Nvidia CEO Says Intel's Test Chip Results For Next-Gen Process Are Good
Intel-built Nvidia GPUs could be on the horizon.
Nvidia on Intel 3?
Nvidia CEO: Intel Test Chip Results for Next Gen Process Look Good
- Thread starter benb
- Start date
Nvidia on Intel 3?
For now, let's see if that continues. For Meteor Lake, and their current designs, it might be playing it safe.
But, also, who says NVIDIA is looking to make their GPUs there? They might be, but as unpopular as it is to say it here, it could easily be argued that Intel even now has the best CPU process available, and I suspect I3 will continue that tradition.
I definitely understand what you're saying though, and share the same thoughts. If they don't make their own GPUs there, how can they expect others to? It's not a good look for them, if it's ongoing.
My question to everyone is, what will they be making in their fabs when Meteor Lake comes out? I'm baffled. I mean, it's a shrink (and likely fewer cores as well), and it's only the CPU die (and the transposer made on a cruder node). So, it's going to be much smaller than Raptor Lake and Alder Lake. And given mobile is around 80% of the market, how are they going to keep their fabs busy when Meteor Lake ramps up? Will the move to I3 and transition everything when the capacity frees up? And then, how do they fill up their fabs with I3 orders? It just doesn't seem like they have the volume for that.
So, I'm confused how they are going to keep their fabs busy as they transition to Meteor Lake. It seems like there's going to be a huge gap there.
MKWVentures
Moderator
Companies will tryout all foundries for their chips as they always do. Then they decide who to go with (price will be a big factor). We shall see who get the contract and volume orders. I will be interested to see pre-payments and volume loadings intel reports over the next 2 years.
Very unlikely.
Nvidia CEO Says Intel's Test Chip Results For Next-Gen Process Are Good
Intel-built Nvidia GPUs could be on the horizon.www.tomshardware.com
Nvidia on Intel 3?
I believe it is 18A. Nvidia is all in on TSMC N3.
Very ironic.
Nvidia is also embracing chiplets so they can use multiple foundries with relative ease. The test chip may be for a chiplet which would be easier to yield on a new process.
hist78
Well-known member
I think it's not just you that has such question. Many people, probably including Intel, are confused about how Intel can utilize "all" its extra fab capacity when Intel's major products move to new generations of node technology.
Ideally all those new available capacity can be filled by new foundry customers and/or migrate to the new node generations. But it is not an easy task.
At least for the next five years Intel's own product divisions will be the largest customers of Intel Foundry Service and use the majority of the IFS' capacity. Any new Intel product generation will possibly create some major capacity utilization issues.
Manufacturing to a spec!
Only two real variables if the spec is met.
Price and cycletime.
If you cant compete on one its the other.
But, in this case, it's even worse than a normal shrink. If you look at the dies of current processors, a lot of it is logic that will be made at TSMC. So, it's worse than a normal shrink, by a lot, and then we have the processors having fewer cores. It's really quite a big drop, even compared to normal node progressions.
Ideally, yeah, I'd love to see fab customers chew up all that extra capacity, but I'm seeing mostly "wait and see" rather than jumping in big time on Intel 3. It looks like most companies are looking more seriously at 18A, and it's also difficult to see companies going in fully with Intel until it can prove it is a partner that can be trusted.
Are you viewing it differently? I sure hope I'm wrong, but I just haven't seen a lot of enthusiasm for Intel 3 yet. Is it too early?
“Irony”—The TSMC deal was inked in the Swan era of Intel, which hedged on whether Intel could get to 3. To be a technology CEO must be very hard.
”Future”—Intel’s future, like TSMC and Samsung, is with AI. We will have to wait and see. But the Grace Hopper chip modules Nvidia was showing off are massive. The AI era seems it will need all the fabs it can get.
”Future”—Intel’s future, like TSMC and Samsung, is with AI. We will have to wait and see. But the Grace Hopper chip modules Nvidia was showing off are massive. The AI era seems it will need all the fabs it can get.
hist78
Well-known member
That means it's coming from an Intel R&D environment. What's the catch there?
hist78
Well-known member
1. A dedicated foundry, like TSMC, builds fab capacity based on customers' commitment and prepayments. On the other hand Intel chose to make IFS to serve both internal and external customers. IMO, it's a tough situation for Intel to deal with when time is good or bad. Will IFS sue internal Intel brothers and sisters for not fulfilling their order commitments?
