M1 Max 32-core 64GB vs. i9 iMac Radeon Pro 5700XT Geekbench and comments

[Makosuke]

I've finished setting up my shiny new 2021 MBP (16") with M1 Max and upgrading my 27" iMac final gen to macOS 12, and given that they're both very similarly equipped and running freshly installed OSes I figured this was as good a time as any to see what the (presumably) final generation of top-of-line prosumer Intel Mac desktop can do versus a year later's top-of-line Apple Silicon Mac laptop.

Leaving aside the standard caveat that the only benchmark that really matters is whatever you're actually doing with it, these two computers are extremely similarly equipped. Presumably this will give a pretty good comparison between what a near-future top-of-line M1 Max based iMac and the current top-of-line Intel iMac can do, at least in Geekbench.

iMac 27" 2020 (20,2):

Intel Core i9-10910, 10-core​

64 GB RAM (2x 32GB sticks)​

Radeon Pro 5700 XT w/16GB​

1TB storage​

12.0.1 freshly installed with users migrated​

MacBook Pro 2021 (18,2):

M1 Max, 10-core​

32-core GPU​

64 GB memory​

1TB storage​

12.0.1 freshly installed with users migrated​

In both cases I rebooted, waited a couple of minutes for startup stuff to finish, opened Geekbench 5 and quit the Finder so it was the only thing running, and also ran a couple of times to make sure results were consistent.

Results:

iMac:

Single-core: 1,364​

Multi-core: 9,730​

OpenCL Compute: 53,451​

Metal Compute: 59,637​

Blackmagic disk speed test: peak ~2800MB/s write, ~2350MB/s read (not sure why write is faster)​

MacBook Pro:

Single-core: 1,781 (+30%)​

Multi-core: 12,597 (+29%)​

OpenCL Compute: 64,473 (+20%)​

Metal Compute: 66,624 (+11%)​

Blackmagic disk speed test: peak ~5350MB/s write, ~5350MB/s read (+91%/+127%)​

None of these results are particularly surprising, other than the OpenCL result on the MBP being so close to the Metal result, since it seems like in general OpenCL registers significantly slower, and when I ran the benchmark earlier with a bunch of stuff open it was lower than this although the other results were similar.

Basically, assuming that the first generation of AS 27" iMac uses the same M1 Max, the top-of-line is going to be a solid year-over-year upgrade. I would guess that whatever this year's high-end iMac-grade desktop AMD GPU would have been will be somewhat faster than the M1 Max in Geekbench Compute benchmarks, although the real-world performance will vary widely depending on what exactly you're doing with it.

A more general comment, I really like the new form factor and design, but coming from a 2018 15" it makes me feel like Apple should really start making a 15/16" "ultra Air" or something like that, more along the lines of the older ultra-thin MBPs that sacrifices some thermals, battery life, and ports in favor of a thinner and lighter form factor.

For a truly "Pro" machine the new 16.1" is a perfectly acceptable size and weight but it really is noticeably bigger and heavier than the old one, and I'm 100% sure there are users who would like the bigger screen but also would prefer a thinner, lighter, and slightly less-industrial computer. I know of at least one person who bought a lower-end 16.1" that's in that category, and even I might have considered it.

I also, personally, miss the Touch Bar. I like hardware keys, and wasn't a huge fan of the Touch Bar, but I really have noticed situations where I reach for it and am disappointed it's not there. Kind of wish they'd just squeezed it in above the row of physical function keys--seems like there's enough room for it.

 

[vladi]

The biggest difference maker in the actual real speed is always storage speed adn then CPU or GPU. That iMac is storage is super gimped.

 

[ADGrant]

Very useful comparison, thanks. Do you have any "real world" performance comparisons you can share? Or other benchmarks such as Cinebench.

It is interesting that Apple's top of the line MBP is faster than their top spec Intel Desktop (aside from the absurdly priced Mac Pros).

 

[ADGrant]

The biggest difference maker in the actual real speed is always storage speed adn then CPU or GPU. That iMac is storage is super gimped.

