The Absolute Peak of Tech Snobbery: “Am I Getting 2021 Tech or 2020 Tech in My (Rumored) Late 2021 14″/16″ MacBook Pro (or Mac mini Pro)?”
I mean, just how spoiled can we get? The M1 is so good, just add 4 CPU performance cores, double-to-quadruple the GPU cores, raise the SSD/RAM limits, and add at least three independent TB3 ports, and hardly anyone would seriously complain.
And to keep things very quick for those thinking Apple simply goes this M1-derived route to the chagrin of tech snobs everywhere (thanks for crushing our dreams)? If this “worst-case scenario” does happen, perhaps overall CPU performance would be around 90-ish percent “as good as it could have been” by favoring “A14-platform” silicon over the just-now-shipping “A15-platform” silicon.
But it’s not November 2020. Mid-2021 is in the rear-view mirror. Enough time has passed that it’s actually quite thinkable to hope, to expect, that Apple will employ newer-than-M1 technology in its prosumer-grade Macs.
Why? Apple has its pride as a premier silicon designer, and also knows its market segmentation.
Back when it didn’t have much of a choice, Apple made do with
• dual-core Core i3 chips (*involuntary shiver*) in base-model MacBook Airs,
• quad-to-hexa-Core i5 through i7 for the mid-range,
• and up to 8-core i9 chips for the 16″ MacBook Pro and up to 10-core i9 on top-end CTO 27″ iMac (still on sale…for now)
Now that Apple’s taking complete control of Mac silicon design, Apple will soon run a grand total of two chip families (“M1” and what most tech journopundits call “M1X”), expected to cover every single Mac SKU except the top-end Mac Pro workstation.1 This takes the concept of “modular” to new heights. Of course, it’s likely that the “M1X” chips will have two or more distinct varieties to choose from, since prosumers like options, particularly the ability to add on Moar Cores™ for their demanding multithreaded workloads.
Mac prosumers have accepted that their future laptops and desktops with the possible exception of Mac Pro will not be upgradeable (well, at all) after purchase. If the cases of their MacBook Pros and iMacs must remain sealed to them for years aside from repairs or some rather-brave iFixit-style DIY, they at least want up-level staple features, such as:
• 8 or more performance CPU cores
• discrete-level GPU performance with starting power at least in Radeon 5500M-ish territory (a mid-range option in the current 2019-launched MBP16)
• RAM limits as high as 32GB+ for the rumored MBP14, 64GB or higher for MBP16, and 128GB or higher for prosumer iMacs
• SSD max capacity way over 2TB
• plenty of Thunderbolt 3/4 lanes, at least four ports’ worth
I didn’t mention mini-LED, high refresh rate displays, or HDMI, because all of these technologies are enabled by the Apple M1 already. It’s “high-overhead” items like Even Moar Cores™, pro-level I/O, massive amounts of RAM, and 8TB-level onboard SSD which seem…not well-suited to the M1 architecture, which is laser-focused on maximum efficiency.
And if you think about it, would Apple be better off:
• further stretching the boundaries of a very-low-power-type architecture,
• or instead, starting fresh with newer “building blocks” to create an efficient Mac system architecture specifically for high(er)-performance computing – with a max sustained thermal overhead of, say, 40-50W on laptop form factors?
Given that all the Apple users who don’t intend to purchase an M1 Mac represent the Mac performance-enthusiast demographic2, I really do think that Apple’s obsessive, productive focus on perf/W ultimately led them to Give the Peoples What They Want™, and utilize A15-based tech in its yet-to-be-announced prosumer Macs. For one, the “big” CPU cores are supposedly more performant at around the same power draw as A14-type cores – about 10% faster in single-core, and potentially with very significant year-over-year performance gains in those much-lower-power efficiency cores.
But there’s more. Apple introduced the concept of a 5-core GPU with the A15. And on iPhone 13 Pro/Max, it can drive very impressive performance gains in the 50%+ range (GB5 Metal compute score basis) when comparing to the 4-core GPU of the A14 Bionic (inherently faster A15 GPU cores than A14, clocked a bit higher on the Pro iPhones to boot). The A15 also brings along Apple’s 4th-gen Neural Engine, claimed to be capable of slightly over 40% more operations per second from the same number of cores.
