AMD rumored to be working on an Arm-based Apple M1 rival

midian182

Posts: 9,745   +121
Staff member
Rumor mill: Apple has shaken up the processor industry with the arrival of its M1-powered Macs, offering excellent performance despite being a low-power chip. But could another Arm-based competitor be on its way? According to a new rumor, rival AMD is almost ready to lift the lid on its version of Apple's SoC.

The rumor comes from hardware leaker Mauri QHD (via NotebookCheck). They write that AMD is working on two versions of the chip—one with integrated RAM and one without. The prototype is "almost ready," apparently.

AMD has explored the possibility of an Arm-based chip in the past. Back in 2014, news broke that it was developing its own, custom 64-bit ARMv8 CPU core codenamed K12. And in 2016, it announced the K12 Core. The project was to be led by then AMD lead CPU architect Jim Keller, but the SoC was shelved. In May of this year, however, leaker Komachi Ensaka created an AMD roadmap that included a "K12 FFX" with a vague release date of between 2017 – 2022.

Interestingly, AMD boss Lisa Su will present at CES on January 12, 2021. We can expect to hear about the Zen 3-based Ryzen 5000 mobile chips, codenamed Cezanne, as well as the Epyc "Milan" server lineup and possibly RDNA 2-based mobile GPUs. We'll have to wait and see whether K12 gets a mention.

As with all rumors, take this one with a pinch of salt. But given AMD's history with the K12 Core, and how well-received the M1 Macs have proved, there might be more to it than mere speculation. With AMD chipping away at Intel's CPU dominance in the latest Steam survey, introducing an Arm-based SoC of its own could be a shrewd move.

Permalink to story.

 
AMD going back to start growing their ARM operation can only be good for them IMO. They had to shelve K12 before to focus on Zen, when the company was struggling. But with Zen proving to be just as power efficient as the best ARM cores, its easy to question if AMD even needs to make ARM chips.

Zen 3 Cores and RDNA 2 Graphic cores all on one die with IO controller and other accelerators, Along with on package memory. It would be a huge win for Mobile computing. Intel has been so lackluster in this area for last few years.
 
To me, that's how I see Lisa at the moment...

1607089058089.png

Sabotaging the x86 platform that's been feeding the company.
 
To me, that's how I see Lisa at the moment...

View attachment 87243

Sabotaging the x86 platform that's been feeding the company.
I actually think this is AMD being quite smart. X86 is doomed, risc is about to boom. Businesses need to adapt to stay relevant. Intel appear to be going after data centers, cloud compute and enterprise solutions and AMD seem to be going for OEMs.

 
I actually think this is AMD being quite smart. X86 is doomed, risc is about to boom. Businesses need to adapt to stay relevant. Intel appear to be going after data centers, cloud compute and enterprise solutions and AMD seem to be going for OEMs.

X86 while using CISC instructions, the chips themselves are RISC. They convert the operations. Intel and AMD started using RISC designs over 2 decades ago. They have plenty experience and really these days X86 isn't a huge deal. This is why X86 Zen 3 arguably has the best performance per watt of any commercial processor. But don't go thinking that X86 isn't RISC, every modern CPU is RISC. There have been plenty of upgrades to X86 that not all ops are CISC, but X86 compilers still uses CISC opcodes. So X86 chips do have die space that is dedicated to this decoding and honestly would be better to move off X86 at some point. But to say X86 is going to die any time soon is a lie. That being said, ARM isn't all perfect and moving to a open future platform like RISC-V would be best.
 
With x86 apps being emulatable I don't mind, Apple has already done a pretty good job already.
 
X86 while using CISC instructions, the chips themselves are RISC. They convert the operations. Intel and AMD started using RISC designs over 2 decades ago. They have plenty experience and really these days X86 isn't a huge deal. This is why X86 Zen 3 arguably has the best performance per watt of any commercial processor. But don't go thinking that X86 isn't RISC, every modern CPU is RISC. There have been plenty of upgrades to X86 that not all ops are CISC, but X86 compilers still uses CISC opcodes. So X86 chips do have die space that is dedicated to this decoding and honestly would be better to move off X86 at some point. But to say X86 is going to die any time soon is a lie. That being said, ARM isn't all perfect and moving to a open future platform like RISC-V would be best.
Sure, I didn’t meant to upset anyone, it was just an exaggeration. X86 isn’t doomed it’s just going to to become far less prolific in the future...
 
You may have the performance crown to rendering but an Arm chip can do rendering just as well if not better if it has a dedicated chip on it. This is the advantage of Arm, the customizability of it.
 
You may have the performance crown to rendering but an Arm chip can do rendering just as well if not better if it has a dedicated chip on it. This is the advantage of Arm, the customizability of it.

Yes but for arm to be just as feature rich it requires instructions, and arm is already becoming just as bloated as x86.
 
X86 while using CISC instructions, the chips themselves are RISC. They convert the operations. Intel and AMD started using RISC designs over 2 decades ago. They have plenty experience and really these days X86 isn't a huge deal. This is why X86 Zen 3 arguably has the best performance per watt of any commercial processor. But don't go thinking that X86 isn't RISC, every modern CPU is RISC. There have been plenty of upgrades to X86 that not all ops are CISC, but X86 compilers still uses CISC opcodes. So X86 chips do have die space that is dedicated to this decoding and honestly would be better to move off X86 at some point. But to say X86 is going to die any time soon is a lie. That being said, ARM isn't all perfect and moving to a open future platform like RISC-V would be best.

