I'm gunna upgrade so hard. Have been waiting for this one!
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.
M5 Ultra Chip Coming to Mac Studio in 2026
- Thread starter MacRumors
- Start date
- Sort by reaction score
This is exactly the sort of well-informed, comment that we could do with more of here. Thank you.lol what
pytorch runs fine
Local models work fairly well despite the drawbacks of M4 (low inference speed mostly).
M5 will hugely improve this: per GPU core neural acceleration, matrix math improvements, etc. Inference should be significantly faster depending on how high they get the memory bandwidth and what they do with the interposer and SoIC / chip stacking, if it happens this generation.
There are metal optimized libraries that work pretty well and there is an apple-funded CUDA -> MLX workflow in progress.
This machine is going to be excellent for personal and professional AI development / fine-tuning, and for ~$12k it’s going to cost what one nvidia card with 1/5 of the GPU memory does.
If you need really big data nvidia stuff you can just connect to the cloud, but for local work nothing will compare to this for a long time that isn’t 10x the cost.
I say all this also owning a 5090 that’s going into a Linux workstation next week for local CUDA work by the way.
Now, fi you were talking about Apple Intelligence… fair point! I have low faith especially given the new rumors that they are not teaming with Anthropic which is an enormous mistake, if true.
Apple’s small models are doing pretty cool stuff but they’re going to take a couple more years to be really relevant relative to the best of what else is out there now, and that is also dependent on them retaining talent. They are doing groundbreaking R&D work with on-device models particularly in RAM constrained areas, but everywhere else they are trailing pretty badly.
Once I get one more thunderbolt BUS/port I'm down, not for this, but for MacBook Pro. Any word on M5 gen port expansion? 😅
They need to release an M5 Extreme chip for the Mac Pro. Such a chip would probably outperform the best current Threadripper chips. The Mac Pro needs more differentiation from the Mac Studio. Support for eGPUs and more expandability or modularity would be good, too, especially upgradable memory.
Sadly the Mac Pro is not a priority for Apple, and now that they’re all-in on integrated GPUs and memory, I don’t think expandability or modularity are either.
I have the M3 Ultra with 512GB Unified Memory. It’s amazing for AI/LLMs. Cannot imagine what the M5 will have. A day one upgrade if it has 1TB Unified Memory.
Think it is very possible for it to be unveiled at WWDC along with the new Mac Pro and Mac Studio. Should definitely be very powerful.
Don’t wish your life away, current chips are already very powerful so weigh that up when deciding on new computer
I think he / she meant the process node, as in going from 2 nm or 3 nm litography down to A16.
Great, now put a memory card slot on the front of the Mac Mini, and the power button on the back
I think some (very few) people want to use very specific and expensive PCI cards and an old (but still quite powerful) Mac is better than nothing for them.
So, the price makes sense. It's there just for those very few ones, it has to be justified for Apple. Apple moved away from computers with any sort of customization a decade ago. Who still wants that, has to pay.
And of course it's not consumers, it's businesses. If you compare that with any sort of pro-oriented low-production machine for very specific businesses, it's not that much of a crazy price.
If it has that (M5U + 1TB) and it comes in under £4000, it’ll be a day one purchase for me, too. Lol. Sadly, I think I might be a tiny bit out on price. 😂
I’m taking the liberty of selectively quoting a couple of sentences from your excellent post, I hope that’s okay.
IMHO, the point you allude to about memory is important and although the raw performance of Nvidia big GPUs is incredible (not least for 32-bit FP and integer, as well as the lower precision AI optimisations), I do think that for use cases where it’s good to run locally (the type of things you mention; also inference, as opposed to large-scale training, where there are privacy concerns) the Apple unified memory model is a big plus.
So the idea of an M5 Ultra with 512GB or 1TB of unified memory is really quite attractive. With respect to raw compute performance, it’s pointless (from my slightly simplistic perspective) to compare Apple Silicon with dedicated Nvidia GPUs that, by design, consume way more power (and I also think that Nvidia has major advantages from a developer’s perspective) but for what they appear to be focusing on, the value proposition of something like an M5 Ultra Studio may be quite interesting.
Wasn't there a rumor that Apple was going to move away from local silicon interconnects (LSIs) that are using to bridge Max chips together in the Ultra package? I guess this would either be a single monolithic chip (high $$$) or a full chiplet design using a full interposer (e.g. CoWoS).
I bought M4 max studio with 64GB of RAM few months back. It’s a beast for me as I dont have a heavy workloads like video editing. I paid 3750 eur for it, more expensivw than M3 Ultra version in US 😬
If M5 scales properly, we're going to witness Apple m5 Ultra bypassing Nvidia's 5090 in Blender. 😊
I think the Mac Pro will get the m5 ultra, and the Mac Studio will get m5 max + m4 ultra option. but that's just speculation. Unless they are thinking of killing the Mac Pro heh
Can't wait to take a massive hit on my M3 Ultra and then pay through the nose for the top-of-the-line M5 Ultra with loads of RAM and storage. Ah well, can't take it with me...
I would so love if they could figure out a way to have a GPU that can match the RTX 5090, with the amazing integrated RAM of Apple Silicon. I don't even need gigantic CPU performance…