An Introductory Reminder that iMac 5K is Still an All-in-One of a Kind

Here's a quick survey of the "competitive field" and key specs:

>> Dell Inspiron 7000:

• 27" 1080p max resolution (yes, as in 2MP) touch-panel-option display, 11th-gen Intel Core i7-1165G7 quad-core-laptop-ish CPU, up-to-GeForce MX330 with 2GB (yes, TWO gigabytes) of dedicated VRAM, up to 1TB max SSD with a suspiciously Fusion Drive-looking 256GB SSD + 1TB SSD (2.5" HDD, 5400RPM) midrange storage option - $1300(US) list price reasonably well-equipped

>> HP ENVY 34" All-in-One (apparently just launched and pre-ordering for end-of-November 2021 shipping!)

• 34" 5120x2160 non-touch display(~11.1MP@ 500 nits max), Intel 11th-gen Core i7-11700 8-perf-core CPU, NVIDIA RTX 3060 GPU w/ 6GB GDDR6, 16GB RAM with four total SODIMM expansion slots, 1TB onboard SSD with two expansion M.2 slots, an iMac-like port array including two TB4 and six other USB-A (mostly 10Gbps), and a 330W power adapter - $2,350 list price as described

>> Lenovo Yoga A940 (a funky base-is-the-computer design which is used to support the adjustable touch display, think of a chunkier Surface Studio but with a 9th-gen Intel Core CPU vs. 7th-gen, and it's actually in stock)

• 27" standard-aspect ratio 4K (8.3MP) touch display, antiglare, 350 nits max, Intel i7-9700 8-perf-core CPU, Radeon RX 560 w/ 4GB VRAM, 32GB RAM, same-as-Dell 256GB SSD + 1TB 2.5" HDD Fusion-ish Drive, the usual ports, mostly USB-A, one Thunderbolt - $2,620 list price, $1,740 on Black Friday special pricing (NOTE: It's common in the PC industry for OEMs to regularly discount their products vs. list/MSRP)

Yes, it's true that there are some all-in-one PCs with starting prices above the iMac 5K. And the HP and Lenovo both have stronger starting specs in the CPU and GPU department.

However, the iMac 5K remains in a class of its own - largely because the PC competition doesn't focus its product lines as much as Apple, but also because, like in Apple's case, the money and volumes for the industry are mostly to be found in laptops, which have added touch panels and convertible features over time.

The HP Envy is the closest I've ever seen a major PC player come to Apple's 5K (~14.7MP)display in an all-in-one, but even then, it's via a much larger ultrawide-format monitor with fewer pixels at 34" vs. 27" (lower ppi), and you can still outfit the iMac 5K with a more powerful 10-core 10th-gen Core i9 + Radeon Pro 5700XT w/16GB of GDDR6.

Because Apple only has one all-in-one desktop line with two variants, and because it actually cares more about this particular integrated desktop space than the other PC competition (which all sell a variety of minitowers), it continues to stand apart from the crowd in overall capability, even in an x86-powered package.

This oddly positions it both well (objectively or so I think) and "awkwardly" (in the minds of some tech enthusiasts/commentators) as iMac 5K almost certainly prepares for its transition to Apple Silicon.

In Part 3 of my humble home game Apple Silicon mini-series, I'll attempt to cover three topics:

(1) In very general performance terms, what will iMac 5K be like on Apple Silicon?

(2) Will there be an iMac Pro replacement (the one-and-done Xeon variant from 2017)?

(3) Is Alder Lake particularly relevant to iMac 5K's "Apple Silicon Mission"?

(1) iMac 5K: Now and in the Near Future

We're going to keep things very, very simple by focusing on CPU and GPU, since most of the rest of the Apple Silicon "Pro/Max" platform (basic speeds and feeds like memory bandwidth, SSD and RAM configs, A14 Bionic-class Neural Engine, dedicated media encode/decode blocks, etc.) is a fairly known quantity. Yes, CPU and GPU alone are a somewhat incomplete picture, but still will provide a general idea of relative performance uplift moving to a new silicon platform.

