One question remains.
"Gelsinger pledges to make Chipzilla great again"
Is he promising to build a wall, and make the Chinese pay for it?
Intel is cranking up its research spending to fix past mistakes, catch up with and overtake the competition, and build a foundation to grow in future. The US giant spent $15.19bn on research and development in fiscal 2021, more than 20 per cent of the company's $74.7bn revenue. That was about a 12 per cent increase from …
It's the accountants with their (penny decimals set to four digit) Excel sheets cutting corners to reduce overall costs by any means.
In their minds, quality is an excess that can be removed, They are the reason modern corporations outsource IT, why todays burgers taste like cardboard and why services have user fees.
Intel accountants felt that cutting back on R&D would improve the bottom line and damn the consequences.
HP still makes lots of quality products, like printer ink... and eh...
I actually saw first-hand what Carly did to HP. When she had ambitions for the US presidebcy, she must have thought 'If I can run successful businesses in to the ground, a whole country should be my bext challenge'. (She use to preside over loose-end as well before she came over and messed up HP).
As did Apple, GM and a host of other companies. Blame Jack Welch and the cult of shareholder value that has convinced too many that trilemmas (good, fast, cheap) don't exist. But also blame loose monetary policy for favouring fiancial engineering over the real stuff.
Intel has a great track record, but the market now is much more challenging because it's both more technologically and geographically diverse. Not only is future growth in x86 limited, but TSMC, Samsung, et al. have been really busy developing their our process and engineering IP, with the Chinese catching up fast. This means you can't just buy the competition or poach the best chip designers.
I'm sure it was noticed that AMD's recent renaissance neatly aligns with the rise of their "technical CEO" in Lisa Su - I hope Gelsinger can do similar for Intel.
Being vertically integrated, Intel has a riskier path ahead - but with that risk comes the potential to reap greater rewards.
There seem to be some promising noises about their return to the discrete GPU market - currently an over-priced duopoly constrained by TSMC.
If they can nail that (or get close enough) and then crank out the wafers 'at cost' - they'll be untouchable as they can pick their retail price by just adjusting some internal budgeting to their desire to destroy their competition.
Maybe 5 years of crushing - before they decide to reappoint a bean-counter and the cycle will continue.
That's a vague ask. Competent in what? Accounting? Engineering? Project management? Sales?
The bottom line is that technical companies need to ensure that technical people are at the executive level. It is very easy for a sales type to advertise his metrics. much harder for a technical type. A company must make conscious effort to recognize and grow these talents.
Intel's best chance of success would be combining an ARM processor and an Intel Processor in one little-big hard wired form factor.
Intel are never going to beat ARM at a particular fabrication die size for speed, ARM will always win out, but by recognising and routing ARM instructions and Intel instructions to dedicated chips on a single SoC die, it could combine the best of both worlds and win against the likes of Apple's M1 ARM chips through better backwards cross compatibility.
Rosetta 2 is good, but it's still not as good as a dedicated Intel chip for running X86 code. But then again, maybe the future is the likes of a virtualised software stack for X86 code on an M1 or equivalent, and Intel's X86 days are already numbered against ARM.
It's certainly not going to be easy for Intel to crunch those Apple M2, M3 benchmarks at 3nm die size heading towards 1nm die, going forward. Apple are showing their metal on the hardware chip design front, has to be said, but they are heavily reliant on TMSC for all of this, and preventing their best from jumping ship.
I think you're making a CISC/RISC argument there. However I think Intel has made some noises about RISC-V - the soft CPU offered for Intel FPGAs has already moved over to RISC-V.
As a RISC fan myself, I think it's fair to say that Intel won out with CISC due to their process technology being ahead of everyone else's - and the economies of scale from serving the PC market.
However I have to admit that Intel's RISC efforts, from the i860 to Itanium, have not been crowned with glory. Itanium may have been faster than contemporary Intel x86s on floating point heavy workloads, but that wasn't an inherent limitation of x86 as AMD demonstrated with the Opterons.
So despite my dislike of x86 as an architecture, I think the size of the ecosystem and Intel's economies of scale mean that, on an equivalent process node, there won't be massive differences in system-level performance.
The thing is, RISC originated at a time when there had been a radical increase in memory, but not CPU performance - so RISC does well when the memory hierarchy is fast, allowing a high CPU instruction retirement rate usually gained by higher clock frequency. Now with the practical end of Dennard scaling the RISCs run at pretty much the same frequency as CISCs, negating that advantage.
When it comes to power efficiency I'm not so sure, in principle a simpler design with fewer transistors should be more efficient...
When Intel ruled, and Pat Gelsinger was the CTO, x86 was on a growth path, PCs and on premises servers.
Now PCs are being replaced with ARM powered devices and the cloud has consolidated x86 servers with some of them becoming ARM servers.
So what's left for Intel? Just make ARM chips for other people or come up with some radically more efficient/compelling architecture to compete with ARM? The latter is incredibly difficult to do and pull off commercially.
Unless you're talking about a specific model of Mac, or perhaps some models of Chomebooks, no 'PC's (as in Windows) are being replaced by ARM other than a few experiments here and there.
