Apple Announces The M1 Pro / M1 Max, Asahi Linux Starts Eyeing Their Bring-Up

[smitty3268]

Behold the beast: https://browser.geekbench.com/v5/cpu/10476727

Almost as fast as Ryzen 7 5800X both in ST and MT performance.

Looks almost like an 11700k.

[lucrus]

Behold the beast: https://browser.geekbench.com/v5/cpu/10476727

Almost as fast as Ryzen 7 5800X both in ST and MT performance.

But, for some reason, I suspect it costs like twice or three times...

[amxfonseca]

Unless most of that is doubled on the Max, the GPU is still giant. 57bn for the max - 34bn for the pro leaves the GPU side at ~23bn for the pro and ~46 for the max unless more than just the GPU is getting added on the max version.

You're correct that the CPU is probably smaller, though.

As far as we know the biggest difference between the Pro and Max is indeed the bigger GPU. So we can extrapolate a bit from that.

There’s also more video decoders, and twice the memory channels (400GBps vs 200GBps bandwidth) on the Max. That should also take some extra transistors. But not much to make an huge difference on the numbers. But I am no silicon expert.

Hopefully some die shits will be available soon for analysis. AnandTech usually makes great deep dives.

[tildearrow]

But, for some reason, I suspect it costs like twice or three times...

And consumes half or one third of energy...

[sinepgib]

Well, you can put an actual 8086 inside a modern CPU and it would only use 29000 transistors 😛. Which is not much in todays billion transistor CPUs.

I think legacy doesn’t help, but I don’t think it’s a transistor count limitation. Especially in modern microcoded architectures where the instruction set is independent of the internal microarchitecture.

Fair, but as said, it doesn't help. Plus, even one transistor used for that is too many transistors :P

I think Apple just focus on key areas, and they aren’t trying to make profit on the SoC alone, so things like die size and manufacturing process is not as important. But if you are Intel and AMD and your business is only to sell chips then you need to keep the costs down.

Which makes me think the extra middlemen don't help either. As bad as Apple is, maybe taking chip designer company's + ODM's + OEM's profits away from the equation helps. Ethically questionable as it may be, vertical integration does have potential for cutting costs after all. And when there's (real) competition, those costs are possibly translated into lower consumer prices (when there isn't it just means a wider margin for the company instead).

There’s a lot of other ARM CPUs on the market. But they also fall short when compared with a modern Apple developed core when it comes to performance and efficiency. So I don’t think it’s architecture (ISA) related.

I think it's relevant. Of course, ISA alone won't make or break, but no factory or package will fix a bloated ISA.

[coder]

I believe a big part (not the only one) of Apple technical advance is due to their exclusive access to the best manufacturing node at TSMC.

It certainly helps that they have access to the latest nodes a bit before others, but even comparing theirs to other mobile SoCs on the same manufacturing node consistently shows Apple with a massive lead of several generations' worth.

I think part of the problem competitors face is that they're all using off-the-shelf cores that were designed by ARM to address a broader swath of markets, and probably with smaller area targets than what Apple is designing. That's not to take away from their technical prowess, which is also the best in the industry.

I really hope Intel will manage to catchup on TSMC at some point, otherwise X86 PC will never catchup on Apple ARM.

x86 will never catch the best ARM cores on perf/W or perf/mm^2 (and thereby perf/$). I doubt even IPC, since most ARM cores target lower clocks, which enables a longer critical-path and therefore higher IPC (and further perf/W advantage). That x86 cores can stay competitive is mostly by virtue of higher power-budgets and higher clock targets.

I'm quite certain that both Intel and AMD have post-x86 cores in development. Jim Keller's last project at AMD was the K12 -- a custom, in-house ARM core that I think they wisely chose not to bring to market, due to financial constraints and the ARM server market being too immature at the time.

[coder]

I think legacy doesn’t help,

It complicates and constrains the front-end. It also imposes tighter memory-ordering semantics than ARM has.

Especially in modern microcoded architectures where the instruction set is independent of the internal microarchitecture.

This is largely a myth. The internal implementation is still constrained, in various ways, by the ISA.

[Developer12]

Well, you can put an actual 8086 inside a modern CPU and it would only use 29000 transistors 😛. Which is not much in todays billion transistor CPUs.

There's a 468 in your intel igpu, doing instruction scheduling and power management. Tons of 8051 cores too.

I think legacy doesn’t help, but I don’t think it’s a transistor count limitation. Especially in modern microcoded architectures where the instruction set is independent of the internal microarchitecture.

I think Apple just focus on key areas, and they aren’t trying to make profit on the SoC alone, so things like die size and manufacturing process is not as important. But if you are Intel and AMD and your business is only to sell chips then you need to keep the costs down.

There’s a lot of other ARM CPUs on the market. But they also fall short when compared with a modern Apple developed core when it comes to performance and efficiency. So I don’t think it’s architecture (ISA) related.

x86 is a fucking boat anchor. The instruction set is a living nightmare to decode into uops for an out-of-order risc core to process (mostly due to being variable-length). Not surprising, since the instruction set was invented 10 years too early. Before we knew that cisc (think DEC VAX) was poison to CPU design.

They've been throwing transistors at the problem ever since. Back when risc and x86 chips could still go head-to-head, risc chips used half as many transistors for the same performance. Ultimately, economies of scale and sales volume of the PC won out anyway and killed off most of risc.

There's only so much you can do with that transistor budget though, even if you sink it into huge decode caches. Past a certain point the complexity becomes unmanageable. AMD Zen managed 4-way parallel decoding to bring instructions into the core and keep it fed with uops to crunch. Intel moved mountains for 5-way. The M1, apple's first effort, is 8-way. First in history afaik.

There's lots of arm cores out there, but none have been designed with performance in mind. It's always taken a back-seat to power consumption. This is the first time serious focus has been given to performance and there's still probably a lot of catching up to do.

[zeehio]

Wasn't one of the selling points of the Apple M1 the "one processor for everyone" and how they were breaking the market by challenging the i3, i5, i7 and similar market segmentation into a single processor for everyone?

I don't recall, maybe Apple never said that, and it was just said by others. I definitely read that somewhesomewhere.

[coder]

The M1, apple's first effort, is 8-way.

Huh??? Apple has been designing their own ARM cores for over a decade! You don't just come out of nowhere and make a chip like the M1 on your first try! Is that what you thought happened?

There's lots of arm cores out there, but none have been designed with performance in mind. It's always taken a back-seat to power consumption.

That's not really true. There's Fujitsu's A64FX supercomputer CPU and ARM's new Neoverse V1 core. We could also talk about a lot of performance-oriented mobile cores, which would fall in the same category as Apple's "performance" cores (i.e. those emphasizing efficient performance).

This is the first time serious focus has been given to performance and there's still probably a lot of catching up to do.

It's true that ARM's own cores are always a balance of performance and other factors. They traditionally haven't been able to target them as narrowly as Apple has. However, ARM's new V-series server/workstation cores look set to change that. The V1's are X1-derived, IIRC. However, I think the V2 might be a more purpose-built speed demon.