It's entirely possible that Apple will switch to RISC-V, not now but by 2030, because Apple always plan years before these switch. And consider how much experience they're on Switching Architecture, it's not a big deal for them for another switch.
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.
Apple Possibly Exploring Open-Source Alternative to Arm Architecture
- Thread starter MacRumors
- Start date
- Sort by reaction score
All true, but there’s little reason fro them to consider RISC-V, at least not unless they make some pretty massive changes to it. It has some crazy design choices that make it bad for products like computers. Certainly no advantages over Arm. Things can change, but right now the only real RISC-V advantage is no license fees (which does not really matter to Apple, which probably pays very little for its license, if anything). And the lack of license fees isn’t going to help you when you get sued for infringing someone’s CPU patents, and then there’s no Arm or Intel or AMD there to indemnify you.
Of course you're waiting. ARMv9 was introduced in the spring of this year and isn't expected to be in devices until early 2022.
The price of a product is not determined by the cost of materials, but by what a consumer of the product is willing to pay.
Apple has used RISC based architectures for most of it's existence before switching to Intel X86.
They actually went back a few steps in the Intel transition as the old IBM-Moto chips were full 64 and the older Intel Chips were not!
RISC has more in common with ARM64 than X86-CISC based architectures so this is simply good forward planning
They actually went back a few steps in the Intel transition as the old IBM-Moto chips were full 64 and the older Intel Chips were not!
RISC has more in common with ARM64 than X86-CISC based architectures so this is simply good forward planning
Last edited:
No, apple didn’t use RISC for most of its existence. The Apple II and original Mac were not RISC. The first RISC machines they used were PowerPCs. They didn’t start that until 1994. The Intel machines started in 2006. So they used CISC from 1977-1994, RISC from 1994-2006, CISC from 2006-2020 (ish).
And the PowerPCs apple used were not 64-bit other than the G5, I believe?
I stand corrected.
From when I started dealing with Apple - early 90s
Still they have a long history of RISC based processors from Moto/IBM.
There were rumors of the Intel skunk lab in Apple years before announcing the switch.
Likely same thing this time around as they certainly have moved to own the whole process so not beheld to supplier constrains etc…
The last one took ~12 years though. From the acquisition of PA Semi (chip design company) in mid 2008 to the launch of the M1 for laptop & desktop chips in 2020 and corresponding software transition.
I’d be very interested to hear some more of your thoughts on “crazy design choices that make [the RISC-V] bad for products like computers”. I’ve come to trust your take on much about processor design through your posts over the past year or two.
There’s a good history lesson there. Had to do with IBM and their design. Ultimately x86 is very inefficient compared to ARM. Do some digging on YouTube you’ll find a good documentary on it.
Well, it ranges from little things to bigger. It’s only little-endian, whereas Arm is bi-endian. x86 is little-endian and PowerPC is big-endian, So it’s nice to support bi-endian so you can remain compatible with future cpu changes. Not a huge deal.
It uses a single instruction for jump, call and return. That’s a mess, and complicates the branch predictor, likely causing either an IPC hit or slowing cycle time. It’s also just “clever for the sake of being clever.” Which describes a lot of it - it takes four or five instructions to index into an array (vs essentially 1 - or 2 depending on how you count - for x86 or Arm). The poor code density will have some effect on effective IPC, either in the instruction fetch path or in the issue path. It’s terrible at coherency, which is necessary for multiple cores. It’s really best suited for single-core, out of order issue, in order retire, machines. Sort of like what we had in the mid 2000s. Again, they could change stuff (a lot), but right now it doesn’t seem to offer any advantages over Arm for “real” computers.
UPDATE: I’m reminded of two CPU designs from my past. RISC-V gives me the feel of these, which isn’t surprising considering Hennessy and Patterson are involved and they influenced me in these 2 designs.
In my senior year for my B.S. in E.E., we had to design a CPU to fit into an FPGA (which were not so big back then). I designed an ISA that was so “clever” about being easy to decode - saving a NAND gate here or a multiplexor there. Reusing one instruction for every kind of jump. No move instruction (like RISC-V!), just re-use other instructions and assembler macros. Oh so clever. Not very good for the real world.
Later on, my grad project involved a GaAs CPU design, using CML logic (all bipolar transistors), and wafers were small, yields were bad, we were using multi chip modules, and we had to keep everything small. So we took, essentially, MIPS, stripped out everything we could, and modified a few things. Yes, it was a Turing machine, but not a very good CPU. Super fast, though! (1 GHz in 1996, but IPC was around 0.7 or something).
Last edited:
It goes father back than that. KHTML from KDE's Konqueror was the initial codebase that Apple worked with.