I don't agree that the iMac storage is "super gimped". It's performance is on par with other recent T2 Macs which is to be expected since they all share the same SSD controller.

 

[wilberforce]

I don't agree that the iMac storage is "super gimped". It's performance is on par with other recent T2 Macs which is to be expected since they all share the same SSD controller.

Yes, what is super gimped is a Fusion drive at ~200MB/s. Horrific that they put these in iMacs until last year. How we have gotten spoilt, that 2500MB/s is now considered slow, one year later

 

[Makosuke]

The biggest difference maker in the actual real speed is always storage speed adn then CPU or GPU. That iMac is storage is super gimped.

It's clearly not state-of-the-art, but in what world, in 2020, was nearly 3GB/s "super gimped"? I have a 4x 1TB NVMe TB3 RAID array connected to that same Mac, and it's slower than the internal drive. Sure, it could obviously be faster, but that's a perfectly respectable 1TB SSD.

Edited to add: Past a certain point I'm not even remotely convinced that storage speed is the biggest difference maker in real-world performance. Going from a spinning drive to a high-random-IOPS SSD, the difference is absolutely massive for "average user" tasks, and even for a lot of pain-point pro things. A large multiple more than any other upgrade for anything other than entirely CPU or GPU bound tasks like rendering or 3D gaming.

But I would like to see the benchmark that shows any significant real-world difference between a fast 3GB/s SSD and a fast 6GB/s SSD for major use-cases. Sure, you'll get faster app launches and boots, but with any task in which an M1 Max would even be noticeably different from a M1 MacBook Air, the disk isn't going to be the limiting factor.

Very useful comparison, thanks. Do you have any "real world" performance comparisons you can share? Or other benchmarks such as Cinebench.

I do not at this point; I'm a prosumer who overbuys and keeps for a while, and the main things I do with my Macs are image and photo related, so not easily benchmark-able like a render or video transcode. I did run through some of the relatively slow ML features in Pixelmator Pro (de-noise, super resolution) on the same image, and without putting a stopwatch on it they feel similar (and are quite fast).

As for other benchmarks, looks like Cinebench is free so I'll give that one a spin and post the results. If there are other free benchmarks or specific tests folks would like me to run for comparison, I'd be glad to.

 

[Makosuke]

As requested, Cinebench r23:

iMac i9: 1,224 single, 14,061 multi (11.49 multiplier)
MBP M1 Max: 1,523 single, 12,307 multi (8.08x multiplier)

So the M1 Max comes in about 24% faster single core, and the i9 about 14% faster multi-core. I take these results to imply that the i9 is slightly hamstrung when it's not taking advantage of hyper-threading, while the M1 Max scales very linearly, but doesn't seem to get much of an extra nudge from the two efficiency cores.

A less impressive result in absolute terms, although I'd say still respectable, and impressive in a moderate-sized laptop that can run at these levels off battery.

What was astounding to me, though, was the thermal profile. During the 10 minute multi-core test, the iMac's fan was ramped up high. It's an impressively quiet computer with the fan cranked up, and the noise profile isn't grating, but you could hear it from the next room.

The M1 Max MBP, after running the CPU flat-out for 10 minutes straight was... effectively silent. If I put my ear right next to the keyboard, I could just barely hear the fan, so it was technically running, but it was hard to believe how quiet it was. My wife is on a FaceTime video call five feet away while the benchmark was running and didn't even notice. The bottom was bordering on hot to the touch, but I could still use it on my lap wearing jeans without it being uncomfortable.

Compared to the very loud and slightly grating whine that the 2018 MBP i9 this replaces made when under load, or the fan noise on literally any other even midrange laptop I've ever used when under heavy load, it's absolutely mind-boggling.

Just to confirm, I did a quick check on the M1 Max with the Power Adapter Energy Mode set to "High Power" instead of "Automatic". It made no measurable difference in performance. Manually forcing 20 threads instead of the default 10 also made no difference one way or the other, as you'd expect.