2. Probably many people here do not agree with me on this. IFS is forced to operate under an awkward situation. There are many conflict of interests and contradicting goals they need to pursue. What you have pointed out is just one of the difficulties Intel is going to face. To me the best solution is to spin off IFS to set them free. It's good for both Intel product divisions and IFS.
But Pat Gelsinger thinks Intel must keep both product divisions and IFS together in order to survive. I think it's a high risk and unnecessary gamble.
When people buy products from AMD, Apple, Qualcomm, Nvidia, and Mediatek, they don't really care who actually manufacturered those products. They buy a particular product is because the superior price, performance, quality, features, and availability. None of the above mentioned companies operates a fab and all of them are doing just fine.
Intel is a product company, not a service company, because majority of Intel revenue are generated from selling semiconductor products. Intel biggest problem is that it doesn't have a killer product. To spend billions to create a new service business (IFS) is placing precious money, time, and management attention on wrong priority. IFS won't be able to bring revenue to Intel that is large enough and quick enough to offset Intel product division's shortfall.
Last edited:
Actually, Intel said they will also build out capacity based on commitments, but they wanted the "shells" in place because they have the longest lead time.
They do have conflicts of interest, but they also have advantages "half" companies don't have. Like they have processors and logic, and can assist companies in unique ways, by offering their processor IP. It's good and bad. I doubt AMD will use Intel fabs, but we already know that NVIDIA will if things line up. And NVIDIA competes pretty directly with Intel, and likely will even more as they move into processors more aggressively, and Intel moves into GPUs.
Agreed, people don't care about who makes the chips, or even know in most cases. But, that ignores the advantages Intel has over companies like AMD, which make their products more attractive to customers. For one, look at the clock speeds they already reach, which AMD/TSMC can not. Raptor Lake won the performance crown, to a very large extent, because of the clock speeds. On top of that, with Intel getting 87+% of the PC revenue by the last report, they clearly have an advantage somewhere. Is it their design? Maybe a little, but a lot is AMD's malicious pricing, because they have the TSMC lamprey sucking the life out of their profits at 5 nm. Intel has the initial costs of development, but the iterative costs after that are much less, and we've seen the devastating effect they have when demand is low. AMD has to counter by offering obsolete processors on obsolete TSMC nodes to compete on price. While desperately trying to hold onto their sub 20% market share.
I see the fabs as a HUGE advantage. Intel designers have access to the fastest node available, and can use TSMC if they have to as well (or even Samsung). AMD is pretty much stuck with what TSMC and Samsung have, and while it isn't necessarily a painful limitation in terms of quality of their processors (like right now), it has been, and can be in the future. And moreso now, with the expensive 5 nm node making it very difficult for them to compete with Intel in the PC market on price, and losing money there last quarter, while Intel made money. And as we know, Intel can and will use TSMC if needed, although clearly they can't do it entirely, but they can for specific products, while still using their own nodes extensively for other products that might not benefit, or don't need an external node as much. For example, I'd expect them to use TSMC for servers if they ever get as far behind as they did with 14 nm. I'd hope, anyway.
I could see the point where you could say Intel's design would benefit from being separated, but I disagree with it. There would be benefits though, but I think the negatives are more substantial, by a lot. But, how would it benefit the fabs? You'd lose a HUGE customer with advanced designs, and lose access to their IP, not only with hardware, but with software. You'd lose scale. What is the benefit? AMD might come in, with their 18% market share? It's not even likely. The other big players don't seem shy about using Intel as is.
But, that's what keeps horses racing. I think it being one company is a huge asset, and as Jerry Sanders famously said "Real men have fabs". AMD, Qualcomm, Apple, NVIDIA are all at the mercy of another company, which by the way has some nice margins and profits. Intel can use them, but they have other options, and those options have allowed them to have the fastest processors, and have allowed them to remain profitable in the PC segment, despite it being a complete mess, while AMD could not be.
On the other hand, the 10 nm debacle shows how it's not always favorable, but if you look at the company longer terms, the vast majority of time it's been very positive for Intel.
