On Mar. 8, 2022, Apple introduces the all-new Mac Studio with M1 Ultra. How it may be the 2nd death knell of the x86 empire, long live the ARM king.
Next transition
In WWDC 2005, Jobs confirmed the rumor of the 3rd transition from PowerPC to Intel processes. It took Apple 5 years to make OS X singing Intel/x86 to start a new journey in bringing the best computer to users by looking ahead instead of milking the current PowerPC products.
In 2019, Apple made another giant leap by transiting Intel/x86 to Apple M1/ARM processes. Once the stable notebook industry finally starts to dance in a very different tone. For example, the M1 chip helped fuel the 70% growth in Apple’s Mac revenue. Whereas, Intel’s inventory continues to increase in 2021 despising the global chip shortage. While Intel may still feel its “best year ever”, 1% growth in the revenue is hardly exciting.
But, this transition is not done yet. This March, Apple introduces Mac Studio taking desktop to a whole new level. Only this time, it signs ARM instead of Intel/x86. Just like iPhone & MacBook do. So, where to go next?
A mini-mill for the cloud
So what? Intel still the king of the host computing today. In Q4 2021, Intel’s CPU market share is 74% for overall x86 & 89% for server. But the question is what is the best High Performance Computing architecture by looking forward. Mac Studio & M1 Ultra have shade a light to that, even they are not really designed for “HPC”. However, Mac Studio & M1 Ultra are reinventing the desktop for power users.
It becomes one more reason for more applications to migrate to ARM instructions in Apple’s ecosystem. Furthermore, the competition pressure will drive Windows ecosystem to match, and the spillover effect will accelerate the migration for more apps. ARM has started to eat x86’s lunch even in server & HPC as the theory of Low-End Disruption predicts.
- Mac Studio becomes the 1st choice of professional creators because of performance, productivity & better UX. For example, M1 MacBook Pro not only last longer, but also run faster, including compiling WebKit & exporting 8K video.
- The creator section is the most profitable market. With the highest density of paying customers, app developers will prioritize their product & service development.
- With more professional apps on Apple machines, more users will pick Apple computers & be locked in Apple’s wall garden. Apple’s “unified app empire” will rule them all. Which only signs ARM going forward.
- As it does, more will start to reuse their ARM software inventories for other computation problems on the cloud/server. This is how Intel/X86 became the King from PC to server. And, now this is how ARM is going to become the new popular tone.
M1 Ultra — a lean innovation
The Machine That Changed The World is a fascinating book with a lot of great insights also relevant to the tech industry. It tells a lean innovation story of how Japanese carmakers make 4-cylinder engines beating their 6-cylinder counterparts. Japanese carmakers were late “high-tech” 6-cylinder engine development because of a forecast mistake. But, they managed to win back customers by adding every available “low-tech” upgrade to boot the performance of their 4-cylinder engines. Most of upgrades were simple, such as fuel-injection. The improvement was so good, even to a factor of two in some case. Many consumers were even convinced Japanese cars were now “high-tech”. What a lean innovation, turns “Low-Tech Weaklings to High-Tech Wonders”.
Similar pattern happed to computers. CPU is considered as the engine for a computer. There was a time, higher CPU’s clock speed is the key to boot performance. Soon, the diminish return kicked in as the clock speed reaching the heat & power limits. People started to add more core into a new CPU. In parallel, GPU & special purpose accelerators are also pushing the performance of heterogeneous compute forward as the diversity of computation workflow keeps increasing. Today, discrete CPU & GPU are still leading on performance in the expenses of power. But, connecting them effectively & sharing data across are limited by the external interfaces, and therefore the total performance per watt suffer. Adding another pair of cylinder is not the answer going forward.
Apple tends to focus on the value that users perceive instead of better specification of parts. Therefore, while some expected M2, a “6-cylinder engine”, Apple delivers M1 Ultra, two “4-cylinder engines” connected by UltraFusion, Apple’s custom-built packaging architecture. Which removes the bottleneck of the system performance instead of pushing the performance on the core part. This is a typical way to scale performance in the industry. With new packaging technologies, M1 Ultra connects 2 with lower latency, higher bandwidth & less power consumption by skipping the bottleneck of the motherboard. As, this is a known secret now, many will follow to complete.
Inspiration for smarter cars
According to Apple, M1 Ultra runs as fast as a high-end discrete CPU by consuming 100W less power, and as fast as a high-end discrete GPU by consuming 200W less power. In theory, M1 ultra may power the best computer in smarter cars today. Only, Apple has its own plan for smarter cars. More importantly, why will Apple sell its secret sauce to competitors?
Today, Tesla is using AMD CPU to power its infotainment system, and has to deal with its power drain. But, Tesla will not stand still. Tesla has designed its own HW3 chips for self-driving. Tesla also designs its own D1 chips and connecting 25 into a Training Tile. Sooner or later, Tesla will design it’s own “M1 chips”. Which will soon be powerful enough to compute self-driving & infotainment without draining too much power/miles. The question is HW4 or HW5. Even I wish HW4, but HW5 may be a safer bet. What do you think?
Full Disclosure
The opinions stated here are my own, not those of my company. They are mostly extrapolations from public information. I don’t have insider knowledge of those companies, nor a whatever expert.
