Apple’s M2, the successor to the M1 and the SoC variant expected to ship in higher-end systems, is said to have now taped out on TSMC’s N5P process. N5P is said to be a small improvement over TSMC’s N5 node, offering an additional 10 percent improvement to power consumption and a 5 percent performance improvement over N5.
This news comes from Asia Nikkei, but we have to note an error in the story that’s been introduced by either the authors or Google Translate, since the said error is being propagated online. The Asia Nikkei story (or at least the GT version) claims that TSMC is manufacturing “chipsets” for Apple. This isn’t the first time this error has popped up recently, so it’s time to set the record straight. A “chipset” consists of the components integrated into an SoC to handle data transfer between the CPU and its various peripherals. We don’t normally talk about chipsets with advanced, highly-integrated SoCs, because this silicon is integrated into the SoC itself. Calling the entire SoC a chipset, however, is inaccurate for the same reason it’s inaccurate to refer to a PC chassis as the “CPU.” Apple’s M1 and M2 are not chipsets. They’re chips. The chipset is essentially invisible.
According to Nikkei, the M2 will be available as soon as July in future MacBook Pro systems, and will likely ship in systems beyond the MacBook thereafter. In June 2020, Apple declared it would transition to its own silicon, and that the transition would take about two years. This is similar to the timeframe the company announced when it moved from PowerPC to Intel back in 2005, but it only took Apple about 18 months to make that transition at the time.
As we’ve discussed, Apple is positioning the M1 in a way that effectively mocks the x86 practice of offering dozens of products across a wide range of SKUs and price points. The M1 is the Apple CPU of choice for a wide range of systems and the M2 (M1X would also be a plausible name, given Apple’s past naming conventions) could also follow suit. Right now, the M1 is available in the MacBook Air, MacBook Pro, Mac mini, and 24-inch iMacs. The M2 could debut in upper-end MacBook Pro systems and a new line of 27-inch iMacs.
If Nikkei is right, and the M2 is built on N5P, it implies the CPU might include other architectural enhancements over and above the original M1, rather than simply being an expanded version of it with more high-performance cores. Apple hasn’t really clued us in to the ins and outs of its new silicon development strategy, so one possibility is that the company intends to iterate new desktop and laptop chips yearly, just as it does with smartphones and tablets. This would imply the introduction of specific tiers, where the M1 family would address some products, the M2 others, and Apple would have to then increment the branding for each family per year, as new hardware is introduced.
The other question is whether Apple will lead with smartphone processors or non-smartphone chips going forward. Apple has historically led with smartphones and introduced new iPad processors that were augmented variants of iPhone chips. The M1 is a scaled-up A14. If the M2 resembles the A15 but debuts first, it’ll indicate that Apple is shifting its CPU debuts to focus on higher-performance hardware. If the M2 is a scaled-up M1 and the A15 introduces new performance-boosting capabilities or features that the larger chip lacks, we’ll know that Apple is still leading with mobile and waterfalling features and capabilities into its higher-power, higher-performance chips.
ET expects a future M-class chip intended for the Mac Pro market to offer many more high-performance cores than the current SoC does, but with the same emphasis on low power operation and high overall efficiency. Twelve to 24 FireStorm cores would be a reasonable target for any machine intended to replace the current Mac Pro, probably paired with four to eight high-efficiency cores.
If you care about CPU performance, the CPU market is looking brighter today than it has in years. AMD and Intel will either demonstrate themselves capable of matching the improved performance per watt Apple has built into the M1 and presumably its successors, or they won’t. x86 will either once more rise to the challenge, cementing its rule as the reigning champion of the personal computing industry for the past four decades, or it’ll prove incapable of doing so, and ARM will replace it. But the M1’s improvements are real. Performance and power tests show Apple’s CPU is faster, in performance per watt and performance per clock, than the x86 CPUs it competes against in relevant price bands and products. For now, the direct competitive threat is limited because Apple only competes in certain markets. But Qualcomm bought the high-performance ARM CPU company Nuvia last year and intends to reenter the laptop market with a new, custom high-performance architecture by 2023.
CPUs function as the “compute unit of last resort” in the modern world because we increasingly seek to improve efficiency by offloading work to accelerators, rather than performing general-purpose computations in software. CPU performance improvements will always be useful for speeding up the most difficult code.
Apple, TSMC On Track to Move 3nm Into Risk Production by the End of 2021
TSMC's 3nm node is on-track for risk production by the end of 2021, with volume shipments and Apple as a launch partner in 2022.
Nvidia: GPU Production Will Be Substantially Constrained for Most of 2022
We're coming up on a year since Ampere launched, but anyone hoping for a new graphics card at MSRP in the near future may be out of luck. According to Nvidia CEO Jen-Hsun Huang, GPUs will remain constrained for the "vast majority" of 2022.
Microsoft Has Asked AMD to Boost Xbox Series S, Series X Production
Microsoft's feeling the heat, in terms of console shipments — so much so, it's asked AMD directly for any help it can provide.
Report: Packaging Issues, PS5 Demand May Be Hurting TSMC Production
The hardware shortages currently hitting most of the PC market may be caused by a shortage in a necessary component in chip manufacturing, not low yields on TSMC's 7nm node.