>> CPU Comparisons

With a big hat-tip to Primate Labs' Geekbench Browser, here's the single-core and multi-core CPU scores of existing iMac 5K SKUs (last updated to 10th-gen Intel Core CPUs in Aug. 2020) compared to what we currently already see in the MacBook Pro 14.2" and 16.2" lineup:

It certainly seems like the CPU performance is there, so how about price?

Let's assume a worst-case-type pricing scenario where Apple's new-design iMac 5K, which is rumored by the very accurate Ross Young, CEO of Display Supply Chain Consultants to add ProMotion and miniLED to a presumably-5K-or-better display, gets a $300 price hike on the base model.

This might be too conservative, considering that the base model (top-binned 10-core CPU) MacBook Pro 16.2" is $100 more than the Intel MacBook Pro 16" (with 6-core Core i7 CPU running GB5 1000 single-core/5300 multi) that it replaced. Then again, an it's-about-time 16GB baseline of unified RAM would be double the loadout on all standard iMac 5K SKUs available today...even the $2,300 model.

Anyway. On a pure CPU basis (yes, accepting that some apps still need to support Apple Silicon natively), the estimated-$2,100 iMac 5K M1 Pro base model, if 8-core-binned-CPU, still ends up comfortably faster than the $2,700 top-end Intel iMac 5K SKU with 10-core CPU upgrade. The lowest cost of entry for that 10-core Core i9 is the $2,500 mid-range +$500 CTO upgrade option, which sticks with a lower-performing Radeon Pro 5300 GPU.

What if the 10-core CPU isn't standard? We've already seen that the MBP14.2 with double-binned M1 Pro offers an initial $200 upgrade option to the full 10-core CPU ($100 per performance core, I suppose). So at worst, you can expect to pay around $2300 for an onboard CPU 50% faster (in both single and multi-core performance) than the current standard high-end iMac 5K 6-core Core i7 config with the exact same $2,300 price.

I'm mindful that $2100 or $2300 feels even more expensive in these tough pandemic times, year-on-year inflation aside...but at least the CPU value is definitely there for those who want or must have the iMac 5K platform for their computing needs.

And as far as the competitive set goes, it will probably take until 12th-gen Intel Core or AMD Zen 4 for the all-in-one PC competition to achieve performance parity or superiority...that is, if they even expend the effort right away, for the reasons I mentioned in the introduction.

Next up, iMac 5K GPU comparisons, which are much more interesting.

>> GPU Comparisons

Generation-over-generation comparisons get little fuzzier between the current iMac 5K's AMD Radeon GPUs and the unified-memory Apple Silicon GPUs slated to replace them. So we'll have to stay in the realm of semi-educated guesses, despite the concrete Metal benchmarks and TFLOPs ratings we do have.

First, we'll compare the somewhat-"controversial" Geekbench 5 Metal test results of all four iMac 5K SKUs shipping today and all four M1 Pro/Max GPU configurations available on the newest MacBook Pros. To be clear, the "controversy" might just be a GB5 optimization thing more than anything else, since Apple Silicon GPUs on Mac were only launched one year ago (and the M1 Pro/Max GPUs, about one month ago). I'll explain further after the chart.

You probably noticed that the GB5 Metal results don't scale particularly well for the 32-core GPU. Generally speaking, people were expecting a GB5 Metal score of around 80,000 or so, and just shy of 70,000 isn't that.

The chief culprit? The GB5 Metal test doesn't invoke the full wattage of the 32-core GPU, thanks to the findings of Vadim Yuryev at Max Tech (on YouTube), aided by the asitop macOS powermetrics GUI designed by Timothy Liu. Not even 60% of the 60W rated power maximum, in fact - just 35W peak during the test.

