New Mac Studio Can Be Clustered Together for Faster AI Performance

C'mon Macrumors bot/editor... do better.

How is this different than the M3 Ultra/M4 Max Studio models that also use RDMA?

But what kind of AI would you run on it without CUDA? If you just get any old gaming PC with some nVidia GPU for a fraction of the price, you'll be running AI maybe a hundred times faster.

There was a time when Macs supported nVidia GPUs... There was also a time when external GPUs seemed to be a thing. Wouldn't it be better if you could just plug in an external nVidia GPU with CUDA and just run whatever you want on whatever you want orders of magnitude faster and for a fraction of the price?

The neural engine is great for Siri and genmoji I guess but you can't do any real work with it no matter how fast is, no matter how much RAM it has.

1.2 TB/s of bandwidth is faster than any consumer GPU available today like 4090 etc. I don;t know what you've been drinking.
Prompt processing is also many times faster thanks to their compute improvements.

A few notes that probably should have been mentioned in the article:


* RDMA applications are not limited to AI models. RDMA is also useful in areas like scientific/grid computing, big data, etc.
* RDMA on macOS was introduced in Tahoe 26.2 and should work on any Apple Silicon Mac with Thunderbolt 5. It's not limited to the 2026 Mac Studio.
* You can cluster two, three, four, or even five Mac Studios together (though if you do five, you need to use a ring topology, which significantly hurts performance since it becomes impossible to directly connect each Mac Studio to every other Studio because of the number of available Thunderbolt ports). I'm not clear on whether you can do more than five machines; the developer documentation doesn't address this. I've also read that clustering more than two systems is buggy, but this will surely improve as time goes on.

RDMA is niche for now, but it's worth pointing out the very low power consumption on these systems. Running a cluster of Mac Studios could potentially save you big money over time when you consider electricity and cooling. We've come a long way since XGrid ('https://en.wikipedia.org/wiki/Xgrid') and the Virginia Tech System X supercomputing cluster ('https://en.wikipedia.org/wiki/System_X_(supercomputer)').

This article sponsored by Rocket Mortgage ... which you'll need to afford the Mac Studio cluster setup.

"Shopping for a new Mac? We've got great rates on HELOCs. Call today!"

But what kind of AI would you run on it without CUDA? If you just get any old gaming PC with some nVidia GPU for a fraction of the price, you'll be running AI maybe a hundred times faster.

There was a time when Macs supported nVidia GPUs... There was also a time when external GPUs seemed to be a thing. Wouldn't it be better if you could just plug in an external nVidia GPU with CUDA and just run whatever you want on whatever you want orders of magnitude faster and for a fraction of the price?

The neural engine is great for Siri and genmoji I guess but you can't do any real work with it no matter how fast is, no matter how much RAM it has.

Gaming and AI are 2 completely different world. Also, AI is mainly limited by the available VRAM, even the highest end 6090 will support 32GB Max, while the Apple architecture allows for up to 512GB (minus some for RAM use).

This was possible with the previous model…