Because Apple’s self-developed M1 replaces Intel processors, Microsoft Windows On Arm seems to be a bit of a real game, and Arm server processors seem to have some gains in the market, so the view that “Intel is bound to rebuild the Arm processor product line” is again Another hype (although we have good reason to believe that 87% of these types of articles are typed on the Mac). And the arguments are generally over-contextualized, and the success of Arm’s beach grab is so light-hearted, is it pretending not to see so many sages and martyrs who have crossed the river over the years (like the Qualcomm Centriq whose bones are still cold)? Do I need to write another article “Arm Server 10 Years of Struggle” to pay tribute to the market pioneer who died heroically?
The author has previously written an article to analyze the possibility of AMD restarting the K12 plan and attacking Arm servers. From the perspective of the server market, I re-introduced the chronicle of the 20-year x86 server war between Intel and AMD. Now the author examines and analyzes the possibility of Intel doing this from a more alternative angle.
First of all, Intel’s two worst decisions in the past 30 years are all related to DEC:
The server market does not continue the accumulated achievements of Alpha (including the FX32! binary code translator that allows Alpha systems running Windows NT to execute 32-bit x86 Windows applications), and bites the bullet with HP to develop the IA-64 instruction set and Itanium processing It also planted the seeds for AMD to take advantage of the situation.
Due to the “x86 Boxer Rebellion”, the mobile processor market has inexplicable confidence in the so-called “huge IA software ecosystem”. It is believed that x86 processors can meet the needs of “all markets”. DEC StrongARM’s XScale was sold to Marvell, replaced by “Atom”, an attempt to attack every conceivable emerging market, followed by a tragic 10-year “failless” journey. After all, the x86 instruction set architecture and Intel’s business model have fundamental weaknesses.
Every time the author thinks of the Wireless MMX instruction set dedicated to XScale released by IDF in the fall of 2003, I feel a little sigh.
There are different opinions on why Intel voluntarily abandoned the Arm processor product line with the best performance at that time, but generally it is nothing more than “lack of matching high-quality baseband chips”, “users in the pre-smartphone era did not pursue performance so much” , “high power consumption”, “expensive”, “smartphone market share is too small” and so on.
In 2007, the year after Intel sold the XScale, as we all know, a mobile phone called the iPhone came out, which opened up the vigorous development of the smartphone market, and the above five problems were gradually solved.
The late Intel’s fifth CEO, Paul Otellini, once said when he left office that “refusing to produce chips for the iPhone was the worst decision.” Pat Kirsinger, who recently returned to Intel’s eighth CEO position, also revealed this on the show. This is Intel’s biggest misstep, and it also clearly states that “OEM Apple chips is one of the main goals” in the IDM 2.0 strategy.
It is worth mentioning that Pat Kirsinger, who was Intel’s first chief technology officer, is not a member of the x86 Boxer, just like Mike Fister, general manager of the enterprise platform business group who jumped to the EDA factory Cadence earlier. One of the big supporters of Itanium. When he first left Intel and moved to VMware, he was probably unwilling to Intel “release” Itanium, and once said to the media that “Itanium actually made money”, and Intel was forced to issue a press release to “refute the rumor.”
To put it bluntly, not everyone in Intel likes the x86 card so much, not to mention all the senior executives and practitioners in the processor industry, who are not the private disciples of the two contemporary RISC masters David Patterson and John Hennessy? Remember the saying “x86 only loves those who made it”?
However, we also know that it is still very difficult for Intel to catch up with TSMC in terms of process technology and service level (don’t forget that the essence of foundry is “service industry”). In addition, the smartphone chip market has long been saturated, and the mobile phone brand has a long history. The three big ones (Apple, Samsung, Huawei) have all developed their own chips, and the rest have been eaten up by Qualcomm, MediaTek, and Spreadtrum. Foundry may be too profitable for Intel investors to accept. Even if Intel does its best to develop a world-leading Arm processor and surpass Apple, Apple is unlikely to go back and give up its independent research and development team.
Back to the topic, if Intel wants to return to Arm instruction set compatible processors, what are the incentives? The author can roughly summarize the following points:
If Microsoft is serious about operating Windows On Arm, it will allow Apple’s M1 transfer method to be replicated in the Windows ecosystem, creating an Arm individual that is “cheaper and can execute existing x86 applications at high speed through emulators”. For computers, unless Intel has absolute certainty, by returning to the rhythm of the pendulum, and having the ability to maintain the advantage of performance and energy consumption at the micro-architecture level, Intel will have to follow up with the introduction of Arm processors that are compatible with x86 platform pins, and even It has to imitate IBM’s PowerPC 615 (all-in-one PowerPC and x86), and spare no expense to launch a product compatible with two sets of instruction sets, so that the “paradigm shift” can be changed from the left hand to the right hand.
At the moment when cloud giants are developing their own chips, Intel wants to ensure the hegemony of the data center market. There is probably only one way left: quickly customize the chips they want for customers and provide production services. This may be the real purpose behind Intel IDM 2.0. And the difficulty of having to re-invest in Arm instruction set-compatible processors: the long process of developing new products for 5 years, difficulty in Time To Market and quickly responding to emerging applications, are the biggest chronic problems of x86 processors.
The above is the author’s “most optimistic” personal speculation. Intel must have enough research and development energy to develop a top-level Arm instruction set compatible processor, but “business” and “politics” are still far more important than technology. Microsoft, whose business focus has shifted to cloud services, is a huge question mark whether it is necessary to rebuild a new PC ecosystem. It’s one thing if the technology can do it, it’s another if it’s profitable to do it with a lot of effort.
In addition, the ubiquitous industrial computers and IoT applications are almost exclusively dominated by x86, and they are also the most conservative customer group (no wonder Intel’s IOTG supply period has been extended from 7 years to 15 years, and AMD’s Embedded now has 10 years), As a result of Microsoft and Intel taking the initiative to destroy the x86 ecosystem, the final beneficiaries are probably nothing more than Nvidia, RISC-V systems and open source software camps such as Canonical. Neither company could have thought about this possibility.
Regardless of how the situation evolves, let the world’s largest processor manufacturer get rid of the burden of x86 and make every effort to develop an Arm instruction set compatible processor, how will it make people’s eyes dazzle, or repeat the mistakes of Itanium, it is quite impressive. Looking forward to–if this happens.
Siemens (formerly Mentor) EDA/IC Design Webinar on May 28
The Links: FF100R12KS4 SKKH460/22EH4 IGBT-MODULE