If the Geekbench Metal results don't (yet) tell the entire story, we do have one traditional metric of GPU performance to refer to - good old TFLOPs. Yes, it's not a perfect proxy for "graphics horsepower" or anything like that, but I'd argue it's at least decent - especially since AMD, NVIDIA and Apple still publish teraflop ratings on their GPUs. Here's AMD's helpfully published data on the current-gen iMac 5K Radeon GPUs alongside the known/estimated M1 Pro/Max GPU TFLOPs ratings:

The Radeon Pro lineup in today's iMac 5K models steps up fairly consistently from level to level: 4.2, 5.3, 6.2 and 7.6 TFLOPs from the humble Radeon Pro 5300 (with a not-great 4GB of dedicated memory) all the way up to the midrange-ish Radeon Pro 5700 XT with a much more robust 16GB of dedicated VRAM ($2800 minimum iMac price). (Note: Due to thermal/space constraints, iMac GPUs are detuned versions of the often-150W+ GPU cards you'd see in minitower PCs.)

The current M1 Pro/Max GPU lineup is less consistent, with 4.5, 5.2, 7.8 and 10.4TFLOPs ratings on offer in the newest MacBook Pros. Looking strictly at TFLOPs, the 14-core and 16-core M1 Pro GPUs are comparable to the Radeon Pro 5300 and 5500 XT. The 24-core and 32-core M1 Max GPUs compare more favorably against their anticipated Radeon Pro 5700 and 5700 XT counterparts.

>> The Potentially Tricky GPU Value Equation

Interestingly enough, iMac 5K represents the one segment of the Mac lineup facing potentially awkward compares to the current generation in the GPU specs department. If the "Apple Silicon iMac 5K" keeps GPU wattages and upgrade pricing the same, it might cost $300 simply to upgrade a base-model double-binned iMac to a GPU of the same TFLOPs class as the Radeon Pro 5500 XT. And from that point, it might cost $600 more to upgrade the GPU to the 24-core M1 Max level - since at present, you can't upgrade the MBP14.2/16.2 to a 24-core GPU (nominally $200) without also paying $400 to increase the unified RAM amount to 32GB and upgrade to the M1 Max higher-memory-bandwidth SoC. So the gains in value from not having to upgrade the CPU past the 10-core level are very quickly erased.

Now to be fair, Apple Silicon is a very different beast from the traditional x86 CPU + dGPU arrangement. In theory, a ~5TFLOPs Apple Silicon GPU is still better than a comparable dGPU (assuming equivalent optimization for the app in question) thanks to the unified memory architecture, a more efficient design and tile-based deferred rendering model originally adopted due to smartphone-class resource constraints.

However, iMac 5K is not supposed to be a system where efficiency matters nearly as much - especially not now, where Apple can easily design for cooling at least one 60W GPU, since iMac 5K systems already provide around 130W in power to Radeon GPUs today.

How could Apple get around this GPU value problem? Well, starting the entire iMac 5K-class platform from the top-binned M1 Pro 10-core CPU and 16-core GPU for $2100 or less would be a good start. Finding a way to deliver users ~8TFLOPs-class GPU power for under $2700 would also help.

The "easy" (read: actually quite nontrivial) approach would be adding power, allowing the GPU to extract perhaps 20% or so more performance with 40-50% more wattage. Except that it seems to go against everything Apple Silicon stands for so far.

Could Apple add memory bandwidth and/or GPU cores without requiring an automatic upgrade to 32GB of RAM? Could it save on cost by allowing higher GPU core SoCs to have a binned M1 Pro-level Media Engine? Can a 24-core GPU even function optimally without the 32GB minimum unified RAM within the M1 Max SoC platform?

To me, the upcoming iMac 5K GPU situation is even more intriguing than iMac Pro and the theoretical threat of Alder Lake and Zen 4 - though we'll be getting into those still-interesting storylines right now!

