Macs could soon take performance leadership for 3d work

#1

As I’ve stated before: I’m platform agnostic. I currently use Windows for all my 3d work, but also have a history w/Apple and own a MBPro.

I’m beginning to see that the M1 is poised to change everything, not only for entry level laptops and NUC-sized devices…but for high end computing.

Apple’s initial foray of devices is merely a tease for what is coming. These devices (small laptops and a Mini) feature four low-power cores (IceStorm) and four performance cores (FireStorm). These devices are ultra-competitive in their class.

But now consider: rumors now abound that future Mac products, such as a more compact MacPro might feature 7x the FireStorm count. 28 Firestorm cores and 4 IceStorm.

Since Apple has perfected parallel scaling, such a machine would push an already impressive Cinebench score up another 700%. With modest improvements in IPC and clock…That’s an expected (corrected) CB score of 75,000-90,000. :astonished: And similar gains are coming on their GPU side.

But it’s not just performance: it’s cooler, quieter and less consumptive of electricity. And their machines feature an array of custom dedicated chips for AI, decode/encode and (likely) for rendering.

Such a MacPro likely won’t arrive until 2022, so my Windows PCs are safe for now.

#2

Current M1 products feature 8 Core GPUs.

According to Bloomberg, the highest-end Macs coming will feature 64- and 128-core graphics processing.

#3

Am I missing something here? A 3990X (released nearly a year ago now), clocks in at over 74K on CineBench R23?

If future Apple chips are expecting to hit 12,000-16,000 on CineBench, even assuming the rest of the processor world stands still, that’s not exactly what I’d call a performance lead.

#4

I think the 12,000 to 16,000 would be the single thread score, which for the 3990x is 1342.

#5

There’s something dreadfully wrong with that 28 Xeon Platinum score.
That’s a 28 core part from 2017, that is massively outperformed by the 32 core 3970x.

The passmark score for both should illustrate that fact

Threadripper 39070x = 64,249 | Xeon Platinum 8280 = 37,575 |

The benchmark neglects to mention that there’s very cleary two processors ($6000 a pop) in there and that in single threaded applications, the Platinum’s suck. I scored a couple off of Ebay, built a machine and promptly dismantled it. Thankfully I was able to get a quick sale on the procs with minimal losses.

As far as the OP’s post about Mac performance. While Mac neglects to support anything meaningful on the GPU side and while AMD are making the gains they are. I don’t foresee Apple taking any kind of leadership, any time soon.

They may make a viable alternative. But come this time next year, the Threadrippers are going to be superior to what they are today, and they’re already quite impressive. I’m running a 32 core for productivity and a 64 core for rendering. And honestly, I could stay at this power for another 2 years without any complaints.

#6

I’m as big a Mac fan as they come and I tend to agree. I don’t see how Apple has suddenly created some magical chipsets that will suddenly beat the best that AMD has to offer. However, since Apple is going all-in on its own silicon, the signs are at least good that it’s not going to be left in the dust (certainly Intel chips are being easily bested). Personally I’ll believe any of this stuff when I see it, but if it means upgrading my Mac Pro to some 128-core monster in a few years, then fine by me.

#7

Xeon 8280 where released in 2019 not 2017, I agree that the Cinebench score in the link is likely from a dual CPU system.
Anyway, those CPU are conceived for server and are not suitable for workstation usage since the single thread will be limited compared to Xeon W processors and the NUMA architecture introduce latency. By using a dual 8280 (or a dual Epyc BTW) you have simply made the wrong choice.
Single CPU systems are better for most workstation tasks no matter if you use a Xeon W or a Threadripper(and yes a TR will be significantly faster than Intel at least for multithreaded tasks).