Even if that’s “all” of the notable architectural improvements A15 brings over A14…I mean, they seem like nice-enough advancements. Particularly so if this added performance can be attained at more or less the same power consumption as last year.
How to Know the “Engine Model” of Your Prosumer Mac Within Moments After Announcement or Press Release? 20, 30, or 40 GPU Cores (Likely) Means 2021, While 16 or 32 GPU Cores = 2020.
If you’re planning on watching the rumored (yet inevitable) prosumer Mac launch event, you’ll probably get your answer within seconds after Apple announces either the GPU core configurations or the performance rating of the Neural Engine during the specifications-heavy portion of the presentation.
For everyone else, it’s as easy as take the GPU cores, and see if the number on the rumored highest-end MacBook Pros or Mac mini Pros divides by five, because that’s the “unit of measurement” introduced by iPhone 13 Pro/Max and iPad mini A15. If this rumored “M1X” is derived from the A15 as I’m hoping it will be, it will logically employ “units” of 4 performance cores and “units” of 4-5 GPU cores (possibly due to the same kind of binning that saw M1 come with either 7- or 8-core GPUs).
Yes, it’s not what well-sourced Apple tech journos including long-time big name Mark Gurman have been predicting as far as GPU cores are concerned (the consensus has generally settled on 16-core or 32-core GPU options), but that was before the A15 was officially announced. Now, Apple could announce GPU core configurations in the 20-core or even 40-core range, given these “units of five”. And if that might be a little too taxing on battery life or take up too much space on the SoC, there’s even the option to dial down the 5-core packages to 15-core or 30-core bundles. After all, since a 5-core iPhone 13 Pro GPU can perform as much as 50%+ better than the 4-core GPU within any given iPhone 12, that means multicore GPU maths could work out like this:
M1 GPU = 1x baseline performance per 4 cores = 2 units of performance @ 8 GPU cores (assume minimal multicore penalty)
= 4 units of combined performance @ 16 cores
= 8 units of combined performance @ 32 cores
“M1X” GPU based on A15 = 1.5x baseline performance per 5 GPU cores
= 4.5 units of combined performance @ 15 cores
= 9 units of combined performance @ 30 cores
So even if you assume the GPU gains are a little more modest than what we’re already seeing in the iPhone 13 Pro vs. iPhone 12 (which seems doubtful), you still get “better-than 16- or 32-core” M1-platform GPU performance out of these newer-generation GPU cores without even needing 16 or 32 GPU cores.
And don’t get me wrong. If Apple can squeeze a 40-core GPU configuration into the MacBook Pro lineup, it would be splendid.
Throwing Out Geekbench 5 Score Guesses Just For Fun
Finally, let’s estimate some theoretical GB5 single-core CPU, multi-core CPU, and Metal scores for giggles. Thanks to numerous benchmarks posted from recent Apple devices (and real-world benchmarks), we know that modern A-chips and the M1 multiprocess very well, so given certain assumptions we can get a fairly decent guess on how these (yes, “synthetic”) benchmarks might play out.
CPU
First, the “M1X” single-core CPU score. M1 scores in the 1700 range, which is around 8% higher than the A14 performance cores, largely due to higher clocking. A15 itself scores in the M1 single-core range, as confirmed by numerous GB5 runs by tech reviewers around the world. Why would we expect the “Mac version” of an A15 performance core to be any less upclocked than M1? Assign a 1720 average to the A15 single-core GB5 score, add 7-8%, and you get a score of at least 1840, a number that almost matches the Core i9-11900K — which has a base frequency of 3.5 GHz, but turbos to 5.3GHz (something that no A-chip or M-chip to date resorts to). In any case, a high-performance M-chip theoretical CPU single-core score ranking either first or second among all PC processors isn’t a bad place to start.
Then add whatever multicore multiplier penalty you like (which I personally think is modest, since Apple CPUs multiprocess well and don’t turbo) and multiply by number of perf cores. The rumor consensus seems to be 8, so right off the bat, you can imagine a score of 1840 x .95 x 8 = nearly 14,000. Even the best of x86 require at least 12 cores (often 16+) and rated TDPs of 100W+ to achieve those kinds of multi-core scores.
