It would've appealed better if it had 2x the memory bandwidth of the M4 Max, but the bandwidth is unfortunately on par with the old M1 and M2 Ultra chips aside from the larger memory size. At this stage, I'm waiting for ASICs and language processing units to make an introduction between 2026-2030.
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.
M3 Ultra Chip's Massive 80-Core Graphics Performance Revealed in Benchmark Result
- Thread starter MacRumors
- Start date
- Sort by reaction score
As i can remember it was said about M3 as well. And they said not every generation will have ultra. But haven’t said specifically M4. If it should be M5 it may come in two years then not next year.
And graphics performance come short of some 300k predictions.
Last, real life tests especially in AI will be important and of course any tasks that needs hoards of ram will just fly.
GPU is not only about games though. There are ML or DL, But I will not talk about it; but I will talk about my Domain 3D where many program uses GPU acceleration. Thing such as GPU accelerated rendering in Arnold or Karma XPU which predominantly use CUDA or Optix. Truth to be told, Support for AMD card is also doesn't exist or even if it exists it is very low.
Texturing also are heavily GPU accelerated. Painter & Mari both supports OpenCL & CUDA. But CUDA perform generally better.
Real time Physics are dominated by GPU acceleration whether Havok, Chaos physics in Unreal, Houdini engine in Houdini or Bullet physics in Maya. Apart from Houdini engine most run on OpenCL only Houdini use CUDA but still supports AMD hardware. Don't know how Apple GPUs perform here.
Physical or Offline Simulations also heavily depended on GPU acceleration like Bifrost in Maya exclusively uses CUDA or computation & Optix for Raytracing, Houdini use both CUDA & OpenCL, recently Houdini made a move to Vulkan as well with 20.5 release.
Now from what I have heard Houdini, Maya, Unreal or Havok does support Metal, but performance clearly lags behind Nvidia & AMD.
I haven't seen much benchmark of Apple hardware in this domain.
The jump in performance from M2 Max to M3 Max was also only iterative on pure raster performance, which is what the metal benchmark is using with GeekBench.
A couple of early takeaways and caveats for these benchmark leaks:
1. We already assumed the single-core performance would be worse than M4 Max. That's a given.
2. The multi-core benchmark in GeekBench doesn't scale linearly with cores. There are likely many apps that will better make use of the cores available.
3. The GPU has ray tracing and mesh shading. Neither of which are utilised here in GeekBench. Blender will be a better test for this GPU's new capabilities. 80 cores with hardware ray tracing should be a massive jump over the M2 ultra's 0 cores with hardware raytracing.
Finally Doom will be playable on a Mac. Been waiting for this moment ever since 1993. Knew we’d catch up with the DOS PCs
It used to mainly matter to gamers. But now with AI models being able to run locally IF they have the right resources and memory, it will effect a LOT more. And really makes the Mac not a development machine for any serious AI developer working directly with many parts of the process.
Sure, you can load some models onto a new Studio, but it will never come anywhere close to matching what a good external GPU or more focused AI PCI card could offer. Even if it caught up or beat them, with external GPUs you can just add another, and another.
But nope. Not anymore with external GPU support removed. Very bad timing for a move like that which to me shows that indeed Apple was completely blindsided by the AI wave. And the AI wave is not just all fun & games, there is a LOT of real work that can be done, for AI developers. Which sucks.
I mean it does. It's an Ultra the other is a Max. Consumers putting too much emphasis on the number in the processor name, they're just marketing names, they don't mean much else.
If they'd gone the way of Intel and just given long forgettable strings of numbers to things no one would care - because over there too their top end Xeon processors are a generation or more behind.
Well no. One is an M3, the other is an M4. Apple makes it very clear that when they have a new gen of processors, that is the best one. This is the first time they're going against that.
Yeah but I thought those (bad) days were behind us now that Apple doesn’t have to use Intel?
I disagree. It makes a lot of sense. Those in the market for a Mac Studio differ from those in the market for an iMac or Mac mini for instance. What I expect to do on a Mac Studio vs an iMac is extremely different. The needs of someone using an M2 Ultra for graphics performance is better than someone using an M4 Pro. Is CPU performance is the underlying factor than the M4 Pro has an advantage. There should be an expectation for the consumer to know the computing needs that they have. The type of consumers that need 192gb of memory are no where near the same as some that barely uses 16gb of memory. Again we are talking M3 Ultra. For my use case the M4 Max is better suited for my needs. The M3 Ultra could definitely handle my use case, but would be overkill at a price point where I can purchase two Mac Studio M4 Max's.
