A12X benchmarks and ARM Mac prospects

Hardware
1

Many of the reviews of the new iPad Pros have made mention of the Geekbench scores of the new A12X processor (see Gruber and Panzarino, for example). The stats are pretty impressive. Most of these reviews are comparing the A12X to last year’s iPad, this year’s iPhone and to Apple’s current laptops. Take this chart from Gruber’s piece, for instance:

                Single Multi
                Core   Core
2018 iPad Pro   5,007  18,051
2017 iPad pro   3,894   9,242
iPhone XS       4,851  10,534
15" MacBook Pro 5,653  21,737
(2.9 Ghz i9)

That i9 is the fastest processor you can get in a MacBook right now. So the A12X seems to be able to cover pretty much all of Apple’s laptop line. In the past few years we’ve gone from the iPad being able to match the MacBook Adorable, to matching Apple’s mainstream consumer laptops, to matching their Pro machines.

This is the comparison that I think it really interesting, though:

                 Single Multi
                 Core   Core
2018 iPad Pro    5,007  18,051
8 core iMac Pro  5,020  30,664
10 core iMac Pro 5,139  35,439
12 core iMac Pro 5,141  40,501
18 core iMac Pro 5,288  46,919

One of the objections to the idea of Apple transitioning the Mac lineup to ARM has been that while the A-series processors could work for the laptop line, but what about the Mac Pro/iMac Pro? While there’s still a gap between the multicore performance of the A12X and the Xeons in the iMac Pros, the A12X is already about halfway there!

What’s more, if you look at the performance history of the 12.9" iPads:

                 Single Multi   Number
                 Core   Core   of Cores
2015 iPad Pro    3,144   5,367    2
2017 iPad Pro    3,974   9,562   3+3
2018 iPad Pro    5,007  18,051   4+4

They’ve roughly doubled the performance in each generation. I think a Mac Pro/iMac Pro class processor is well within reach.

2 Likes
2

All of this is without any active cooling, and with being concerned about battery life / energy consumption. Once those restraints come off (for replacing something non-portable at least) they might get even more of a boost.

3 Likes
3

Ars Technica has an interesting article on the A12X processor.

4

What does Geekbench actually prove? Just because the A12 scores good with a synthetic benchmark, doesn’t mean it’s going to be any faster in real world use.

iOS devices are locked down, highly integrated devices, it would be much harder to achieve the same efficiency in a PC environment.

Also, just because huge gains have been made in the past, doesn’t mean future releases will continue that pace. If this were the case Intel would have maintained their curve. For all you know, the past performance gains were low hanging fruit, and it will be harder to see improvement going forward.

1 Like
5

Just because Intel hasn’t managed also doesn’t mean that Apple can’t. :smile: Not saying that they will, but they have definitely made some serious progress on this, and for all we know they’ve been doing this in parallel with getting the technology ready for setup in Macs.

If the technology is already designed to allow for more cores etc and cooling, it shouldn’t be that far fetched that the power would already be more than sufficient for the lower end laptops - and I assume that this would be cheaper for Apple than Intel stuff?

6

Clearly a (fellow) Performance guy. :slight_smile:

I think there is nothing inherent in ARM architecture that it would eventually scale more linearly with time than Intel.

There is a degree of “taking up the slack” with ARM and eventually it will run into similar issues to Intel and the rate of improvement will slow down.

Note already multiprocessor effectiveness is a dominant factor, rather than single thread speed.

By the way we’re saying “core” here and I wonder if multithreading will come into play and we’ll talk both core and thread speed (with multiple threads per core). (Multithreading might be here already.)

As a Performance guy I know performance varies enormously by configuration, workload, environment. So I agree synthetic benchmarks are nice and illustrative but not representative of real life.

1 Like