(2) iMac Pro's Near-Certain Return

Based on Apple's only iMac Pro ever (the up-to-18-core Xeon edition launched in 2017, which saw no updates throughout its lifetime), I understand why there's some doubt that Apple will revive a similar product in Apple Silicon form.

But to me, simple math demands its return. There's only one way for the current Apple Silicon building blocks to achieve 128GB of system memory and well over 10 TFLOPs of GPU performance within a Mac. Also, Apple's been rumored to be entering the "chiplet" arena of smaller, higher-yielding "fused" chips taking on the performance of much larger, lower-yielding "monolithic" chips. So why not start with a 20-core CPU, dual-SoC package in an Apple Silicon iMac Pro?

Even today, Apple's iMac 5K display (a 2014 launch, by the way) is still a high-water mark in its class, and a "mere" ProMotion + miniLED backlight makeover should be more than enough of a display revamp for the next few years, to say nothing of the overall design change. And several years ago, Apple clearly saw (and attempted to target) pro user demand for far-more-than-typical-compute power stuffed into an all-in-one form factor...even if it ended up being a bit of a stopgap before the 2019 Mac Pro, and perhaps even the entire Apple Silicon Mac platform.

Indeed, arguably the most famous power user in the world - none other than Marques Brownlee of MKBHD fame - was well-known in pre-pandemic times for taking an iMac Pro while travelling because it was just portable and fast enough for him to work his pro-video-production magic after importing video from high-end RED camera gear.

The hypothetical M1 Max Duo, at its peak config, should provide more than sufficient power for many 2022-era pro needs. Twin 10-core CPUs, assuming no upclocking from current M1 Max chips plus a hefty 20% multicore efficiency loss, will still allow for a GB5 CPU multicore score of around 20,000 - around 50% better than the top-end 18-core Xeon CPU in the 2017 machine while using only around 60W max CPU power. An M1 Max Duo iMac Pro would also rival the fastest Mac ever tested - yes, this means the 28-core Xeon-based Mac Pro minitower launched in 2019. I also can't leave out the massive estimated 800GB/s of total system memory bandwidth, dual Neural Engines, or twin Media Engines with a combined total of four ProRes encode/decode blocks.

Meanwhile, M1 Max Duo, in addition to allowing the 128GB RAM ceiling currently seen in prosumer iMac 5K models today (Apple Silicon pricing TBD, of course), would also allow for up to 64 GPU cores, for a combined ~20.8TFLOPs of FP32 performance. This is a healthy uplift over the 11 TFLOPs available on iMac Pro with the top-end Radeon Pro Vega 64. And with currently known M1 Max configs, the 64 GPU cores would have a minimum of 64GB unified memory available to access, versus the 16GB of HBM2 discrete memory from the Vega 64.

How much would adding a entire extra M1 Max SoC cost? Well, I'm sure it wouldn't be cheap - easily $2,000 more than a single-M1 Max config? Apple can name its price to some degree, but it's also aware of the $4,999 iMac Pro base model of the past. Also, in addition to a binned M1 Max Duo with twin 24-core GPUs(~15.6combined TFLOPs) in a chiplet, there's also the possibility of M1 Pro Duo configs for those who need maximum CPU performance but not as intense GPU performance. So it's possible that iMac Pro can provide far more performance at the base model level for several hundred dollars less than the first-gen iMac Pro's $5,000 base price.

(3) Apple's Statement of Performance Intent, Independent of Alder Lake (For Now)

Finally, since it's trendy and timely, I'll spend a little time on the Alder Lake news. I'm on social media record as being extremely pro-competition. Even though I'm obviously Team Mac, I love the disruptive implications of Apple Silicon - because it's sending shockwaves throughout the PC industry, and kicking Intel's, AMD's, Qualcomm's and other chip players' competitive energies into extreme overdrive.