Back on topic, my prediction is that a 32core CPU form Apple will get a score near to 90.000 in Cinebench23 (top of the line 64core TR score 74.000) provided is a 32 high performance + 4 efficiency cores.
The math to get this number is quite simple, today an M1 (remember that’s an entry level CPU conceived just for ultraportable and SFF) score about 1530 in CB R23. Apple CPU advancement in single threaded performance is remarkably linear and the increase is about 25% YoY, so in a couple years we will have a 50% improvement in core speed for the same clock, that will give about 2300 CB ST score.
Multiply that number for 32 and you get about 74.000, but that’s only if they’ll keep the clock at 3.1GHZ, while realistically in a tower design with proper cooling it will be easy to add at least a 15% increase in clock speed, so we are looking at about 85.000 points, add the 4 efficiency core to the math and you can easily get near 90.000 points (with a 32 core CPU!).
That number will be surely superior to Intel offering, I suspect AMD in the mean while (2022) will still have faster CPU for multithreaded tasks, but slower for single threaded.
Then it will come a 2023/2024 model with more core, and that will beat both Intel and AMD in both ST and MT tasks while consuming less energy (remember that AMD was able to get from a 16core TR to a 64core in no time and without the financial resources from Apple).
Of course I can be very wrong, time will tell;)

All that being said, is myopic to look only for the bigger score in benchmarks since most of the advantage of Apple CPUs comes from other stuff that are much more important in everyday usage, like a tight integration between software and hardware, ethereogeneus memory architecture, dedicated neural engine, image signal processor etc.

#8

Here are ways Apple is changing the game:

Big / Little
Having 4 efficiency cores coupled w/16, 32 or more performance cores is awesome. Sip power until you really need it.

SoC
Putting key components in tight proximity yields performance gains

Targeted accelerators
If we’ve learned anything CUDA and RT change the game. Targeted hardware acceleration changes the game and Apple is doing this in a number of areas.

ARM (efficiency / >IPC)
The fastest super computer in the world is an ARM processor, not x86. And in terms of efficiency it’s not even a competition. It’s the beginning of the end for x86.

Unified Memory
x86 must shuttle memory from RAM to GPU and back again. Apple’s Integrated Memory approach is faster.

iOS apps / crossover
This is going to yield benefits w/camera apps and tech, cam, sensors, spatial rec, utilities and more. Massive software library.

#9

Forgive me. I was multiplying against a different benchmark entirely.
Current M1 w/4 performance cores = 7,760 - designed for sub-notebooks.
32 perf. cores w/out any improvement= 62,080
With likely improvement from cooling, IPC, chip speed: 75,000-90,000

Scores reference:
https://www.cpu-monkey.com/en/cpu_benchmark-cinebench_r23_multi_core-16

This single chip will likely zip past current 64 core Threadripper, and with much faster single-thread performance…and much better system optimization.

#10

For your reading pleasure:

New Apple Silicon could be death blow for Intel

Arm technology is now powering the world’s fastest supercomputer

Nvidia Brings CUDA to ARM *


*this may or may not impact Apple Silicon as it is not vanilla ARM. Apple leverages it’s Metal technology.

Side Note: Since Nvidia is also bringing CUDA to x86…Octane and Redshift could soon utilize GPU and CPU simultaneously.

Why Is Apple’s M1 Chip So Fast? (nice geeky dive)
https://debugger.medium.com/why-is-apples-m1-chip-so-fast-3262b158cba2

#11

Competition is good. Apologies for my slip up on the Xeon platinum date, the one I had was from 2017 and they were similar in spec.

I don’t care for Apple. I’ve been working for them for the past 3 months (about the 4th time too). Everybody on the zoom call seems to live in constant fear of their evil overlords; anxiety ridden; fearful of being let go if they do or say anything remotely against strict company policy - and it is strict. Needlessly and painfully so for the most mundane and innocuous of things.

Rotten company, to the core. No pun intended.

In short, even if Apple were to overtake AMD. I wouldn’t change, for both their company politics or the harsh difference in price.

#12

I hear you there. Same w/Google, Facebook, twitter, etc. There is a mental/spiritual virus that’s spreading and it’s a darkness. Orwellian.

Forget everything you learned from Jesus, Ghandi and Martin Luther King. And forget about human rights. You play by their rules…or else. You can be a turd of a person but be sure to use their words and phrases. Virtue signal out your ass.

I’m excited by engineering advances, design advances. But that’s under the shadow these days. Science, data, facts…and HUMANS are getting mowed down by agenda.

#13

I’m very curious to see how fast Octane is on the M1.

#14

Not very fast I’ll wager, but if it allows me to do look dev on my couch I’ll be ordering an M1 MacBook Air.

