Three years ago Apple switched from Intel x86 processors to their own ARM based M series chips. Last month Microsoft and its PC partners released the Copilot+ PC based on a Qualcomm ARM based chip. The ARM technology has much lower power usage for doing a given amount of work. Intel is transitioning its business away from proprietary X86 CPU's to a foundry manufacturing semiconductors based on other companies designs similar to the business model of TSMC. This is also driven by a US national security imperative to manufacture vital semiconductors domestically and billions in subsidies to encourage this goal.
Yes. The Wintel cycle is coming to a close. Intel is well managed. Intel has led and devoted significant resources to open hardware+software standards benefitting the entire industry equally. They had a very good CFO who retired, Andy Bryant. They have made mistakes, but not as bad as GE. They are very strong on software supporting their customers. They have generally been managed by engineers, and as former head of the Labs, Gelsinger, the current CEO, has a deep knowledge of the business and the technology.
Intel made mistakes on the GPU business, on not adopting ASML's deep UV lithography, failed on wireless, and they wasted a lot of money on side businesses/internal startups they subsequently killed. They had big, unforgivable, design quality problems which were a contributing factor to Apple's rejection. Their sales group is fierce for which they have been criticized, including on benchmarksmanship.
Intel used its process technology only for internal products, that meant that the business needed high gross margins to finance the capital investment in the fabs.
If you look at the foundry business it is TSMC at about $US75B, Samsung $US3.4B. TSMC is well managed, like most Taiwan tech. TSMC has a lot of debt. Their gross margins are good. Intel likes to keep their balance sheet clean and pay dividends. I think Intel will be successful in foundry. I don't think TSMC's business is secure under Xi-thought. South Korea is not the most secure country in the long term.
To ARM vs X86, those represent the reduced instruction set computing - RISC and complex instruction set computing - CISC philosophies.
In computing you will have dozens to hundreds of process. Moreso with virtualization, containers, and the like. Processes wait in a priority queue and are assigned to run when they have the data they need, on an interrupt, and various priority measures as determined by the scheduler.
When a process runs, its registers are copied into the hardware CPU registers, and all of the memory it needs is populated, called a context switch, then the process runs, after, everything is copied back out to memory, and the next process comes in. In code, subroutines/objects start a context switch. All that copying context into the CPU registers and cache, running, then saving it back out takes time.
In RISC one of the theories is that you will improve performance by having faster context switching.
In CISC, the theory is you will improve performance by having floating point, large vector instructions, and instructions to speed up video, machine learning, and cryptographic processing.
Larger and faster on-chip caches benefit both architectures.
I'm not an expert today, but in general the ARM architecture has become more complex and the X86 has added some RISC-like attributes. Intel is also the master of hardware accelerators off to the side, particularly on Xeon. So the ongoing discussion in chip design is what do cloud data centers need, what are their evolution cycles, and what will they pay. That has led to cloud providers developing their own chip architectures.
Early iterations of Microsoft’s bolder ideas have tended to seem more like demos than fully realized products.
www.theregister.com