This sets up ideal conditions for everyone - including Apple, a company with plenty to prove in the PC industry - innovating and competing at high levels, to consumers' inevitable benefit.

Let's get this out of the way quickly - Alder Lake and M1 Max are not in direct competition, period.

Why? Alder Lake and M1 Max are diametrically opposed compute solutions that happen to both be destined for desktop-class computers (well...probably, in the case of the M1 Pro/Max). The most powerful known Alder Lake desktop processor (i9-12900K), an enthusiast-focused 8-performance-core + 8-efficiency-core design, has a rated maximum all-turbo wattage of 241W. Base frequency power consumes 125W, or nearly half. By the way, even the efficiency cores have turbo clocks, with a max turbo frequency of 3.9 GHz, over 60% higher than the base clock. Meanwhile, Apple's M1 and M1 Pro/Max efficiency cores probably use fractions of a watt, and no M-chip CPU to date uses turbo clocks.

Alder Lake also benefits from advanced active cooling for best results - Ars Technica, for example, used a liquid cooler for its test machine, which is something we haven't seen from any Mac since the twilight years of the Power Mac G5 minitower era back in 2005 or so. While today's liquid coolers aren't that expensive ($100 or less) or difficult to install, there's no question they're an aggressive thermal solution, while Apple prefers air cooling (and in the case of the M1 MacBook Air, no fan in the first place).

Putting aside the tendency for new-gen CPUs (like Alder Lake) to require a new motherboard at minimum to go with the new-PIN-layout CPU (with prices ranging from under $200 to over $500), a ~240Wmax CPU just doesn't work in an iMac 5K, which has maximum thermal output of around 300W, or in an iMac Pro, which has 370W measured max output. Those systems would not have enough power overhead for everything else. Honestly, I'd be incredibly surprised to see a full-power i9-12900K or anything close to it in any all-in-one PC in the foreseeable future. 200W+ max output CPUs really aren't suitable for such thermally constrained devices.

This is the fundamental reason why Apple made the decision to move on from Intel - there is simply no other chip designer on Earth that can deliver the PC-paradigm perf/W that Apple requires from its Macs, aside from itself.

Apple set a CPU power threshold of 30W and a GPU power threshold of around 60W, then challenged itself to build the most performant pro-level technology possible within those energy/thermal constraints. It's true that these constraints are somewhat artificial in the case of desktop PCs, but this is only the beginning of the the M-chip era. As the competition evolves, so will Apple, adjusting core counts, transistor budgets, and perhaps even desktop SoC wattages to a certain degree. However, "right or wrong", the current Apple philosophy will continue to be driven by maximum efficiency within a compute form factor, which in turn defines the operating parameters of the silicon.

Ironically, this means Apple's near-term challenge is not to avoid any thermal ceiling - it's actually to avoid leaving performance on the table and underserving customers in the name of low power draw.

In the meantime, some (not all) observers of the Mac space should take note of who iMac 5K and iMac Pro are for. Yes, they are halo devices, as Apple's iconic all-in-ones. Yes, they need to be suitably performant for the price class. Yes, there might be value issues with users who expect more GPU power from Apple at the $2,500-$3,000 price band.

But Apple's primary objective with these systems isn't as simplistic as "beat Alder Lake". It's to win over the existing prosumer-plus iMac userbase - one it's followed since at least the mid-2000s, when the iMacs updated to single-core G5 and later dual-core Intel CPUs. The company has done a fantastic job addressing user needs and wants with its mainstream-through-performance laptops (plus the entry-level Mac mini and iMac 4.5K) so far. And it has done so with three fundamental M-chip building blocks (M1, M1 Pro and M1 Max), rather than the 20+ processor choices AMD and Intel each offer across their premium-tier mobile and desktop CPUs alone.

At least for me, it remains difficult to bet against Apple, especially if there really is a modern-workstation-grade ~180W-combinedM1 Max Duo SoC juggernaut potentially set to launch in an iMac next spring.