M2 Mac Studio Max or M2 Mac Studio Ultra

[kirkster501]

Agreed, 512G (on the MAX) as the default config from Apple is ridiculously mean, especially at current SSD prices that have halved in the last six months. Can now get a Samsung 4TB NVMe now for $200. And frankly, 1TB on the Ultra is pathetic. I get it that Apple is a business and wants to make money with upgrades but the default is insane.

 

[spaz8]

I'm a Houdini user, I don't plan to do much rendering but I plan to do a lot of simulation work and large procedural tools. I have been flip-flopping between a Maxed out MS M2 Max .. 96/38/1TB and the base MS M2 Ultra 64/60/1TB .. you get like twice the machine for the $1000 difference ( 2x cpu, 1.5x gpu, 2x NE).. The thing is I don't think I can get by with just 64 GB.. I've been doing 3d professionally for nearly 20 yrs, but this is just a personal hobby machine, to do personal projects and make demo reels. I am weighting if I can scrape by with 96 and patience. My employer supplied work machine is a 5950x, with 128GB and a 3090, and a can go into swap space with some setups at least temporarily. I did not see a ton of CPU's being pegged in my tests.

I am so used to getting the best GPU possible apple with give us, as a MP 6,1 owner with D700's I am concerned about how quickly the 38 core Max is gonna age. So part of the lure of the Ultra is future proofing as I run my personal machines for 6+ yrs. I oscillate saying F-it and just getting the M2U with 128/60/1tb .. and not thinking about it vs.. saving the $2k to get a 4090 for a PC ML build.

 

[kirkster501]

I'm a Houdini user, I don't plan to do much rendering but I plan to do a lot of simulation work and large procedural tools. I have been flip-flopping between a Maxed out MS M2 Max .. 96/38/1TB and the base MS M2 Ultra 64/60/1TB .. you get like twice the machine for the $1000 difference ( 2x cpu, 1.5x gpu, 2x NE).. The thing is I don't think I can get by with just 64 GB.. I've been doing 3d professionally for nearly 20 yrs, but this is just a personal hobby machine, to do personal projects and make demo reels. I am weighting if I can scrape by with 96 and patience. My employer supplied work machine is a 5950x, with 128GB and a 3090, and a can go into swap space with some setups at least temporarily. I did not see a ton of CPU's being pegged in my tests.

I am so used to getting the best GPU possible apple with give us, as a MP 6,1 owner with D700's I am concerned about how quickly the 38 core Max is gonna age. So part of the lure of the Ultra is future proofing as I run my personal machines for 6+ yrs. I oscillate saying F-it and just getting the M2U with 128/60/1tb .. and not thinking about it vs.. saving the $2k to get a 4090 for a PC ML build.

I think you have answered your own question then - get the Ultra with 128G or even 192G of memory. Expensive for sure but if you have those requirements then that's what you need. Only you can say for sure what you need. My counsel is that there is a difference between "need" and "want".

Or build a Threadripper PC with a mega graphics card?

 

[tiguanito]

I'm a Houdini user, I don't plan to do much rendering but I plan to do a lot of simulation work and large procedural tools. I have been flip-flopping between a Maxed out MS M2 Max .. 96/38/1TB and the base MS M2 Ultra 64/60/1TB .. you get like twice the machine for the $1000 difference ( 2x cpu, 1.5x gpu, 2x NE).. The thing is I don't think I can get by with just 64 GB.. I've been doing 3d professionally for nearly 20 yrs, but this is just a personal hobby machine, to do personal projects and make demo reels. I am weighting if I can scrape by with 96 and patience. My employer supplied work machine is a 5950x, with 128GB and a 3090, and a can go into swap space with some setups at least temporarily. I did not see a ton of CPU's being pegged in my tests.

I am so used to getting the best GPU possible apple with give us, as a MP 6,1 owner with D700's I am concerned about how quickly the 38 core Max is gonna age. So part of the lure of the Ultra is future proofing as I run my personal machines for 6+ yrs. I oscillate saying F-it and just getting the M2U with 128/60/1tb .. and not thinking about it vs.. saving the $2k to get a 4090 for a PC ML build.

For simulation works, you need to find benchmarks and compare with the 4090. Not sure the M2U reaches the power of the high end RTX cards. And as a bonus, you can play games with the 4090.

edit: here’s one compute benchmarks

The Apple M2 Ultra GPU is on par with the GeForce RTX 4060 Ti in Geekbench 5 - the RTX 4090 graphics card is 150% more powerful

The Apple M2 Ultra regularly appears in benchmarks after its presentation. We've already written about the CPU tests in Geekbench 5 and Geebench 6. Now it's time for the GPU.

gagadget.com

 

[koyoot]

For simulation works, you need to find benchmarks and compare with the 4090. Not sure the M2U reaches the power of the high end RTX cards. And as a bonus, you can play games with the 4090.

edit: here’s one compute benchmarks

The Apple M2 Ultra GPU is on par with the GeForce RTX 4060 Ti in Geekbench 5 - the RTX 4090 graphics card is 150% more powerful

The Apple M2 Ultra regularly appears in benchmarks after its presentation. We've already written about the CPU tests in Geekbench 5 and Geebench 6. Now it's time for the GPU.

gagadget.com

Stop using godawful benchmark that Geekbench is to compare anything to anything.

Its without any question the least realistic benchmark on the planet for GPU performance comparison.

 

[spaz8]

I'm mostly flirting with building a PC box for ML experiments ( pretty much a Nvidia gpu wrapped in a PC), and a Mac Studio as daily driver and 3d work.

I did a bunch of napkin math this morning to look at things from another angle - since the GPU in my mind is the thing that ages the fastest in any computer, and concerns me the most about obsolesce.

