ARM Longevity

[Fomalhaut]

Mostly IO related. Only 2 external monitors on the M1 mini and only 1 on the M1 MacBooks. There seems to be an issue with some USB SSDs not getting full speed of 10 Gbps. Many issues with waking monitors from sleep. Those are the ones that I know of.

In my experience, there is an issue with every USB SSD not getting its full speed on the M1 Macs, even if they do connect at 10Gbps. I've tested 6 different external SSDs and seen drops of about 100-250MB/s (read speed is most affected) compared to a MacBook Pro 16.

If there is a USB SSD that works with an M1 Mac, and has less than 5% drop in the speeds it can get on an Intel/AMD computer, please let me know!

 

[Significant1]

So Apple could sit on the same version of ARM or even fork off of it into their own ISA and still deliver the performance. Now forking could have issues in the world of software compatibility but in theory they could do it.

I imagine Apple would very much like to implement the hardware mitigation of Spectra type attacks that armv9 will bring and I understand apple silicon is vulnerable to. My faint hope though is, that armv9 features is not completely news to apple and secretly they already have some of its features in hardware. But that might be a kind of fork, as you mention, and armV9 is all or nothing.

 

[Bodhitree]

In my experience, there is an issue with every USB SSD not getting its full speed on the M1 Macs, even if they do connect at 10Gbps. I've tested 6 different external SSDs and seen drops of about 100-250MB/s (read speed is most affected) compared to a MacBook Pro 16.

Apple have implemented so much new stuff in silicon with these new ARM processors that I think that a few bugs are quite likely to come to light. If read/write throughput on the USB bus is the only thing that has taken a bit of a hit then I think Apple have done well. Perhaps it’s a software problem in the driver of the new SoC.

What I think we can expect is that as the SoCs mature these kinds of problems will go away. But I doubt it will significantly impact the longevity of the early devices. I think it is more that because Apple have complete control of the software and hardware stack that they can be supported for longer, although of course future OS upgrades are going to introduce features that will make them slower. Still the hardware performance is so good now that I don’t see it being a problem.

 

[dmccloud]

ARM has little to do with M1 performance outside of the ISA having consistently sized ops - the big drivers are the die size and (moreso) the microarchitecture. The Firestorm cores in M1 and A14 are not only faster than Intel/AMD cores but also faster by a good deal than the rest of the ARM world. Having 8 decoders and being able to push 7+ instructions per core simultaneously is a big driver.

So Apple could sit on the same version of ARM or even fork off of it into their own ISA and still deliver the performance. Now forking could have issues in the world of software compatibility but in theory they could do it.

Apple licenses the ISA from ARM, but not the reference designs that others such as Samsung and Qualcomm rely on. For all we know, Apple's ISA license was already upgraded to v9. Changing to a new instruction set won't make an immediate difference in performance, capabilities, etc. because the software has to be rewritten/recompiled to take advantage of the ISA improvements first. Apple has already forked off of the base ARM ISA, and has basically built a superset of their own instructions on top of ARMs existing ISA. I believe some of those improvements are now part of v9 as well. Apple's performance advantage over competing ARM parts from Qualcomm, Samsung, etc. is due to a couple of factors. First, since Apple is designing their own SoCs from scratch rather than plugging existing designs into a die like Legos, they can design the chips to take advantage of USB4, Thunderbolt, DisplayLink, UMA, etc. Apple also has the advantage of designing the OS and first-party software from the ground up, so they can code the software to take advantage of the hardware and vice-versa.

 

[leman]

Apple licenses the ISA from ARM, but not the reference designs that others such as Samsung and Qualcomm rely on. For all we know, Apple's ISA license was already upgraded to v9. Changing to a new instruction set won't make an immediate difference in performance, capabilities, etc. because the software has to be rewritten/recompiled to take advantage of the ISA improvements first.

It’s the same instruction set. ARMv9 just adds some things that were optional in ARMv8, plus some new features (which seem to be more relevant to servers anyway, but thats just my opinion).

Apple has already forked off of the base ARM ISA, and has basically built a superset of their own instructions on top of ARMs existing ISA.

I really wouldn’t say anything like that. The only Apple ISA additions that we know are the matrix coprocessor instructions. They are well isolated from the regular ISA , have virtually no interaction with the regular processor state and are not accessible to the user.

 

[quarkysg]

Regarding whether Apple’s SoCs uses ARM v8 or v9 ISA, I think it’s a moot point really. I don’t think programmers code using the ARM assembly? Any software build for Apple SoCs will be using Apple’s compiler and linkers, tho. open sourced software may be using other compilers? v8 or v9, it’s really transparent to most folks.

ARM will likely stay for a long while, as at least Apple is already all in with ARM tech. for the next 10 years at the minimum, IMHO.

 

[cmaier]

I just worry that when Apple very soon adopts armv9 my m1 will be like my core duo when core 2 duo arrived with 64 bit support.

Armv9 doesn’t add a whole lot - it’s a much smaller change than previous revs. I wouldn’t worry.