#15

I agree with everything you’ve said.

People are getting excited without knowing what price will have to be paid for the privilege of owning an ARM based Mac Pro.

AMD construct their whole product stack from EPYCs down to the cheapest Ryzen with chiplets which means there’s virtually zero waste in silicon. AMD sells millions upon millions of these chiplets in their product stack so they have massive efficiencies of scale. Same with GPUs, nVidia and AMD sell GPUs in the millions.

How does Apple compete on those terms with their monolithic CPUs and their own discrete GPUs (discrete GPU does not also imply upgradeable GPUs) which they’ll make in far less quantity than either AMD or nVidia. With such small quantities how will Apple offer 8, 16, 24, 32 core options? Maybe they’ll use multiple SOCs? Whatever they do the Mac market is still well under 10% of the enormous PC market which Intel, AMD and nVidia are shipping millions of units into. So how does Apple compete on price?

Apple’s Afterburner Card, a simple FPGA board for ProRes decompression is priced at $2000. How much will discrete GPUs cost if that’s a benchmark for Apple custom silicon?

By the time Pro level ARM based Macs ship AMD and nVidia will be on 5nm too, we could see a 24 core Ryzen for $800 and PCIe 5 GPUs for $1000 msrp and 64 core Threadrippers for $2.5k and a new 96 or 128 core high end version. I don’t think Apple will have any sort of advantage when you factor in Price/performance.

There is one huge advantage to buying a PC that is knowing that you’re not lining the pockets of the particularly oleaginous Tim :pray:t2: Cook.

#16

I love using Macs, but I finally sold my Mac a while back and went PC again. It seems like I’m always waiting for the next thing to come to Apple. So if they get these great chips and GPUS, How long until something competitive with the crazy RTX line comes in for 3D work?

But it feels like it’s a long way off. I can comfortably wait it out on the pc and see how the wind blows.

#17

My plan?

I’ve been waiting to buy the new AMD 5950x and a pair of 3080 cards for my 3rd PC build. Just can’t find any inventory.

I’m overdue for a system upgrade and won’t wait on an Apple Silicon MacPro.

Am glad Apple is pushing industry forward w/innovation. And hate the fact that my new PC will likely be drawing over 1,000 watts when rendering! The future carbon tribunal will find me guilty as charged.

#18

“I’ve been waiting to buy”…

just get a lenovo p620, off the shelf.
immediately available, solid builds, “no leds or stupid crap”, and not bad prices. Available right now, no pissing around. “all you can eat”
Put it to work, and move along.

#19

Paul I appreciate the no-fuss of such a proposition, truly, but from what I’ve seen such a system would be much more expensive AND involve compromises.

-I want dual 3080’s or 3090’s. Not Quadra cards as in that Lenova system.
-I want the latest 5950x as it features fast single core performance plus 32 threads
-I want control of power supply and NVME. I’ll likely be adding a third GPU to drive monitors. So I need massive MOBO and 1400watt PS, and may still need risers to connect all cards.
-Quite possibly need to go open case

And…I don’t mind doing the assembly. Came into some family money this month, but want to spend thoughtfully.

#20

Apple makes more CPUs (well, SOCs) annually than either Intel or AMD do in the PC arena, so lets stop acting like they are some little upstart (Apple makes (via TSMC) the SOCs for ~200+ million iPhones, iPads, and now M1 Macs. The entire PC market is around 260 million annually. AMD has about 1/3 of that share, Intel has the rest).

The M1 has a relatively small die size, so there is plenty of room for Apple to move on up to 8, 16, or more performance cores.

And the GPU is pretty interesting. Much better performance than typical integrated graphics, at the performance of about an Nvidia 1050. Again, the GPU cores are easily replicated on a larger die.

What is really interesting is the “terrible idea” of shared memory. That’s a not optimal in conventional PC designs, but it confers huge advantages if you can get everything (good CPU, Good GPU) on the same package. A massive amount of GPU render inefficiency comes from shuttling data from storage or RAM to VRAM. That completely goes away in a high performance shared memory design. James Orbach from Octane has already commented on those efficiencies, and indicates that Octane performance on M1 is really good…