes, there's no way its going to make as much money as Intel's CPUs. So the business-minds / bean-counters kill the product.
As each day passes very bright students are forced to not persue a higher level education in the U.S. Tuition and costs of living have doubled to tripled since 2010 with scholarships and funding dwindling. Why persue engineering, especially complicated engineering, when the companies that need it refuse to fund education. Intel, Apple, Microsoft, Google, all the big most wealthy tech companies refuse to put money into the future of U.S. engineering.
Startups for chip tech is a joke. The knowledge and domain specific equipment necessary is so great that you can't just pick up a tutorial through Google and use a framework to make your product. We need to tax these behemoths and put that into the education and tooling of people. That's how we create competitive markets and innovation.
They are throwing money wherever they can find an audience.
> "Startups for chip tech is a joke."
Last time I checked, there was a startup called SiFive in the US from a bunch of UC Berkeley professors which successfully had its founders invent a whole new instruction set, RISC-V, a compiler for it, and are now building the highest-performing RISC-V chips on the market.
> "We need to tax these behemoths and put that into the education and tooling of people."
OK, by your own admission there is a ton of industry-specific knowledge. As linked above, fabs are trying to get into universities where they can, but finding qualified people is difficult. I don't know how taxation would help solve this problem in any way, because government is typically even less efficient at teaching domain-specific knowledge than industry.
Marketing. Unless you have receipts in hand Intel is not giving out money. They selectively choose specific institutions of where they draw from a pool of some graduates.
The problem we find ourselves in is that we have no competition. If Intel is the only one funding, at very specific institutions, for their own gain, that's not competition. You need competition to innovate. You need a lot of money and equipment to learn. Intel is not putting that in the hands of any institution they don't control.
I'm saying that you can't make a startup for chips out of nowhere like you can software startups. We have to have the underlying education first. The U.S. does not have this, and continues to fall over itself because tech companies have significant Democrat control that won't tax them.
It is weird to me how the perspective on Intel flipped a while back.
For a long time, the discussion was that Intel had failed, failed and failed again to get their next generation fabs operating at acceptable yields that they were a couple of generations behind TSMC.
Which was a huge departure from their decades of first or least competitive on fabs.
Intel announced they would start having some chips produced by TSMC,
so they could try and harness benefits of the latest, or the generation
behind latest for their chips.
In short Intel was seen as incompetent and could no longer compete with their competitors. The long Intel domination was sliding away.
Then people started worrying about Taiwan.
Intel got a few billion dollars in a sweet contract with the military,
building fabs in the US became the major push in tech.
Since then, the perspective has been much more positive for Intel and
their role in building domestically.
Intels inability to build a latest generation fab has not changed.
New leadership and a more urgent geopolitical landscape has forced many to change their actions if not their outlook. We just need fabs. We have money. Intel has been saying for a long long time that we need fabs, so lets give them the money.
personally, I think the gloom and doom around Intel is hate-hype. It's so fun to look at the giants and laugh, but they're still the giants. If you look at the landscape right now, and ignore the past, you see Intel and AMD in a slugfest for domination, and no (?) real other contenders. If AMD were the incumbent, we'd all be cheering Intel, but they're still neck-and-neck regardless. And Intel has other focii, like compilers, HPC software, kiosk form factors, etc, that help round out the offerings (and those are just things I'm familiar with).
20 years ago Microsoft was evil and google was the little-man's angel. Now I'm much more a fan of MS than Google, and feel MS is moving in the right direction while Google is not. It seems generational.
This stuff takes years to turn around, before Zen AMD was in an even worse spot. Intel is down but not out, they have capital to spend and are making the right moves to get back on track.
If you mean Chip ACT, they don’t have that money yet. All other announcements are tax breaks which they haven’t got any either; not until they started building the fabs they’ve announced
> Intels inability to build a latest generation fab has not changed.
That is going to take ~5 years, according to Intel's publicly released roadmaps (and if they actually succeed, which they may not).
That being said, Intel's "Intel 7" manufacturing node that they are shipping Desktop and Laptop processors on currently is surprisingly competitive. In the markets Intel is competing in, only Apple is currently using a more advanced manufacturing process to my knowledge. TSMC's N7 process, used by AMD and many others seems to be pretty similar in performance and efficiency.
The problem is that N7 process, is from 2018. thats 4 years that TSMC have been developing their N6, N5, N4, N3 processes.
N5 (which Apple uses), was late 2019 with the A14 processor with AMD starting to jump in this year with it.
N3 should be out this year, more than likely with another flagship apple product.
Now. As per Intels roadmap, https://www.anandtech.com/show/16823/ , They should have Intel 4 out later this year. Which should be slightly faster than N5. The big question is whether they stick to their schedules.
I think the sentiment always was that Intel would be doomed, if it were not for the fact that it’s too important an asset to the US gov. Same for Boeing.
What I’d say matters the most is whether Intel can keep up / take back the lead in terms of innovation with TSMC / AMD / all ARM things. The next 5 years will be super interesting to see how things develop.
This comes across as desperate to me. I don't think Intel can compete meaningfully with TSMC at this point, and process is literally all that matters as we hit the limits of physics. This sounds like a last-ditch effort to find _something_ which will help them out of the hole their currently in with waves hands ... hope?
Intel is roughly a quarter to a third the size of NVidia, a tenth the size of Apple, Amazon, and Microsoft... AMD is starting to eat into the server market, and ARM is more of an existential threat now that Apple sells ARM-based computers and Amazon is deploying ARM based servers. They are no longer a powerhouse, and IMHO, very likely won't ever regain their lead. I think Intel is going to end up like IBM, largely irrelevant, but still hobbling along somehow how as a ghost of its former self. Once single core performance on ARM meaningfully exceeds x86 (which it likely will because energy efficiency will eventually be the defining factor to enable higher speeds) for their most profitable markets, like Fintech it'll be "game over"[1] for Intel.
Intel's management has as a matter of fact: fucked around for the last decade. There is little they can do at this point short of a miracle, to undo their chicanery and prevent the "find out" portion as the market leaves them behind. It's just too bad it's the employees and retail stock holders who will get fucked over in the end.
[1] "Game over" for a multi-billion dollar juggernaut that lasted decades is hobbling along shilling their shittiest technology for decades (think HP).
The vast majority of the semiconductor market doesn't use bleeding edge nodes, even at TSMC which only has something like 20% of their current production on industry leading nodes. Even if Intel never catches up to TSMC and remains a year or two behind, that still leaves them a huge percentage of the market that they can potentially serve. There is a lot of money and volume in boring, unsexy chips like automotive and industrial microcontrollers and the like.
I'm not sure I have much hope for Taiwan remaining independent once we don't need them to make chips for us anymore. I think it's a great and necessary move for Americans to manufacture at home though.
Probably the most important factor in Intel's success is its (relatively) new CEO's ability to fix cultural problems and internal dysfunction that inhibit any technical improvement.
At least this is a sign that Intel recognizes this and is reaching out to startups - only time can tell if it works.
reply