I tried to place all the Mac Studio GPU configs to Nvidia equivalents myself. Lots of guesswork. Compare M2 compute to closest AMD card, find the smallest delta of that AMD card to recent Nvidia card on userbenchmark.

Mac Studio M2U 76 core = approx RTX 3080
Mac Studio M2U 60 core = approx RTX 3070
Mac Studio M2M 38 core = approx RTX 4060 Ti
Mac Studio M2M 30 core = approx RTX 4060

** The issue with the gagadet stats is its using OpenCl, which is abandoned tech by apple. I used Metal scores of the M2's vs, the metal scores of AMD cards, and then used those AMD cards to line up to NV cards.

 

[koyoot]

I'm mostly flirting with building a PC box for ML experiments ( pretty much a Nvidia gpu wrapped in a PC), and a Mac Studio as daily driver and 3d work.

I did a bunch of napkin math this morning to look at things from another angle - since the GPU in my mind is the thing that ages the fastest in any computer, and concerns me the most about obsolesce.

I tried to place all the Mac Studio GPU configs to Nvidia equivalents myself. Lots of guesswork. Compare M2 compute to closest AMD card, find the smallest delta of that AMD card to recent Nvidia card on userbenchmark.

Mac Studio M2U 76 core = approx RTX 3080
Mac Studio M2U 60 core = approx RTX 3070
Mac Studio M2M 38 core = approx RTX 4060 Ti
Mac Studio M2M 30 core = approx RTX 4060

** The issue with the gagadet stats is its using OpenCl, which is abandoned tech by apple. I used Metal scores of the M2's vs, the metal scores of AMD cards, and then used those AMD cards to line up to NV cards.

M2 Max 38 core is actually 10-15% faster in 3DMark Wild life than RTX 4070 laptop.
M2 Ultra 76 Core is approximately 20% faster than RTX 4090 laptop GPU in 3DMark wildlife.

In GFXBench, M2 Max 38 core is 20% slower than 4080 laptop GPU, while being also 35% faster than 4070 laptop GPU, while 76 core M2 Ultra is around 20% faster than RTX 4090 laptop GPU.

 

[tiguanito]

I don't know the benchmarks. But 3DMark Wild Life seems to be a graphics / gaming benchmarks ? (for gaming I don't think M2U can be compared to a desktop 4090 really, with Ray tracing, DLSS3, Reflex, ... )

Some interesting comparisons in LR, with AI denoise using intensively the M2 GPU.

Post-processing & Printing - Photography - FM Forums

Explore Post-processing & Printing photography discussions. Share your photos, discuss gear and techniques, and buy or sell equipment in our marketplace.

www.fredmiranda.com

The M1U 64GPU 64 GB (which I expect to be very close to the M2U 60GPU 64GB ) -> 18s .
3990x + 4080 -> 14s

Of course it's difficult to compare with the x86 platforms because the CPU can impact the results as well. But I think it's safe to conclude in this GPU intensive application, a 4090 will outperform the M2U 60GPU.
I'm not looking at the M2U 76GPU 128GB as the price makes the machine really too expensive anyway. (would need to be compared with a system with 2x 4080s at least )

With that said, I prefer the small form factor of the Mac Studio and ecosystem. And the performance / watt of the M2U is insane.

 

[spaz8]

I don't know anything about Laptop GPU's, I've never owned laptop in my life, bought a couple for my wife... very different animals.. I know desktop GPUs.

A 4090 laptop card is like a 4070 desktop.

No way a 76 core M2 ultra is even close to a 4090 desktop, not even a 3090 desktop... my math says.. its a 3080 desktop. In ML training a 4090 is 2.2x faster that the 76 core M2U.. about a 1080 ti.. so Apple has a ways to go with its igpu still.

 

[koyoot]

I don't know anything about Laptop GPU's, I've never owned laptop in my life, bought a couple for my wife... very different animals.. I know desktop GPUs.

A 4090 laptop card is like a 4070 desktop.

No way a 76 core M2 ultra is even close to a 4090 desktop, not even a 3090 desktop... my math says.. its a 3080 desktop. In ML training a 4090 is 2.2x faster that the 76 core M2U.. about a 1080 ti.. so Apple has a ways to go with its igpu still.

4090 laptop uses 103 die. The same that 4080 desktop is based on.

And 76 core M2 Ultra is faster than 4090 laptop GPU that uses the same die.

Yeah, not even close to 4090, desktop. Not even 3090, desktop.

 

[spaz8]

I think all this GPU research has helped me decide to get the maxed out M2 Studio Max 96/38/1TB. I can optimize and reduce setups to get the point across and fit into 96gb and move to a PC work machine if I have to with more ram. The M2 Max is gonna be a 4x increase at least in every stat coming from my MP 6,1. The 38 core GPU (4060 ti) should be fine to tumble a few million polys around on screen.

The real "ah ha" though is the state of gpu on Apple Silicon is still not good enough even at the highest end M2U 76 core for what I would want to do with it. So it does not make sense to invest the money there. I will buy a 2nd machine PC or Linux box and put the $2K saved into an Nvidia GPU instead of wishing I could do it all on a top spec Mac Studio.

 

[Morgonaut]

my advice - if your workflow is not insanely demanding like mine then get the M2 Max MS with as much memory you cam afford and 1TB disk. Paying for more internal storage would be waste of your money, save it for external storage or the memory. In last few weeks I deeply tested M2 Ultra 192/72 vs M2 Max 32/30 and I was strongly considering to return my Ultra and get maxed out M2 Max. The real life feeling of difference is negligible but price is double.