And no doubt you're right, they have to learn more about being a service company, and the fab industry as well, but there's no reason companies can't do both. So many do. Wouldn't you consider both Apple and Microsoft product and service industries, at least to some extent? And they both have been pretty successful at it. Should Microsoft leave the cloud business? Or spin it off? It seems more synergistic than problematic, and the same with Intel and their fabs/designs. That's no guarantee they'll succeed, but there's no reason to think the company can't do both as well. The rewards so justify the risks to me, and that's why I like Gelsinger, and despised "play it safe" Krzanich.
We'll have to see how it plays out, time may side with you, but right now, I think it would be tragic to fragment the company based on fear and lack of confidence in one's ability. That's not normally a good recipe for success, So far, Intel is doing well with their nodes too, so let's see how it plays out. I've got 5K shares of Intel, so I'm not chirping from the sideline while taking no actual risk. If they spin off their fabs, I'll just sell it, because they'll show a lack of confidence, and worse still, be like many other companies, rather than being unique, and full of opportunities. No thanks.
Last edited:
hist78
Well-known member
1. When market is bad, like the current big drop in PC demand, will IFS sue Intel product divisions due to order cancellation?
2. With proper arrangements, an independent IFS can still license the pre-spilt Intel IPs.
3. Currently one of the requirements implicitly imposed on IFS is to maximize the whole Intel corporation's revenue and profit. Due to the majority of Intel revenue and profit are coming from Intel product division, unavoidably it means Intel product division's revenue and profit are the highest priority.
As a team player in the whole Intel corporation, IFS has few options to pursue its own interest. It will negatively impact IFS's competiveness in the foundry market.
4. To do both IDM and foundry service is not a new idea, granted not in huge scale as IFS is trying to do.
Are those reasons that brought down the IDM model and brought success to the fabless and pure play foundry gone?
What has changed that can let Intel IDM+foundry model reverse this history?
If I was to have a guess they would need to pay a cancellation fee. Design side takes the hit when the orders are 86d, and manufacturing takes the hit if they can't fill it.
As has been stated ad nauseam this shouldn't be a problem. Intel design gets their wafer agreements and external gets theirs if you want more; tough get in line behind those with contracts. Everyone who signs a contract will get their wafers as and when they were promised. This is no different to how TSMC works.
How? What is good for it's customers is good for intel. Better customer service creates more contracts which makes more capacity which leads to lower wafer costs for intel design and money that can be funneled back to them. People who use their IP also get design side a payout and make intel software/oneAPI stickier/more widespread.
I think the IDM model is fine. The problem is scale as you have growing R&D/cost of a minimum viable fab for each new node. It makes no sense for a TI or Phillips to move to the next node with how low their leading edge wafer demands were. There just wasn't a compelling ROIC. If IDM was as broken as you say TI, infenion, microchip and the like would sell their fabs and only buy TSMC 90/40nm rather than use their own process technology.
The way it is done. The old way was "Hey, here's the fab and here is our dev tools. Go make a chip I guess to fill up our fab while we are in a bit of a lull.". If that was what the plan was I see no way that could succeed.
Instead the way things seem to be phrased by both BS (intel using industry standard design flows) and Pat (IDM/external being on equal footing and design will use the best node for the product), is that intel design is or will be for all intents and purposes be a fabless firm with a tight integration with intel manufacturing (ala TSMC-Apple). So in a way those asking for intel "to spin out the fabs" might have gotten their wish, with intel sorting being a holding company for intel design and intel TD-MSO-IFS. However time will tell if IDM 2.0 ends up anywhere near as large of a shake up as Pat/Stu talk it up to be.