I guess the thing is, M3 cores are still very snappy and fast. It's not like anyone is sitting around using an M3 generation processor thinking 'wow my machine is sluggish'. In fact I don't think a Mac has felt sluggish at all since the shift to M1 and onwards. So for those who want a very fast, snappy machine, AND the most multicore performance, AND the most GPU performance, AND insane amounts of RAM, the M3 Ultra studio is STILL a very good machine.
Personally I am currently running an M3 Max fully maxed out MBP16 and if I'm being honest with myself, I only want a new M4 Max Mac Studio simply because I like to have the fastest thing I can get. I do a lot of multitasking with fairly heavy apps, but I am yet to hit the limit of my M3 Max, or my M2 Max before that.
Looking at getting an M4 Max 16/40 core, 128GB RAM and 4TB SSD.
I have had 64GB RAM on my last 3 machines (27" iMac 10 core i9, M2 Max MBP16, M3 Max MBP16), and I feel like I could get away with 64GB again, however I would like to keep this new studio for a longer time.
Personally I am currently running an M3 Max fully maxed out MBP16 and if I'm being honest with myself, I only want a new M4 Max Mac Studio simply because I like to have the fastest thing I can get. I do a lot of multitasking with fairly heavy apps, but I am yet to hit the limit of my M3 Max, or my M2 Max before that.
Looking at getting an M4 Max 16/40 core, 128GB RAM and 4TB SSD.
I have had 64GB RAM on my last 3 machines (27" iMac 10 core i9, M2 Max MBP16, M3 Max MBP16), and I feel like I could get away with 64GB again, however I would like to keep this new studio for a longer time.
The Ray Tracing on the M3 Ultra that is lacking on the M2 Ultra is a big deal. At this point, No one should get an M2 Ultra to try and save a few bucks if you will be rendering Blender or Maya files. That would be a terrible decision.
Having said that, the overall GPU performance gains over the M4 Max are still kinda lacking for Blender work with that hefty $2000 markup. I think all signs point for me skipping the Ultra this year and waiting for a more modern Ultra, as the M4 Max alone is already nipping at it for substantially less, and the M5 Max later this year probably making it an even worse value over the M4 Max for anything other than running LLMS
Having said that, the overall GPU performance gains over the M4 Max are still kinda lacking for Blender work with that hefty $2000 markup. I think all signs point for me skipping the Ultra this year and waiting for a more modern Ultra, as the M4 Max alone is already nipping at it for substantially less, and the M5 Max later this year probably making it an even worse value over the M4 Max for anything other than running LLMS
Last edited:
M3 Ultra 60-core GPU should murder M2 Ultra 60-core in anything that supports Metal's RT acceleration, which got introduced in M3. So your choice of that chip could be well justified, depending on what you plan using it for. Games-wise, I'd wait to see how Cyberpunk Metal port performs on that.
Not to mention their dynamic memory allocating too, so yah real world applications will probably show a big jump.
I was convinced by a friend to keep the 60 core order and not switch down. It's exactly what I would have ordered had they refreshed with M3 last year, and even a smaller 16% bump over the M4 Max is still around the level of improvement between the M3 and M4 putting me about a year ahead in GPU performance which is what I want if I'll end up keeping the machine for 4 or 5 years. Once the base models catch up with that it'll be time to upgrade again.
I wonder if the M5 Max GPU will surpass the raster performance of the M2 Ultra. That would be pretty neat if it did.
I had a Mac Pro with a W6800X Duo in it, and even the M2 Ultra is faster at rendering in Blender. In real-world benchmark scenes the M4 Max is faster than the M2 Ultra by up to 2x, and almost as fast as a 4080 Super. I don't put much faith in synthetic benchmarks, only real-world render times. Some really good examples here (M1 to M3):
These M processors aren't nearly as bad as you're suggesting.
That misses the point of my comment. My comment is that having an M3 chip come out after an M4 doesn't make sense.
I fully understand that someone with an iPad has different needs than someone who wants a Mac Studio. Thats obvious.