Quote: "So what's left for Intel? Just make ARM chips for other people or come up with some radically more efficient/compelling architecture to compete with ARM?"
Why do they need to compete with ARM? x86 primary market is Windows based systems at the home/office, and Linux & Windows in Servers. Windows is not going to be going from x86 to ARM any time soon, at least not on mass, simply due to the existing software base and tooling (compilers etc). ARM and x86 are not competing in the same market.
Also bear in mind ARM is barely more efficient that modern x86 CPUs when on the same manufacturing node and when measuring Watts vs effort delivered.
As an example, the primary reason why the Apple M1 ARM CPUs are more efficient than current x86 chips, is that Apple got first dibs on TSMCs then new 5nm node, something AMD will have access to in a few months time when the new Ryzen 7000 CPUs come out.
Also the current M1 Macs gain a lot of their power savings by using coprocessors, handling some functions in discrete silicon, separate from the M1, as the ARM processor just isn't that good at those specific tasks.
My point is that they need to compete with ARM because the long term prospects for x86 devices do not look good.
Long term:
corporate Windows laptops will stay on a very slow downwards trend as MacOS (ARM) and maybe Chromebooks (ARM) increase.
consumer Windows laptops will largely disappear (apart from gamers)
consumer MacOS (ARM) laptops will increase significantly
Android/iOS devices will get even more capable and become for many the sole platform
Massive x86 server consolidation from on premises DCs to the public cloud hyperscalers, which are being sold to the hyper scalers at significant discount.
Significant migration of x86 servers to ARM servers in the Cloud made possible through platforms, services and portable frameworks (.NET Core/Java) abstracting the need for x86 ISA.
I just think this is going to happen and Intel can't change it by spending more on R&D
“ Research and development pays dividends – figuratively – in the future and if Intel can convince investors it's money well spent, it doesn't become a drag on the stock," Moorhead said.”
It absolutely will drag the stock price down in the short term (the next 5 years). What he needs is full support of a board that understands what he’s doing and are prepared to ignore the bleating of shareholders incapable of thinking in terms longer than the next quarterly earnings call.
Unfortunately share trading has been reduced to what's essentually a casino, with influential people distorting the market by shorting stock.
Those people are not interested in the longevitiy of companies, they're only in it to make a larger profit the next quarter and they'll keep making profit off of the company's bloated corpse all the way down, just before it hits the bottom of the ravine as they themselves jump to their next victim.
This death spiral starts as soon as a company goes public,.
> MIT's del Alamo said any talk of Moore's Law ending is overblown as people have narrowed the definition. Moore's Law means different things to different people, but it still is a measuring stick to drive chip innovation.
If you have to change the wording to make it work then it is not a law and is already broken. Meaning different things to different people proves it is not a law.
That is pretty much the defn of 'law'.
Hey my Royal Society hommies.
It really looks like two massive objects attract each other with a force proportional to the product of their masses and inversely with the square of the distance between them. I think it should be my law.
My Newton why are you talking like that? This is the 18th century ?
I'm hoping that somebody writes a best selling musical about my fights with Leibnitz and I want it to be hip with da yoof.
X86's long standing advantage is it's pre-existing catalogue of software. Modular programming, if not platform agnostic programming have largely eroded that. Certainly in server world. Emulation is proving very good on ARM which might unlock legacy desktop migration too. Looking at you Apple.
That leaves you with hardware performance; which is ultimately about performance per kilowatt; and perhaps reliability, security & value-added service provision.
Intel IME is hardly an appreciated technology and often a burden. So is AMD AGESA. Make living with Intel better.
Make ownership of X86 slick, and deliver on competitive products with AMD and they will do well. 12900K is the new "desirable" desktop for many gamers. Not a market to ignore given sales linked to it.
Innovation for Intel by poking at marginal gains is also tough - they are involved in so many markets already it is hard to see where to expand or do new in terms of semiconductors. Getting more into software / service provision would be a sensible route. If WinTel isn't what it used to be; hook up with Haiku and make that deployable in real world (or other OS of choice).
Quote: "12900K is the new "desirable" desktop for many gamers."
Not really, granted it is currently topping out most benchmarks on average, but it's only single digit % faster than equivalent AMD, is quite expensive, and only really worth it if you're upgrading from an old gen CPU. I've you've got say an 11900K, or a 5800X+ it's just not worth it.
Also the new updated version of the 5800X, the 5800X3D is due out soon, which is aimed specifically at gamers, and is expected to take back top spot. Then we have the new Ryzen 7000 coming later in the year.
So Intel might have top spot for a month or two, but it's not going the last.
Competition is good :-)
Mr. Gelsinger technical background and engineering contributions have been as a computer architect. Therefore, I find architecture innovation, manycore scaling and better exploitation of parallelism conspicuously missing from his recent narrative.
These issues used to be his focus when he was CTO and are as relevant in 2022.
His exclusive focus on advancing semiconductor technology makes me wonder.
For anybody with real money, the profits are in surfing the stock market where YoY returns of well over 10% have been been possible for the last few years. Only because the treasury injected so much liquidity specifically for stock speculators to encourage speculation. The biggest downside of that it is harder for manufacturing to attract investment, unless it is for the purpose of making a quick buck on a fire sale.