Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Acertomacbook

macrumors newbie
Original poster
Aug 6, 2023
5
0
I've been waiting long to move from intel to apple silicon:
I skipped m2 because it was not a substantial improvement over m1

m3: Similar price to m3 pro with 16gb ram
m3 pro: no significant improvement over m2
m3 max: only chip with substantial improvement over m2 max

I need more ram for local LLM inference and maybe ios, vision os development but spending $800 for upgrading from 64 -> 128gb
1699170618541.png
but I can get 96gb upgrade for the base m3 max which just cost $100 more from the 64gb version but that sacrifices CPU, GPU cores
1699170652763.png
 

kitKAC

macrumors 6502a
Feb 26, 2022
805
759
We're now in the extended Christmas return window, so you've got time to buy a machine and test if it's right for you - if not, return it for the other.
 

ChrisA

macrumors G5
Jan 5, 2006
12,789
1,974
Redondo Beach, California
I'm working in the same area. I'm not trainning LLMs, but other ML projects.

My setup is to have an M2-Pro mac on the desk and an HP "Z" class workstation that runs Linux and runs with no monitor or keyboard in a closet. The workstation has 16-core Xeon, 128GB RAM and Nvidia GPU. I use VNC screen sharing to access the workstation. I bought this for $600 on the used market then added some RAM and the GPU.

The workstation sucks up electric power in a big way so I have it connected for a "smart plug" that is controlled via Apple's Home App. I can remotly disconnect the power.

Even if I owned a big 40-core 128GB Mac, I'd not want to use it the ML. It is very anoying to have the Mac tied up for hours or days crunching data. Buy an Intel box run Linux on it and make your life easier.
 

maxrubin629

macrumors newbie
Jul 29, 2024
1
0
I've been waiting long to move from intel to apple silicon:
I skipped m2 because it was not a substantial improvement over m1

m3: Similar price to m3 pro with 16gb ram
m3 pro: no significant improvement over m2
m3 max: only chip with substantial improvement over m2 max

I need more ram for local LLM inference and maybe ios, vision os development but spending $800 for upgrading from 64 -> 128gb
View attachment 2307370
but I can get 96gb upgrade for the base m3 max which just cost $100 more from the 64gb version but that sacrifices CPU, GPU cores
View attachment 2307371
Which did you end up going with? I'm literally in the exact same boat right now. The 30 core gpu model has lower memory bandwidth compared to the 40 core gpu version (300 GB/s vs 400 GB/s). Apple made this so frustrating.

I need more than 64 gigs of ram, but I don't want to be going for the 128 gigs just because the bandwidth is lower in the 96gb :/
I mean, it might just be worth waiting for M4 at this point...
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.