Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
SoftBank Considering Possible Sale of Arm Holdings as Apple Gears Up for Arm-Based Macs
- Thread starter MacRumors
- Start date
- Sort by reaction score
Guys guys: this is SoftBank! Go read about SoftBank and WeWork[1] and then you'll understand that the rule, financially speaking, is: do the opposite of whatever SoftBank is doing. 😅
[1] https://www.cnbc.com/2020/05/18/sof...olish-valuation-falls-to-2point9-billion.html
[1] https://www.cnbc.com/2020/05/18/sof...olish-valuation-falls-to-2point9-billion.html
Mac sale volume is very low compared to the volume of ARM licenses/products for mobile. Apple switch does not do much for ARM business.
As Seoras and JPack noted, Apple just licenses the ARM ISA so they would have no need to own any part of ARM as a company. And since their license is perpetual, regardless of who comes to own ARM, it would have no impact on Apple's plans.
That would mean that Apple owned the development of ARM architecture for all the companies that use it. Apple is not known for doing anything like that. Also, this might raise anti-trust concerns. It is important for the world that ARM architecture stay available for many companies that use it. Apple would be unable to just buy ARM and keep it for themselves.
Don't forget Wirecard. $1 billion that's almost certainly gone.
My guess is that Softbank is investigating if there could be over betting for Arm Holding. After all it is strategically important for Apple (and to a some extent also Samsung). Imagine for a second that some group that could be only a storefront for Intel would have interest in it ? It could sell for much more than it's real value because of it.
Everybody knows Apple is going to be bringing in income, so it's been priced into the value of the company.
It's speculative income. What if Apple's ARM switch flops? What if the first gen is poorly received and there's a delay? It's often better to sell when there's people who believe it can wildly succeed, rather than after the firm results are available.
I don't think ARM makes any money from it. Apple has a perpetual, fully paid, ARM license. I'm sure that cost a lot of money at some point, but Apple is free to make as many ARM chips as they like without extra payment.
That's not how it works. There's two licenses in play. Apple has an architecture license, which allows them to design cores. Other companies can buy licenses for cores that ARM has designed. Those can be time-limited or perpetual. But after that, all licensees pay a per-chip royalty for every device actually manufactured.
As you'd expect, the majority of ARM's revenue is from "royalties" not "licenses".
To complicate things, in some cases, the royalty is paid by the fab, not the customer. Obviously the cost is being passed on, but on paper it looks like the end customer isn't paying a royalty.
Last edited:
Apple's not stupid enough to do that.
There isn't enough popcorn in the world to fully enjoy the antitrust actions that would happen if Apple tried that. I don't think even Apple has enough cash to "lobby" Washington to get away with it.
That debate should have died a long long time ago. Intel processors have had a RISC core and 386 to RISC-like interpreter since the Pentium Pro.
[automerge]1594678306[/automerge]
Yep Newton and eMate used ARM
MessagePad - Wikipedia
eMate 300 - Wikipedia
And that interpreter has grown to be larger and more power-hungry than the actual execution cores themselves. And that interpreter is so complex, it's caused multiple performance-sapping bugs (Jump Conditional Code, Loop Stream Detector). And running instructions through the interpreter is so expensive, we have extensions that try to run instructions once for multiple pieces of data a.k.a AVX and AVX-512.
So no, that debate is more relevant today than ever.
I guess that must be why ARM has created it's own SIMD instruction sets, right?
That shows exactly the differences. Both modern ARM and POWER use 128 bit wide vector instructions, whereas Intel went 256 and now 512. You can't eliminate the front end, instruction decode and the like on any processor, but it's evidence of what the x86 tax costs.
As others have noted, whomever controls ARM Holdings doesn't really impact Apple at all.
The ARM SVE ISA supports up to 2048 bits, although such long vectors really only make sense for specialized HPC processors.
You could maybe make an argument that AVX-512 is a bridge too far for general purpose computers (in terms of transistor count/benefit ratio), but AVX and AVX2 are hugely beneficial for many things such as video processing, AI, packet processing, games and many other things. SIMD is about parallelization and memory transfer optimization, not about some imaginary "x86 tax".
I read it was actually money from the sale of ARM that allowed Apple to buy NeXT and bring back Steve Jobs.
There's nothing magic about SIMD, other than in marketing.
Take AVX-512, it's 16-wide, meaning there's 16 parallel EU/ALUs. If you could, having 16 independent EUs is equal or faster than doing a vector of 16.
The simple case is what if you have less than 16 elements? I could simultaneously run a set of 12, and an independent set of 4, perhaps doing a different instruction. With SIMD, you're stuck doing this as two separate instructions.
Entire point of SIMD is to amortize per-instruction costs, both runtime and hardware, over multiple pieces of data. Taken to its extreme, you get GPU computing.
And that goes back to the original point. The x86 tax is so expensive, it makes sense to force the same instruction over 16 pieces of data, whereas only 4 makes sense with the costs on modern RISC architectures.
Last edited:
Many do not understand the P.A. Semiconductor role with Apple's ARM Architecture . I agree with your assessment.
I do not agree ARM is not static, it is still evolving. How it evolve is strategically important in the long run.
Last edited:
A license for a particular design is perpetual, this does not include upcoming design, which you have to license separately.
[automerge]1594684092[/automerge]
Apple also has licenses for ARM IP cores, where they have to pay royalties.
Apple was even one of the founders with Acorn Computers. And it was Apple that asked them to found this company as Apple wanted an independent party to work with them on the CPU for the Newton (the first failed iPad, which had stylus).
Oh Apple... Maybe you shouldn't have sold your foundational stake in ARM? I don't think a lot of people realize this these days, but Apple was essentially one of the founders of ARM.
en.wikipedia.org