 

[ADGrant]

Yes, what is super gimped is a Fusion drive at ~200MB/s. Horrific that they put these in iMacs until last year. How we have gotten spoilt, that 2500MB/s is now considered slow, one year later

The 2020 iMacs are not available with Fusion drives. They are SSD only.

 

[ADGrant]

As requested, Cinebench r23:

iMac i9: 1,224 single, 14,061 multi (11.49 multiplier)
MBP M1 Max: 1,523 single, 12,307 multi (8.08x multiplier)

So the M1 Max comes in about 24% faster single core, and the i9 about 14% faster multi-core. I take these results to imply that the i9 is slightly hamstrung when it's not taking advantage of hyper-threading, while the M1 Max scales very linearly, but doesn't seem to get much of an extra nudge from the two efficiency cores.

A less impressive result in absolute terms, although I'd say still respectable, and impressive in a moderate-sized laptop that can run at these levels off battery.

What was astounding to me, though, was the thermal profile. During the 10 minute multi-core test, the iMac's fan was ramped up high. It's an impressively quiet computer with the fan cranked up, and the noise profile isn't grating, but you could hear it from the next room.

The M1 Max MBP, after running the CPU flat-out for 10 minutes straight was... effectively silent. If I put my ear right next to the keyboard, I could just barely hear the fan, so it was technically running, but it was hard to believe how quiet it was. My wife is on a FaceTime video call five feet away while the benchmark was running and didn't even notice. The bottom was bordering on hot to the touch, but I could still use it on my lap wearing jeans without it being uncomfortable.

Compared to the very loud and slightly grating whine that the 2018 MBP i9 this replaces made when under load, or the fan noise on literally any other even midrange laptop I've ever used when under heavy load, it's absolutely mind-boggling.

Just to confirm, I did a quick check on the M1 Max with the Power Adapter Energy Mode set to "High Power" instead of "Automatic". It made no measurable difference in performance. Manually forcing 20 threads instead of the default 10 also made no difference one way or the other, as you'd expect.

Thanks, very useful. It suggests that in real wold scenarios the i9/5700XT iMac is at least able to keep up. Of course that is a desktop part with high power consumption and heat.

 

[T'hain Esh Kelch]

Do you have any modern AAA games you could throw at it?

 

[wilberforce]

The 2020 iMacs are not available with Fusion drives. They are SSD only.

That is what I meant by “until last year.” Sorry, admittedly ambiguous

 

[metapunk2077fail]

Even those iMac scores are still very impressive and a lot more power than most people have.

 

[Makosuke]

Do you have any modern AAA games you could throw at it?

I have none. (Not a big gamer, although part of the reason the reason I got the decked-out final-generation Intel iMac was to boot into Windows to run the occasional not-cross-platform-game). If you have anything specific you had in mind that has a demo version, let me know.

Question: I wanted to just see what the GPU can do, so I was previously poking around a bit for anything really graphically intensive (even a demo like some of the stuff GPU makers create to show off) that is AS-native, and didn't really find anything. There's native 3D stuff, but nothing AAA-level I saw or even anything close to that level that's native.

As always, if what you want to do is play a particular non-AS-native game then its performance in Rosetta 2 is all that really matters, but I more wanted to just see what the 32-core GPU can do when pushed so if anybody has even a show-off-ish 3D demo that's native, I'd appreciate the link.

Even those iMac scores are still very impressive and a lot more power than most people have.

This is quite true; this may not be a "pro" iMac, but a 10-core 10th-gen i9 desktop CPU with a 16GB dedicated GPU is pretty beefy hardware by all but the most industrial or hardcore gamer standards.

At least with current levels of optimization, the M1 Max, assuming it ends up in the high-end iMac exactly as-is, will probably be a modest step up from the previous high-end Intel equivalent for some tasks, and a modest step down for others. For a few things that really take advantage of the dedicated hardware or specific optimizations, I assume it will perform exceptionally well, and for a few others comparatively poorly.

What's undeniable, though, is that for the power and thermal profile, the performance is absolutely incredible. It's giving performance that can stand up to higher-end desktop CPUs on battery, and you literally can't even hear the dang fan running.

