News Intel CEO Pat Gelsinger: I hope to build chips for Lisa Su and AMD

[bit_user]

What's the point of chiplets if you can't tap multiple suppliers (that is rhetorical, do not answer)?

LOL, okay!
: D

 

[Bluoper]

If intel is successful what do you think will happen to market pricing of processed silicon? There would be another bleeding edge fab for chip designers to use and all of those fabs would probably fight tooth and nail of some of those big customers like nvidia.

 

[thth]

AMD and Intel have a crosslicense, meaning one can use the other's tech. There hasn't been a real competition for years amongst them. The same with Lisa Lu "CEO" implant to appease the Chinese to buy American tech.

The license is only for the ISA (the x86 instruction set). x86_32(Intel) and x86_64(AMD) are licensed to each other. But the actual CPU design, microcode, etc are an entirely different thing which is proprietary and not shared between the companies. Anyone can read the ISA, it's actually public (but only those who have the license for ISA can make CPUs for it). What makes their CPUs different, for better or worse, is the implementation and design of the internal architecture, which they are not going to share.

 

[thth]

In business I never rule anything out completely, but unless Intel gets out of the business of designing chips for itself this seems more like a fairy tail wish than a real chance of actually happening.

It will depend on if IFS can provide a service as good as TSMC. Samsung is also an IDM that provides foundry services to it's competitors. Companies like Apple and Qualcomm have used their services when they were as good as TSMC.

 

[thth]

If intel is successful what do you think will happen to market pricing of processed silicon? There would be another bleeding edge fab for chip designers to use and all of those fabs would probably fight tooth and nail of some of those big customers like nvidia.

There would be no need for a fab price war on bleeding edge. With the incoming wave of AI, there is too much demand for leading edge chips and not enough supply even with all the new upcoming leading edge fabs. Legacy chips are where the oversupply would be.

 

[digitalgriffin]

I'm pretty sure Intel can buy an AMD CPU online, grind, polish, etch and scan it to see its structure.
And the chance they will pull an AMD and just reverse engineer whole chips is small since AMD can do the same to their chips and cry foul if they find out.

Intel would have additional advanced knowledge of performance targets to hit, but it seems like the large players in the industry already do have the basics.

It seems Intel just wants to make and sell as many chips as they can. And it seems like they are getting lots of interest for making them. There also seems to be a big push for very large companies to make some of their own custom CPUs, GPUs, etc. I bet Intel is trying to upsell those markets some IP as well in the form of letting customers add an Intel e-core or video processing block or memory controller, etc to their chips if they get a fee per unit made.

No! You cannot grind away layers like that and analyze the circuits. You can see individual transistors using electron microscopes with grinding but it's an incredibly slow, laborious, intensive process to look at a few. And seeing the specific set of transistors you are interested in is...near impossible.

 

[waltc3]

I'll believe Intel achieving parity with TSMC when I actually see it...😉 Intel--especially Gelsinger--has fallen into the habit of promoting and selling tomorrow's capabilities and products today, which really isn't much to write home about, imo.

 

[Roland Of Gilead]

Intel can build cpus better than amd any time. But for what? There are two major dogs on x86 market. P4 era amd Wins people still buying intel
Ryzen era first and second gen people still buying intel.
Ryzen 3000 amd better than intel people still buying intel...
When people stop to build with intel they will make better producs. For now profit max you can.

Genuinely, I have no idea of what point you're trying to make. If it's that Intel sell more CPU's than AMD, well we know that's right.

 

[-Fran-]

It would be really cool to see it happen.

As a matter of fact, it would be even easier to compare AMD's uArch to Intels if that ever happens, assuming process parity, where I believe Intel would still keep their edge, so AMD would never take it.

It still would be interesting for other products which aren't in direct competition though. The questions is, which ones? XD

Regards.

 

[bit_user]

If intel is successful what do you think will happen to market pricing of processed silicon? There would be another bleeding edge fab for chip designers to use and all of those fabs would probably fight tooth and nail of some of those big customers like nvidia.

Something to keep in mind is that much of their supply-chain is common. Things like:

• substrate
• photoresist & other chemicals
• photomasks
• equipment (ASML)

I'm sure I forgot some. I don't know exactly how much commonality there is, but we've seen news stories about things like a critical shortage in substrate affecting basically all fabs, as well as mask producers not being able to keep up with the surge in lithography masks being ordered.

The point is that just increasing the number of fabs doesn't necessarily mean total wafer production will scale up by exactly that same amount. There could be other bottlenecks. Still, having more fabs isn't a bad thing, as long as each can still do enough business to keep the lights on.

 

[rluker5]

No! You cannot grind away layers like that and analyze the circuits. You can see individual transistors using electron microscopes with grinding but it's an incredibly slow, laborious, intensive process to look at a few. And seeing the specific set of transistors you are interested in is...near impossible.

It isn't like they would have to be like TSMC is when they grind the vcache and CPU so precisely that the copper atomically welds when touched or anything.

But it would have to be at least a bit better than some youtuber using some homemade equipment, or the metallurgical stuff I used to do a long time ago with the spinning discs.

It's totally doable, just time consuming. I bet Intel has equipment in their packaging plants that could automate the process, much less their labs.

But whether they could is different than will or should.

 

[bit_user]

With the incoming wave of AI, there is too much demand for leading edge chips and not enough supply even with all the new upcoming leading edge fabs.

...so we're told. I personally haven't seen a good business case for how such production volume is going to deliver acceptable RoI for it to be sustained. In other words, we could be looking at another crypto-style bubble.

 

[bit_user]

Genuinely, I have no idea of what point you're trying to make.

In general, it helps to keep in mind that Amdlova's username is ironic.
; )

 

[bit_user]

it would be even easier to compare AMD's uArch to Intels if that ever happens, assuming process parity,

Yes, but... keep in mind that Intel process nodes of the same name aren't all the same!