Oh, but that leaves out a wild card – Apple’s efficiency CPU cores. Some rumors say Apple will only have two efficiency cores. They may well be right, but I think Apple will value efficiency cores more than ever in laptops with higher average power consumption than their M1 counterparts. If you think my opinion makes more sense, add over 1000 points to the multicore score, vaulting an 8+4 CPU core configuration to a GB5 result of around 15,000, roughly a tie for 10th best multi-core amongst any PC CPU. Otherwise, add a similar-to-M1 600-to-700-ish points.
Not bad for an SoC that is essentially guaranteed to draw less than the nominal “45-65W” TDP of high-end Intel laptop chips. And what if Apple is able to add more than eight performance CPU cores to a MacBook Pro over time? There’s no reason why Apple couldn’t add more CPU cores to a new, larger iMac or a Mac mini Pro right now, given a good-enough active cooling system.
GPU
Let’s assume that Apple uses the same basic GPU cores from the A15 in either 4-core or 5-core-binned units. We know that M1 8-core GPUs can score around 20k and change in GB5’s Metal benchmark, thanks to upclocking the GPU cores based on the A14.
Starting with the down-clocked, possibly binned 4-core GPUs in iPhone 13 and 13 mini, we easily see Metal scores of 10,600, which with a 5% upclock yields around 11,000 as a base unit in a higher-performance Mac. At 16 cores, that’s 44,000, possibly a couple thousand points lower – the level of the Radeon Pro 5600M, a CTO upgrade costing well over $500 in the MBP16.
If we assume the 16-core GPU derived from the non-Pro iPhone 13 models is the absolute baseline, we can then move on to iPhone 13 Pro-derived blocks of five GPU cores. iPhone 13 Pros average around 14,000 in the GB5 Metal benchmark – add 5% to that for Macs, and you get around 14,700.
Multiply by 4 for a 20-core GPU configuration and you’ll get a score of perhaps 55,000 or so. For a 30-core configuration, that number jumps past 80,000. That’s a theoretical score higher than the Radeon Pro 5700XT – a $500 CTO option on the last-hurrah 27″ iMac 5K, a desktop machine with a base price of $2,300 and a total system power consumption of close to 300W under full load.
Managing Expectations…Somehow
Is a MacBook Pro with Apple Silicon, no matter how well-specced, going to challenge a 2019 Intel Mac Pro for the performance crown in CPU or GPU? Of course not (right now). A 24-core Mac Pro 🤣 will win in multi-core CPU with a score perhaps 25% higher than a 8+2 CPU MacBook Pro. Then again, the 24-core Mac Pro has a base price of $12,000, and the very fastest Intel MBPs ever built can’t even muster a 7,000 score in GB5 CPU multi-core, while the next level of rumored M-chips could be poised to deliver around twice that.
A high-end 30-ish-GPU-core configuration in a MBP16 also won’t top the GB5 charts in Metal, sadly. It’s still about 25% off the pace of the Radeon Pro Vega II…which costs $4,400 at the Apple Store, and is a full-on hundreds-of-watts MPX Module for use only with a Mac Pro. In fairness to the Pro Vega II, it has a theoretical 28 TFLOPs of GPU performance for all that cash, while top-end Apple Silicon MacBook Pro models may end up with half the teraflops, perhaps a bit less.
Somehow, I’m not the least bit disappointed. PC OEMs, Intel and AMD will put on a brave face, but the Apple Silicon-branded competition will soon arrive quite suddenly at the imcumbents’ doorsteps, with never-before-seen performance-per-watt. The resulting pitched battle in personal computing will benefit Apple Silicon users and x86 PC users alike, which is something everyone on Team Mac and Team PC can celebrate.
—————
Footnotes:
[1] I suppose you could add iMac Pro to the top-end Mac list, if you’re holding out hope for a second generation of that never-once-refreshed Xeon-based Mac running Apple Silicon.
[2] You don’t need to remind Mac users that this doesn’t really include games, which Windows has in far greater abundance. I’m one of at least a few million people who are painfully aware.
The AAPL Tree 