Objective-C was originally developed by Brad Cox and Tom Love in the early 1980s, it was selected by NeXT for its NeXTSTEP operating system and popularized as a result. Regardless, who developed it? Not Apple. Certainly not at first, and it wasn't until *much later* that Apple got on board with it.
Apple did acquire NeXT, and certainly OS X (now macOS) was basically NeXT Step rebranded, but Apple definitively did not develop Objective-C.
MagSafe came back, sorta on the newer iPhone 12 devices I guess. I sincerely wish it would be resurrected as a laptop charging connector. Then again, I also wish Apple would bring back 3.5mm headphone jacks on their iPhone (they certainly retained it on the rest of their product line) but I won't be holding my breath.
OpenCL got superseded by Metal, didn't it? It certainly seems to be more performant: https://blog.neatvideo.com/post/metal-opencl
"RISC-V still does not have core components like SIMD or virtualization approved"
So for starters, there are reasons for SIMD not being a "core" component in RISC-V, that was an intentional design choice, not keeping up with the joneses of whatever SSE/MMX/Neon Intel/AMD/ARM feature parity pissing match that really shouldn't be embraced by a core CPU ISA, indeed, even for Intel, those are *extensions* not a "core" component, further reading on such subjects here:
RISC-V Vector Instructions vs ARM and x86 SIMD
Is old Cray-1 style vector machines coming back? What exactly is the difference between vector instructions and modern SIMD instructions?
Similarly, the RISC-V specifications do have a virtualization mode, perhaps you are referring to the proposed H extension as not yet being approved? As it stands, RISC-V already has more provisions than x86's "real" or "protected" modes, and again, more recent instructions such as VMD, VMX, VMD etc. Intel/AMD64 Virtualization instructions are *extensions* and not "core" components of the original CPU ISA.
"RISC-V specifications describe the current virtualization mode using the symbol V . If V=1 then the system is currently in guest context, otherwise it is in host context (either supervisor or user mode). The RISC-V H extension introduces a full duplicate of CPU state: one copy for the guest and one copy for the host."
These are not bad things! From my vantage, that shows that the designers of RISC-V have been paying attention and are intentionally jettisoning old cruft, and taking a measured approach with what they want to integrate into their specification. Similarly, not having floppy disk drives is not a real detractor against a contemporary consumer computer in 2021, even if it might have been nearly unthinkable in the 1980s.
RISC-V also does not specify an FPU, but again, if you track research in academia advocating for posits/unum as alternatives to IEEE 754, a standard from 1985, I think we have maybe *just maybe* </saracasm> come a LONG long way since the mid 1980s and do not need legacies baked into a 21st century CPU ISA which began in 2010.
It may make a lot more sense to facilitate better alternatives which are already proposed (and some implemented, in free/libre open source github repositories for posit/unums as well as one firm even offering silicon based implementations) than to drag the 20th century legacies behind us.
Making spurious claims that RISC-V's omission of some things which 1970s and 1980s vintage CPU ISAs have kept dragging along with them, seems disingenuous to me or perhaps misinformed or a really superficial read, from the sort of perspective that maybe can only parse a list of features and see if there is a check mark next to them rather than evaluate whether the check box should even be checked in the first place given the place we are in space and time relative to other advancements.
I think it is OK to leave the past in the past, rather than make erroneous claims that a CPU ISA from 2010 is somehow at a disadvantage to its 1970s and 1980s predecessors, for omitting things which may no longer be particularly useful or relevant, or have better alternatives more worth implementing than suggesting "core" component omissions as somehow being necessarily relevant, let alone suggesting that implies some sort of inferiority.
RISC-V also does not handle fax documents nor RTTY natively. I am OK with that! I would be hella skeptical of any new CPU ISA which considered dated constructs part of its core ISA.
Last edited:
Ant to add to that. Apple have an advantage over the rest. Then can tie the OS to the chip, right down to the design schematics of the chip and build the os on top of that.
Hey, we don't even know what apple is cooking in their labs. For all we know, they could be combining both chip architecture to come out with something new. Keep in mind that they can integrate the os to the hardware. They can make their drivers more stable.
If ARM is so great, why have they been keeping us on x86 all this time since like the 80's - 68k AND PPC which is a cousin to ARM.. both are RISC and sadly, PowerPC did not do its job.. and now M1 is the new PowerPC like processor with a kick to it. CISC is bad.
This is really a smart business move for Apple honestly.
When Apple invests in an open-source project it usually does so in a way that allows them to assert control in a serious way. A lot of Apple technologies are open source project where Apple are one of the largest—if not the largest— stakeholder invested: CUPS, LLVM, CLANG, WebKit, Swift, Darwin, etc.
