What would make you upgrade to the M4 Macs?

[PoisonTheWell]

Loaded 2017 27" iMac here. I'll upgrade to the M4 as soon as it's released. I'm not investing money on an M2 Studio Mac if it's mid/end of cycle. Fingers crossed for a surprise at WWDC.

 

[MallardDuck]

True but in some things M3 max is about as goot as a 4060 which is not bad for an integrated GPU. It they get it to 4070 or even 4070Ti level with M4 it would be great.

Assuming the software supports it. That's the thing, it's a niche market...will we see the same investment in software support for Apple Silicon we see for Nvidia?

 

[l0stl0rd]

Assuming the software supports it. That's the thing, it's a niche market...will we see the same investment in software support for Apple Silicon we see for Nvidia?

Yes I know considering how long Apple silicon is out some software support is still lacking or not as optimized as it could be.

Stable Diffusion is a good example, the difference between a M1 Max 32 cores and a M2 Max 30 cores is non existent.

Yes it is 2 cores less but considering differences in some other software it should be different.

Take Blender for example with is optimized with the help of Apple, those same chips, the M2 is about 50% faster in rendering.

I have come to the conclusion that WWDC will decide quite a bit if I bother getting an M4 or not.

I just hope they don’t pull an MS and do something like Recall.

 

[ChrisA]

What would make you want to upgrade to the M4 line, especially if you're already on M1 or M2?

I am just starting to develop and run some AI software using the M2-Pro. In the past, this was exclusively done using Linux on Intel with Nvidia GPUs. So far the move to M2-Pro is going OK.

I will consider upgrading the M2-Pro after two things happen
1) I am convinced the move from Linux/Nvidia will work for larger AI models and
2) I am convinced the upgrade would make a qualitative difference in what I can do, not just an incremental speed bump. I would need to be able to do something I can't do now.


In my case, I need two things to make #2 happen first much faster GPUs or many more GPU cores. And at least a doubling of performance but > 4X would be more compelling. and I'd also need much more RAM. I'm finding that 16GB is enough to run a speech to text model and a 7 billion parameter LLM at the same time with good-enough performance. I was surprized this works so well. But I am using C++ and "metal" on the M2-Pro.

I will wait and see if there is an M4-powered Mac Studio with at least 32GB of ram or perhaps 64GB. But there is no reason to upgrade until my software is limited by the current hardware and that is not yet the case. Today my progress is limited by my brain-power. I've not yet run into a hardware limit.

It might be the case that I don't hit a true hardware limit until the M6 is out and who knows if by them Linux and Nvidia might be well out in front again.

 

[Flowstates]

I am just starting to develop and run some AI software using the M2-Pro. In the past, this was exclusively done using Linux on Intel with Nvidia GPUs. So far the move to M2-Pro is going OK.

I will consider upgrading the M2-Pro after two things happen
1) I am convinced the move from Linux/Nvidia will work for larger AI models and
2) I am convinced the upgrade would make a qualitative difference in what I can do, not just an incremental speed bump. I would need to be able to do something I can't do now.


In my case, I need two things to make #2 happen first much faster GPUs or many more GPU cores. And at least a doubling of performance but > 4X would be more compelling. and I'd also need much more RAM. I'm finding that 16GB is enough to run a speech to text model and a 7 billion parameter LLM at the same time with good-enough performance. I was surprized this works so well. But I am using C++ and "metal" on the M2-Pro.

I will wait and see if there is an M4-powered Mac Studio with at least 32GB of ram or perhaps 64GB. But there is no reason to upgrade until my software is limited by the current hardware and that is not yet the case. Today my progress is limited by my brain-power. I've not yet run into a hardware limit.

It might be the case that I don't hit a true hardware limit until the M6 is out and who knows if by them Linux and Nvidia might be well out in front again.

I'd bet good money on the fact that we are going to start seeing extremely optimised lightweight models pop up around Q3-Q4. Apple's gambit seems to be the ability to run rag and few shot prompting on device.

Seeing how I can already load 7b models on my iPhone. Running some 4*7B MoE on the silicon seems to be what is going to be aimed for. Anything else, might still be offloaded on servers, which means, de facto nVidia.

 

[innerproduct]

Got a m3 max mbp16 maxed out last year but was originally holding out for a real macpro with the next gen 32” + display. Still waiting for that. A m3ultra or m4ultra maxed out would be good enough for an upgrade but not really sure since the m3max solves my actual needs acceptably. I am very seldomly restricted by the m3max but when I am, i need a lot more (3d rendering , ai development. For these tasks I still use PC with appropriate configs. ), so all in all, Mac Studio with a m3ultra would be enticing while some smart solution where I get 4x a m3max gpu with be an instabuy

 

[Matt2012]

...Still waiting for that....

Seems to be the story of my life with Apple