3990X is in a different league since it's 9x faster on Blender CPU than M1 Max 10CPU and projected to be about 6x faster than M1 Ultra 20CPU which is probably more comparable to 12600K/5800X.
Last edited:
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
A bunch of new mac studio benchmarks dropped
- Thread starter dugbug
- Start date
- Sort by reaction score
3990X is in a different league since it's 9x faster on Blender CPU than M1 Max 10CPU and projected to be about 6x faster than M1 Ultra 20CPU which is probably more comparable to 12600K/5800X.
Enjoy your Blender benchmarking, then. I've never had any occasion to use it even for a minute.
Is it because Blender isn't fully optimized for Apple Silicon yet?
Apple just recently announced 1st party support for it, right?
Take a look at Stockfish chess benchmarks.
3990X = 145.000.000
The M1 Ultra should reach: 30.000.000
You are right, for once. Ultra is comparable to the 3970X, not 3990X.
Lol, the 10-core M1 is already faster than those. Ultra is on a whole different performance level. Unless of course you stay in your little fantasy world where the only blue benchmarks are those that disproportionately favor x86 hardware.
Apple Silicon CPU support was added back in Blender 2.93. What you're referring to is Metal GPU support recently added in Blender 3.1. Blender is Apple funded and has Apple developers working on it with Blender development team.
Please just stop to post Blender benchmark BS everywhere... everyone with a grain of salt and a bit of knowledge already know that is not optimized for AS and regardless nowhere near to be representative of general performance using other software.
Don’t you have better things to do with your time than regurgitating always the same stuff? Really you sound desperate...
Don’t you have better things to do with your time than regurgitating always the same stuff? Really you sound desperate...
I'll just quickly step in for a moment.
The reason why the M1 Pro has the 2.1x instead of 2.0x vs the M1 is because of the M1 using LPDDR4X RAM, whereas the M1 Pro uses LPDDR5. The slightly faster RAM gives the Pro that small bump.
Unfortunately, Andrei, who did the in-depth review of Apple Silicon machines, has left Anandtech some time ago. Recent reviews of mobile Alder Lake etc. don‘t even mention M1 chips.
Sigh... Anandtech's best writers (including Anand himself) always seem to move on to bigger and better things
Don't get me wrong, I'm happy for them on a personal level but Anandtech's deep dives are almost like an industrywide public service at this point. We need them to do what they do best!
Anyway sad to hear he's already gone after really hitting his stride but hopefully they've found someone to fill his (large) shoes.
On a different note, I wonder about Apple’s GPU power graphs comparing to RTX 3090. They say the M1 Ultra GPU outpaces the 3090 in their graph, but in practice the actual performance that comes out is a lot lower. I wouldn’t mind reading a write-up on Apple’s testing methodology for that graph, just to get to the bottom of this.
It would not surprise me if optimising for TBDR was a leading cause.
It would not surprise me if optimising for TBDR was a leading cause.
Sigh... Anandtech's best writers (including Anand himself) always seem to move on to bigger and better things
Don't get me wrong, I'm happy for them on a personal level but Anandtech's deep dives are almost like an industrywide public service at this point. We need them to do what they do best!
Anyway sad to hear he's already gone after really hitting his stride but hopefully they've found someone to fill his (large) shoes.
That is sad to hear. Other than Anandtech, are there any other sites that do proper reviews for this kind of thing? I.e. not just running random benchmarks and extrapolating wildly.
While the CPUs are probably going to be fast regardless, I think the GPUs are going to require a fair bit of optimisation to hit that performance and really curious to see what we get with something that's fully optimised for the Apple hardware.
It'd be great if Apple released a reference renderer / metal viewport (perhaps a Hydra delegate for USD as the one in Preview is a bit meh), just to see the kinds of performance you can get from a proper implementation of TBDR. I'm guessing a lot of Apps with go for the moltenvk approach, for cross platform compatibility, which is a bit disappointing.
Speaking of optimisations, I suspect that things on the CPU side are not optimal in some cases. Testing an admittedly trivial scene on my M1 MacBook Air, Karma in Houdini was running slower on Apple Silicon than under Rosetta (1m 54s vs 1m 04). Super trivial scene so wouldn't draw too much from it, but thought it was mildly interesting.
The thing is that as a consumer it does not matter if they are optimised or not.
If I use blender daily, I don’t care if it’s not optimised for AS. AS is still slower for this workload.
When buying any hardware people should look at their own workloads and look at the respective performance.
This is only partially correct, as it is not a kernel scaling issue but a "default allocation" issue. Background is, that Windows partitions the logical processors into groups of 64. With Windows 10 and below a process is assigned to a single group per default. You can change the process to be a multi-group process after creation. The default behavior changed with Windows 11.
Does any single user, small number of users, system handle excessive numbers of cores without specific programed tasks? It's not an easy thing to do. Certainly not any MacOS machines.
To think of what I'd have to do to program the scheduler myself, I'm happy I'm just a business programmer now.
Yes more than anything the GPU advantages require architectural consideration by the software. I like that apple pretty much gave an engineer to the blender team to start this. I wish apple would approach project red or some other dev team with AAA titles and pay for a significant port as a demonstration to the industry.
Demonstration of what?
Their capability? Performance is irrelevant. Switch gets plenty of ports including Witcher 3 and Doom etc.
It’s an economic issue. The number of mac owners using steam is tiny, no matter how good it is they wouldn’t sell enough to be worth their while.
Apple would have to pay several developers to port the games, for a sustained period of time to Increase the numbers to a level that will sustain developers doing it themselves.
Im not sure I understand the switch reference. Nintendo has a gaming market and reputation. Pressing apple arcade+ games to support the mac is not having an effect in the industry. If apple wants to attract established players they will need to do something like this.
-d
That’s totally fine of course
my post was targeting some users here that simply ignore the softwares where AS excels and keep posting only about softwares that favor competitors hardware, giving a totally biased opinion.Thanks for clarifying it! My impression was that this has to do with CPU mask being 64-bit and Windows processor groups is a workaround allowing the system to work with more cores.
It's not about easy or not easy, it's about how you design things. A lot of it has to do with efficiency of implementation, if you limit the number of cores the kernel has to deal with you can use fast basic data types.
MacOS from what we know is limited to 64 cores, but it will hardly be a problem for Macs, seeing how much performance Apple can extract from individual cores.
Absolutely not. We are just commenting on Threadrippers running standard software. It's not really an everyday machine.
The Switch reference makes total sense. It is all about the size of the user install base.
The WiiU - the main reason many developers avoided it as the low install base. Not enough people bought it for the developers to want to make games for it.
The internals of the console were not an issue (even though the vocal minority want us to believe that lie).
The Switch on the other hand started out with mostly Nintendo 1st party games and very quickly the customers came, so the developers were happy to develop for it.
Sony and MS were in a similar situation for their first consoles. Not a recognised force in the console hardware business. But they soldiered on, got the customers to buy their consoles and then the developers came.
The number of consoles sold matters, regardless of if you are recognised at making games or not.
Mind you Sony and MS are not recognised for making console games as they do not do this. They are recognised for bringing in and partnering with talented studios who make the games.