Helping make RISC-V a viable alternative to ARM would allow them a position of strength at the negotiating table with Nvidia, or whoever winds up controlling ARM, when it comes to determining licensing fees, and also allows them to modify and direct it in a way that aligns with their proprietary ARM implementations, that would make a transition virtually seamless should they need to make one. This is the start of their next 10-year plan of hardware and it fully fits with Cook's vision of owning and controlling the core technologies behind everything they make.
When Apple invests in an open-source project it usually does so in a way that allows them to assert control in a serious way. A lot of Apple technologies are open source project where Apple are one of the largest—if not the largest— stakeholder invested: CUPS, LLVM, CLANG, WebKit, Swift, Darwin, etc.
Helping make RISC-V a viable alternative to ARM would allow them a position of strength at the negotiating table with Nvidia, or whoever winds up controlling ARM, when it comes to determining licensing fees, and also allows them to modify and direct it in a way that aligns with their proprietary ARM implementations, that would make a transition virtually seamless should they need to make one. This is the start of their next 10-year plan of hardware and it fully fits with Cook's vision of owning and controlling the core technologies behind everything they make.
PowerPC: there was not a full line of chips available, as neither Motorola or IBM had any interest in pursuing lower power. I was designing PowerPC chips for Exponential to be used in Macs, at the time, and we were only focussed on high performance. That did no good to Apple when Apple cared about low power.
Arm: There have not been competitive performance Arm parts until now because nobody was interested in making them. Who would? Windows (real Windows) didn’t run on it, Linux did, but the market is tiny, etc. So until now, there was no economic motivation for anyone to make high performance Arm - really, Apple is the only one who would do it, because they control their own operating systems.
Arm is pretty great. Would I do things differently? yeah, some things. But in the end, the choice of ISA makes not too much difference, once you have decided between CISC or RISC. They all have their quirks. I’ve designed both CISC (Opteron, K6, Athlon 64, etc.) and RISC (Ultrasparc V, PowerPC x704, F-RISC/G) chips. RISC is better.
Apple was one of the three companies that formed Arm as a joint venture - they provided all the financing. They have a perpetual license. They are not worried about licensing fees, and there’s no chance Nvidia or anyone else could interfere with the rights they surely locked up from the beginning.
True but, I don't believe it's just that simple.
I believe there are ways for ARM's ownership to make changes or new versions of ARM that would fall outside of that license that could deny Apple access to continued advancements or squeeze them from for additional cash. Considering Apple's already rocky relationship with Nvidia I reiterate that I have to imagine this is about bargaining power short term and long term independence simultaneously.
And I think a lot of people made a lot of hey about Apple not buying ARM from Softbank previously, and used your point above to dismiss their need to. I think that's a simplistic way to look at it that probably doesn't tell the whole story.
Apple didn't buy ARM because it would've been an anti-trust regulatory nightmare for them to own something so central to the industry that their direct competitors all rely on, while Apple seeks to make their own chips that lead the industry. If they owned ARM they'd not be able to keep A-series and M-series chip designs to themselves without serious blowback.
By not owning ARM, but using it as a foundation to build their own thing, and then going all-in on an open standard they can eventually steer their own way, while keeping a lot proprietary, they don't have that problem.
Why would this be an issue? The ISA and u-arch are two different IP and are mutually exclusive, as far as I know.
Apple acquiring ARM would mean investing in two separate u-arch design or letting the world enjoy what they have achieved with their A series SoC by rolling the ARM core design team into their existing Apple Silicon team. Not likely for that to happen since Apple Silicon is Apple secret sauce.
You're thinking about it on a technical level. You're not wrong that they absolutely could technologically segment the businesses but from a public and regulatory perception mindset it would be considered an anti-competitive conflict of interests and it would be extremely easy to make an argument that Apple would be keeping the best for themselves and holding back their competitors even if they didn't do exactly that. It would be a huge liability for them.
And that said, for Apple strategically it would be bad business to invest in something they would need to maintain for the benefit of their direct competition and NOT keep the best for themselves.
Ergo, the only smart choice is to not buy ARM, but to position against it while also being its biggest licensee. That gives Apple the flexibility to exert some level of control without owning, and prepare for a future where that is no longer tenable, or it's more profitable for them to go somewhere else.
This is a smart long game position, and it's basically what they did to Intel.
I suspect that if they move off ARM it won't even be a transition that requires a recompile/migration like this one did. It'll be seamless to the consumer, for developers a few things will be deprecated a while beforehand, and then support will end for them with the new chipset, but everything will have been in place for years in advance. This is like a 8-10 year strategy.
After that what'll be left to control? Maybe Apple will start fabricating their own chips... haha