[Tom's Hardware] Apple Considering Moving Macs Away From Intel Chips - Page 21

Thank the lord someone has got it.

Its not going the way of ARM. Its going the way of APU. I'm starting to think that many PC users aren't as proficient in technology as they assume

You do realize that both the Intel and ARM processors Apple typically uses, are APUs, right?

I don't think going to AMD for APUs is remotely as likely as an eventual abandonment of x86 all together.

AMD might be a good stopgap though.

I am well aware that ARM is an ISA, it's just not there yet, and no amount of money is going to get it there that much faster.

4-5 years, yeah I can see that. Maybe even 3. But if that's the case, the point of this article is moot. Anything can happen in 3-5 years.

Maybe.

It takes years of planning and preparation for such a shift, and these are some of the first prominent rumors on the topic.

The transition to Intel didn't happen for a full year after it was publicly announced by Apple (and this was still ahead of schedule), and had been in the works to one degree or another for at least five years before that (every version of OSX, from it's inception, was developed for Intel x86 processors as well as PPC), with the first real rumors cropping up somewhere in between.

You're correct. Most desktop workloads are integer. They're also extremely serialized and subject to "fast enough" performance (things like email, word processing, web browsing (only partly true as the JS spec requires that all numbers be 64-bit floats), or similar apps). The workloads that require lots of processing power are likely to be floating point and also to use lots of vector/matrix operations.

Coremark uses real-world algorithms, but is less compiler dependent than Dhrystone. It is the standard method of comparing performance much like Linpack is used for floating point. I don't know the scores for Loongson, but if China buys MIPS (or licenses all their IP like Broadcom just did) then they would have rights to use the integer unit in ProAPTIV. From a MIPS architecture standpoint, floating point is a coprocessor, so the next Loongson processor could use the current FPU and mesh interconnect while benefiting from faster integer calculations (the best of both worlds).

You're correct when you say that ISA and implementation can vary (this is architecture vs. microarchitecture). The ISA is simply a written explanation of which instructions a processor must support and which bits of those instructions do what. In addition, the ISA makes decisions that force particular general architecture styles and decide which data types must be supported, I/O support, coprocessor usage, memory and cache access and structure, etc (you can look at the MIPS IV ISA in PDF form here).

There are some liberties when designing the microarchitecture (often done in a register transfer language as the concrete microarchitecture is finalized), but the fundamentals that plague the x86 (and x87) ISA can't be fixed with microarchitecture changes (the point of the article you read). Use of 2-register notation, varying instruction length, complex instructions, no cache support, thousands of instructions, few registers, poor native I/O and coprocessor support, etc. The problems with x86 at the architecture level are astounding. The only reason why we use it today (rather than the Motorola 68000) is that it was the cheapest chip IBM could get and one of the reasons it was cheap was that it couldn't do what other chips could (and this carried over into the ISA).

Intel and AMD have great engineers who have done amazing things (with lots of money and time) to work around the kludge that is x86, but starting with a better ISA that doesn't need to be worked around means that fewer designers with less time and fewer tricks will still be likely to make a better product.

Edit: to add to the discussion that was going on while I was typing: If Apple could get an increase in ARM float performance, they would be capable of switching quite soon (all of the "creative" markets that use mac computers wouldn't be too hurt by decreases in integer performance; especially if that decrease were short lived and came with a decrease in price from not paying Intel $200-1000 per chip). The problem with current ARM chips (aside from a lack of 64-bit support and legacy low-power design choices in the ISA) is that they are laid out to decrease die size and power consumption. Change the arrangement and there's a great possibility that the same chips will get better IPC and higher clockspeed at the expense of power consumption.
Edited by hajile - 10/8/12 at 8:52pm

True, you've got a point there. The transition would be over many years. Honestly, having one codebase will definitely have it's benefits and ARM's scalability is amazing. I guess we'll see what the future holds but here's to longer lasting laptops.

Yeah, I agree. Just take a look at the amazing performance phones/tablets deliver. It's crazy to think that we have such highly portable computers in our pocket.

Ironic coming from someone who didn't know ARM was a member of HSA Foundation.

Hehehehe lol That's because they were able to... uh... I'm at a loss for the correct words... but run PowerPC apps through emulation... through Rosetta I think? And it was removed in Lion. ARM isn't powerful enough to do that, and even if it was it's hardly ideal. Plus a Mac would no longer be able to run Windows which would suck.