It's perfectly reasonable to argue that from a pro standpoint that's of negligible benefit in a desktop if and when it eventually lands in one, but as of today that's a hypothetical. For the laptops it's in right now it's shoulder-to-shoulder with the beefiest mobile workstations at a fraction of the power needs, and I don't know that there's anything even remotely comparable.

 

[Makosuke]

Necro'ing my own thread since I happened to find myself in possession of a brand-new, well-equipped Dell Precision 5560 mobile workstation from work that's a pretty darned close match in Dell's lineup. So, I figured it was worth doing the same benchmarks on it.

Specs for Dell Precision 5560:

Intel Core i9-11950H, 8-core​

64 GB RAM (2x 32GB sticks)​

NVIDIA RTX A2000 Laptop w/4GB​

1TB storage​

Win10 Enterprise freshly installed, no bloatware​

Results (comparisons are to MBP M1 Max):

Single-core Geekbench: 1,624 1,781 (-9%)​

Multi-core Geekbench: 8,951 12,597 (-29%)​

OpenCL Compute Geekbench: 59,345 64,473 (-9%)​

​

Cinebench r23 multi-core: 9,855 (-20%)​

​

Blackmagic disk speed test: peak ~2403MB/s write, ~1640MB/s read (-55%/-70%)​

I tried running these benchmarks both with and without the power adapter plugged in. The Geekbench CPU score was nearly identical. The Geekbench OpenCL score was 15% slower. The Cinebench r23 score actually came in significantly (~7%) faster on battery; the score above is the higher one. No idea what's up with that.

Running the benchmarks, the Dell actually did quite well with fan noise, although it was certainly noticeable, if not loud, after Cinebench had been running for a few minutes. The underside, however, was hot enough to the touch that it was painful to hold my hand on it for more than a couple seconds.

---

As for other stuff:

The 5560 is 15" instead of the MBP's 16" so it's a bit smaller physically and weighs just a bit less (145g, ~7%). The Dell has one less USB-C/TB4 port and (surprisingly) no onboard HDMI (it comes with a small HDMI dongle that also includes one USB-A port).

An aside, it also looks rather similar closed--my wife literally said "Can they do that?" when I put it beside our MBP--although when open the palm surface looks like carbon fiber and the profile of the bottom is much less chunky and covered with vent slots, so it's not as much of a copy job overall.

Configured today on Dell's consumer site it costs $4600, about $700 more than my MBP, although sales and corporate discounts can be substantial, so I think my company paid a bit less for the Dell than the Mac cost me with edu discount.

And yes, it has an 11th-gen i9 instead of a 12th gen, but when we went to purchasing to get a top-of-the-line 15" mobile workstation last month, this is what we got, and this was certainly equivalent at the time of the MBP's release.

I didn't do a full test, but extrapolating the battery life is significantly worse--running Cinebench, the Dell looks like it'll get about 75 minutes of runtime with the CPU pegged, while the MBP is closer to 120 minutes doing the same.

In all, if I'm OS-agnostic and want a 15-16" medium-weight mobile workstation, Apple has Dell beat in every possible metric, including (shockingly) onboard ports and price unless you get hefty enterprise discounts, and even then they're close. I you can also upgrade the RAM and SSD in the Dell, which does count for something.

 

[Bodhitree]

Very interesting comparisons, that a MacBook Pro is basically a top-of-the-line mobile workstation is something that I hadnt heard put in quite those terms before.

 

[Makosuke]

Oddly, when I look up other benchmark results for the same i9 CPU as I was testing, the Geekbench numbers are similar (often worse), but the Cinebench is supposed to come in right around 12,000. Not sure why I'm getting erratic performance there, but I've got to assume it has to do with thermal throttling.

Very interesting comparisons, that a MacBook Pro is basically a top-of-the-line mobile workstation is something that I hadnt heard put in quite those terms before.