For instance, Alder Lake and Raptor Lake both supposedly use Intel 7, yet we're told that Raptor Lake uses an improved version of the node. So, how much of the latter's improvement over the former is simply down to node improvements? If a single node can change that much, then what are the chances of AMD and Intel both using exactly the same iteration of it?

Granted, it should probably be more similar than Intel vs. TSMC, but wouldn't necessarily take the issue of process node completely off the table. Especially if AMD's design was oriented towards a TSMC node and then ported to an Intel one.

 

[JamesJones44]

It will depend on if IFS can provide a service as good as TSMC. Samsung is also an IDM that provides foundry services to it's competitors. Companies like Apple and Qualcomm have used their services when they were as good as TSMC.

True. However, Apple heavily reduced their reliance on Samsung once they became a competitor, in fact it was one of the main reasons Apple originally jumped from Samsung making their CPUs (or at least that's what their lawyers would have us believe).

Like I said, I never rule anything out, but I think Intel Foundry, as long as it's attached to Intel as a complete company, would have to be worlds better before AMD jumped onboard.

 

[Roland Of Gilead]

In general, it helps to keep in mind that Amdlova's username is ironic.
; )

🤣

 

[Peksha]

Intel predicts the imminent (apparently within 10-20 years) decline of the proprietary x86. This builds on the ARM/Risc-V pressure with many new independent IP developers. So they feel this weight and they need this new IFS strategy to stay afloat. Just like the GPU. Look at AMD - it can survive the abandonment of x86 right now - it has GPU, AI and so on. Intel is moving in this direction of diversification, and is doing exactly the right thing.

 

[bit_user]

Intel predicts the imminent (apparently within 10-20 years) decline of the proprietary x86. This builds on the ARM/Risc-V pressure with many new independent IP developers. So they feel this weight and they need this new IFS strategy to stay afloat.

I think you have it backwards. It's the rapidly increasing fab costs that have motivated IFS. They're increasing so fast that even Intel's current volumes are at risk of not being able to bankroll their continued fab R&D, buildout, and operation.

 

[digitalgriffin]

It isn't like they would have to be like TSMC is when they grind the vcache and CPU so precisely that the copper atomically welds when touched or anything.

But it would have to be at least a bit better than some youtuber using some homemade equipment, or the metallurgical stuff I used to do a long time ago with the spinning discs.

It's totally doable, just time consuming. I bet Intel has equipment in their packaging plants that could automate the process, much less their labs.

But whether they could is different than will or should.

AMD's grinding for stacking is nowhere NEAR what is needed for transistor level grinding.

This kind of work is only done when they look for failures. Subsystem test and diagnostics built into chips help engineers identify the point of failure and then they grind down to that point to see if it's a frabrication issue. This is how intel discovered there were "holes" in it's cobalt platting used on the power distribution plane. But it is STILL extremely labor intensive just to look at a few transistors.

If you don't believe me look here:

View: https://www.youtube.com/watch?v=TtuUANbaEFI

Look at parts 1 and 2 which involve prep. This is just for ONE cross section and there are millions of cross sections and not every attempt is successful.

 

[Peksha]

I think you have it backwards. It's the rapidly increasing fab costs that have motivated IFS. They're increasing so fast that even Intel's current volumes are at risk of not being able to bankroll their continued fab R&D, buildout, and operation.

This is true in the short term. Eventually nodes (and their costs) will stagnate, but architectural improvements will increase (agree with Huang and Keller))

 

[Co BIY]

I doubt it. AMD is one of TSMC's largest customers.

So is Intel !

 

[toffty]

AMD and Intel have a crosslicense, meaning one can use the other's tech. There hasn't been a real competition for years amongst them. The same with Lisa Lu "CEO" implant to appease the Chinese to buy American tech.

They do have cross-licensing agreements, for x86 and x64, but that doesn't mean they gave agreements for everything.

 

[Co BIY]

I'll believe Intel achieving parity with TSMC when I actually see it...😉 Intel--especially Gelsinger--has fallen into the habit of promoting and selling tomorrow's capabilities and products today, which really isn't much to write home about, imo.

That's because the real product (at the CEO Level) is Intel stock. Not some silicon gizmo. Stock (especially high-yield tech growth sock) is sold on future value not the present.

One advantage TSMC has is that it is not completely "Financialized".

One of Intel's advantages is that it is fully Financialized.

Some of the best scenes in Silicon Valley are with the main character and his new CEO agreeing to protect the integrity of their product. But what is the product?

 

[usertests]

Intel predicts the imminent (apparently within 10-20 years) decline of the proprietary x86. This builds on the ARM/Risc-V pressure with many new independent IP developers. So they feel this weight and they need this new IFS strategy to stay afloat. Just like the GPU. Look at AMD - it can survive the abandonment of x86 right now - it has GPU, AI and so on. Intel is moving in this direction of diversification, and is doing exactly the right thing.

Yet Intel is working on X86S:

https://www.intel.com/content/www/u...visioning-future-simplified-architecture.html

Never buy the ARM/RISC-V hype, they are fragmented disasters for PCs.

Intel needs as many customers as it can get for its fabs, whether it's ARM, RISC-V, analog sensors, whatever, so that it doesn't have to dump its own chips onto the market at cost just to keep the fabs operating. Rather than keeping Intel afloat, fabs will sink Intel if they don't use them properly and of course, slurp up as much free money from the U.S. taxpayer as they can.

 

[Eximo]

RISC tiles? I'm sure there are reasons that wouldn't work amazingly well, but would be a simple way to create cross compatibility.

Rosetta works pretty well too, so legacy stuff having to run a translation layer or emulated might not be bad.