Additionally I would like to point your gaze to Samsung. Let's wind back the clock to the old Samsung logic days. 100% of their "foundry" (if you can even call it that) was Apple. They and the rest of the common platform alliance botched HKMG and simultaneously they stole Apple IP for their own designs. Other than being a good consistent supply source from Apple that din't ever screw them out of wafers. Samsung did everything wrong and they didn't really have any service component whatsoever (and as we all know foundry is all about the service not the manufacturing). After that they had the LSI split (which seems less radical than IDM 2.0 but time will tell), and as far as I know they have never had a customer data breach since. They also then developed the wonderful 14LPE/LPP and got it out in large volumes before TSMC's 16FF. Everyone and their mother then designed some parts at Samsung, heck even Apple went back when TSMC's 16FF ramp was slower than Apple would have liked. Since then Samsung foundry has broken customer trust again and again with poor initial yields on 10LPE/derivatives of 7LPP, poor wafer supply/velocity on 7LPP, and parametric/performance issues on 5/4LPE. According to Dan they also have a poor PDK and are hard to work with. In spite of how much they have bungled it, and how many customers have gone all TSMC or scaled back their reliance on Samsung; they are still the number 2 foundry (albiet with TSMC having a larger share lead then before). I think that had Samsung kept up technologically, had a better PDK, or better customer centricness/support they would be a MUCH bigger force in foundry. If intel could just do the last two items I think they will take lots of share from Samsung and some from modern node TSMC. If they can do the last two, but with technological leadership, then I think intel can be successful enough that we might need to reveres the order that IDM and foundry are listed in. Given this, I think IDM+foundry can work; and Samsung was close to making it so. However intel has to get those basics down to succeed in this business. Fortunately for them. I think there are no fundamental roadblocks that will prevent them from doing so.
That would be silly, and legally impossible.
Which IP, for instance?
You must think Intel's top-line staff are very short-sighted. Spend tens of billions of dollars to get into the foundry business, and then screw the foundry customers for a few billion dollars (maybe) of extra margin for the quarterly bottom line? Would you make that decision if you were Intel's CEO?
Not unless Gelsinger is stupid, and while I don't agree with all of his decisions, he's not stupid.
No, the need for foundries is increasing, not decreasing. This is the trend to "bespoke silicon" in the largest cloud/SaaS/mobile computing companies Dan talks about. Fabs are a lot more difficult to build and operate than chip design is. Orders of magnitude more difficult, IMO. Assuming Intel gets its design act together, which needs fixing badly, there is no reason I can think of why Intel's manufacturing group can't be seen as a foundry for internal and external chip designers. Intel product groups already use TSMC extensively; there should be no difference to the product groups between IFS and TSMC if IFS becomes a competitive foundry. If the design groups were spun out and had to use foundries exclusively they would figure it out. I suspect even the CPU teams have already figured it out, because they're using TSMC too. Some Intel teams have used foundries for decades. IMO it will also improve CPU design efficiency, assuming they execute well on the chiplet (tiled dies) strategy.
The desire and financial need to execute the strategy well even though there will be many challenges and it will be very expensive. The business case is more compelling now, so the level of desire will be greater.
Arthur Hanson
Well-known member
What is the packaging cost and performance cost of going to chiplets over a single complex chip?Very ironic.
That's like asking "how long is a piece of string?"...
Huge chips pushing the maximum reticle size like Nvidia and Intel and Xilinx used (700mm2 or so) have the advantage of being monolithic so not needing any inter-die links, but the yield is *terrible* so the cost per mm2 of silicon is high -- and this cost dwarfs any adder for multi-die chiplet packaging, even if silicon interposers are used. There's also the issue that if you want a family of parts you need a family of chips, and have to pay the development and mask costs for each one which are huge nowadays. With chiplets TTM can be quicker, and spinning out new family versions only needs new interposer/package which is *way* quicker and cheaper than a new chip design.
Depending on what the application is, the inter-die links with chiplets may have almost zero, a small, or a large penalty in performance -- but for big devices a single die will always be more expensive to manufacture, and harder to ship in volume because you throw many bad die away for every good one, some of the biggest chips have single-digit percentage yields.
So if your application values performance over everything else including design/manufacture cost, flexibility, TTM, and high volumes, and it will fit on one big die, and you can sell the end result for a high enough price, then single-die is the way to go. This is what Intel used to do, and why their market share is shrinking.
If cost and flexibility and TTM and big volumes (high yield) are more important, or if it's just too big for single-die, then chiplets are the way to go. This is what AMD have been doing, and why their market share is growing.
There are other advantages to chiplets such as being able to mix different technologies optimised for different functions (e.g. CPU/DSP core vs. SERDES/IO) and not having to make or buy all the (increasingly expensive) IP you need on one bleeding-edge process.
As the cost/effort of each chip development has gone up with each process node, the market is moving more and more to chiplets to keep cost and headcount under control, especially where you want a family of parts not just one.