Problem with this are execution pipelines for x86 instructions. Both Intel and AMD have claimed that more than 4 cannot be implemented due to x86 CICS instruction set where instructions can be from 1 byte to 15 bytes in length. RISC instruction set has all instruction with a fixed length of 4 bytes in ARM64 case, and that makes ReOrder buffer and OoO pipelines much easier to implement when you know that all instructions are the same size. Apple M1 chip has 8 execution pipelines implemented and RoB at least 3 times the size of any x86 chip and that's what makes it very optimised and powerful.
 
With x86 apps being emulatable I don't mind, Apple has already done a pretty good job already.

All X86 apps cannot be emulated

"Some" X86 apps can be emulated

Even the Windows VM that runs on the M1 is just a worthless turd

Try running a real Windows O.S., like Windows XP in a VM on the M1, or the many apps that do not run correctly from a Virtual Machine even on Intel X86 hardware

It won't "ever" work!

Prove me wrong!
 
All X86 apps cannot be emulated

"Some" X86 apps can be emulated

Even the Windows VM that runs on the M1 is just a worthless turd

Try running a real Windows O.S., like Windows XP in a VM on the M1, or the many apps that do not run correctly from a Virtual Machine even on Intel X86 hardware

It won't "ever" work!

Prove me wrong!

Ah, a Windows XP true believer. LOL

Running an ancient x86 OS in a VM on top of an ARM architecture is hardly a typical use case. Emulation is a band-aid. But Apple has the leverage, with their complete control over the ecosystem, to compel developers to recompile their apps for ARM. You won't pick up all types of apps of course. Games likely won't be along any time soon.

I think AMD developing an ARM architecture in a parallel effort to their Zen CPUs is very smart. They've finally got some free cash flow that they can direct into R&D.
 
Not a single mention that Jensen bought ARM for NVIDIA back in September? AMD really picked the wrong chip after that purchase since they have no upgrade path.
Pending regulatory approval, and plenty of future regulatory anti-trust scrutiny. NVidia bought it for a new revenue stream, not because they thought they could bend it to their will and proprietary designs.
 
Why does everyone assume they will abandon x86/x64? There are dual module ARMv8 designs, and x86 cores combined with ARMv8 cores. (Even the M1 has a bit of x86 in its design for its Rosetta.)

AMD would be crazy to fully abandon x86, especially when their current 4xxx mobile processors are more than competitive with the M1. The 5000 mobile processors should be even faster and more efficient.

If AMD is clever, they will have a dual instruction/power processor with x86 and ARM.
 
Why does everyone assume they will abandon x86/x64?

I don't! :) I think expanding their portfolio (when the time is right!) and ensuring they stand on more legs, is actually a good move. I just hope they have a plan, and this ARM venture of theirs won't become a bottomless pit for cash. Though, considering what Ms. Su did so far, that's probably not the case :)
 
With x86 apps being emulatable I don't mind, Apple has already done a pretty good job already.

Apple have done that by making a SoC that is more CISC like, or rather more AMD/Intel like. A lot of the 'emulation' by AS is not emulation, there's actual dedicated hardware there running the x86 instructions.

It's impressive, but Apple don't have some "secret sauce". Microsoft have been doing purely emulation, which is only okay. AMD and Intel simply offer greater x86 compatability by having more dedicated hardware on their CPU/SoCs/APUs for x86 instructions. This is shown by AMD/Intel having '100%' efficiency, Apple in the 70% area and Microsoft in the 50% region.
 
Not a single mention that Jensen bought ARM for NVIDIA back in September? AMD really picked the wrong chip after that purchase since they have no upgrade path.

The deal has not been approved by regulatory authorities or the many countries involved. I for one hope it's killed off even it's by China for selfish reasons.
 
AMD still has very limited development resources compared to those of Intel. That's why their laptop cores are all just powered-down Ryzens, instead of having a second super-low-power core like their old Bobcat or Intel's Atom.
So I would expect them to either make a server part which is basically a Ryzen with only the instruction decode part swapped for one with the ARM instruction set, or, if they did develop a completely new chip, it would have to be some kind of chip for whith there was an obvious big market.
I'm not sure what kind of chip that would be.
 
The deal has not been approved by regulatory authorities or the many countries involved. I for one hope it's killed off even it's by China for selfish reasons.
Couldn't they just ignore China? Wouldn't that just mean that people in China couldn't buy Nvidia products any more?
Oh, wait. No doubt all video cards are assembled in China in order to be competitively priced. Their GPUs aren't yet regarded as such advanced, sensitive technology that they're not allowed to be sent into China, even though China is only capable of making 14nm chips itself, so their 8nm chips are beyond what China can make.
 
I don't! :) I think expanding their portfolio (when the time is right!) and ensuring they stand on more legs, is actually a good move. I just hope they have a plan, and this ARM venture of theirs won't become a bottomless pit for cash. Though, considering what Ms. Su did so far, that's probably not the case :)
The good thing is that any work they do on reduced power budgets, integrating big+little cores, integrating memory on the die... can and probably will equally benefit their x86 cores.
 
Back