It's not "basically" a top-of-the-line mobile workstation, it is a top-of-the-line mobile workstation. The Precision 5560 is exactly what was offered when I asked for a "loaded for bear" mobile CAD workstation with a 15" screen. The only thing that could have been specc'd higher on the Dell is a mobile Xeon CPU and ECC RAM, which won't improve performance in these tasks (I assume the M1 Max's shared memory is not ECC-equivalent with parity, although I haven't heard anybody state this as fact).

I will make a distinction between a "laptop" workstation like the MBP and 5560 series, and a "portable" workstation like Dell's Precision 7560. The 7560 has the same CPU options, but can go up to 128GB RAM (four slots) and can be specc'd with a significantly better GPU (RTX A5000 with 16GB RAM) and up to 3 4TB SSDs. But (aside from the fact it costs $20,000 list and $14,000 on sale maxed out), that system is much thicker, weighs a full pound more and the GPU alone has a 140W TDP. So while it can definitely be moved around, it's intended to be used at a desk and plugged in.

Which is to say that it's only a high-midrange portable workstation, not top-of-the-line, but among laptop workstations that can actually be used on battery it's fully competitive with the best Dell has to offer today.

 

[Bodhitree]

Oddly, when I look up other benchmark results for the same i9 CPU as I was testing, the Geekbench numbers are similar (often worse), but the Cinebench is supposed to come in right around 12,000. Not sure why I'm getting erratic performance there, but I've got to assume it has to do with thermal throttling.

It's well known that Cinebench is not in line with the rest of M1's benchmark numbers, if anything I would expect it to be out of line. Why your particular setup is generating numbers not quite in sync is anyone's guess though ?

 

[Darkseth]

These numbers become way more impressive considering how much Watt the Chips draw during those Benchmarks.

 

[treehuggerpro]

Aside from software optimisations, I'm still expecting the GFX numbers will scale a little better on the 32 core M1 Max, once it finds its way into some desktop machines. There's some decent observations in this Max Tech video for why the 32 core option didn't deliver quite as expected in various scenarios:

Capped power consumption, for battery life, seems the most likely explanation? Not an issue for desktops.

 

[Makosuke]

It's well known that Cinebench is not in line with the rest of M1's benchmark numbers, if anything I would expect it to be out of line. Why your particular setup is generating numbers not quite in sync is anyone's guess though ?

I'm talking about the Dell with 11th gen Core i9, not the M1 Max.

The Cinebench performance of the M1 Max may not be quite as high relative to Intel 11th generation chips as it is with some tasks, but I very consistently get around 12,000 with r23, and that's right in line with all the benchmarks I've seen online.

However, looking up Cinebench r23 results for an i9-11950H on various sites, I see results in the 12,000 - 13,000 range, which is not what I have gotten on a 10-minute test (same one I ran on the MBP).

After experimenting a bit, it's clearly thermal throttling. When I tried a 1-minute Cinebench run instead of a 10-minute one on a cold computer, I got 11,800. On a ten minute run, I get somewhat erratic results in the 9000-9800 range, probably due to moving it around a bit (varying airflow) and/or temperature at the start of the test.

Geekbench is a much shorter test, so that's probably why I get consistent speeds that match others' with it.

 

[l0stl0rd]

I think Cinebench is not that important depending on workflow, especially when it come to 3d rendering as most use the GPU these days anyway.

As far as GPU goes the 32core is about the same as the 5700 XT at least when rendering in Blender (guessing in other 3D software too).

I think that is why there is still no big iMac as in some instances a 27” inch iMac with an M1 Max would not be much of an upgrade unless it gets an M1 Max Duo or so.

 

[darngooddesign]

"A more general comment, I really like the new form factor and design, but coming from a 2018 15" it makes me feel like Apple should really start making a 15/16" "ultra Air" or something like that, more along the lines of the older ultra-thin MBPs that sacrifices some thermals, battery life, and ports in favor of a thinner and lighter form factor."

If Apple sticks anything other than the regular, efficiency-focused M# in a thin chassis, it's proof that they haven't learned anything from their past mistakes.