I'm hoping Apple makes some GPU/NPU improvements to make the Max chips more competitive with nVidia for AI workloads. The main benefit with AS now is (potentially) just having tons of memory to throw at the problem -- up to 128 GB vs a max of 24 GB for consumer GPUs. However, I believe the memory bandwidth isn't really competitive with discrete GPUs and apple has made some strange hardware and driver choices (no support for fp64 or fp8, bf16 is slow). I've been enjoying generating with Flux.1-dev on my M3 Max, but it is very slow (about 8 minutes for a 1MP generation using a specialized ODE solver).
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.
MacBook Pros With M4 Pro and M4 Max Chips Reportedly Being Mass Produced This Month
- Thread starter MacRumors
- Start date
- Sort by reaction score
Yep. It’s so backwards. Yes Apple updated MacBook Pro before desktops with Intel. But never has an X+1 generation Intel laptop chip beat Apple’s best Intel desktop chip X or even X-1 generation.
Just got an M3 Max haha. But it’s more than fine for my needs so I wouldn’t have wanted to wait 2-3 months to get a small speed/efficiency boost that I don’t need. I also don’t think I would have much use in this device for the more powerful neural engine if it’s just for AI, but there’s a lot of unknown there so that remains to be seen. Either way, I think I’ll be ok.
The caveat with that is that in a MBP, the M4 Max will throttle its speed on CPU intensive tasks that result in significant heat production. The Studio form factor has better cooling with fans and so will not throttle as easily. The M4 Ultra Studios are going to be nice machines.
2022 M2 13" MBP with 24GB/1TB, I think I'm good for a while. I am reserving the right to change my mind, but I'm torn between a 14" MBP or a 15" MBA. Has to be an M4 minimum, but I'm thinking if I wait for AI to shake out in Sequoia and until the next release of macOS in 2025, I might just skip M4 and head to M5.
I’ve been running VMware Fusion on a M1 Max with 64GB RAM with Windows 11 ARM. It’s not native performance, but Autodesk Inventor runs great. I also have a few windows only utilities that run under x86 emulation with no problems. So, not as seamless as the old days but perfectly usable depending on the specific needs.
Every time I adjust my volume now, I have to bash at buttons with my neaderthalic club. I used to tap hold and swipe, sooo elegant
Last I checked, the M1 single core score is STILL in the top 50% of all processors measured!
Yeah I'm thinking the same... I would LIKE to have a newer one... just to be new...
But the M1 Pro Max is still doing very well. Maybe I just need to take it in to get all the dust cleaned out from the under the keyboard! They are already competitive as long as what someone’s trying to do isn’t essentially just process proprietary nVidia workflows. Apple will never be as fast as nVidia at CUDA and will never have as deep of a native library as CUDA. Unfortunately, most of what’s out there is CUDA or best with CUDA, so many are still wrapping their heads around how to utilize that on a system like Apple Silicon. Performance is lacking, but it’s more of a software problem than anything, and software problems eventually get figured out.
naw man. I tried to offer it to my mother, she cried out "I have no son!" and slammed the door in my face.
Could be that I wasn't at her house though, and not my mother, or even a woman, and was just some guy. I'm sure I can blame that on this useless piece of aluminum somehow though.
I bet these will be a disappointment. Probably will have the same ridiculous SSD upgrade prices. hard pass if no Wifi 7 and TB5
I never got to use it, but I think I would have found it very handy.
I'm also waiting for Apple to resurrect 3D Touch and Touch ID on iPhones, and smaller iPhones
.
No way there'll be TB5.
M4 shows 25% faster multi-core, that's nothing to sneeze at. And if the rumors are true that they're finally starting at 16gb, that pretty much ensures that you'll get more RAM for your buck across the board.
I don't regret jumping on the M3 Max when I did, but folks who decided to wait are going to be very happy.
Has TB5 actually been implemented by Intel yet? And are they making discrete controllers available to OEMs?
It’s been a while since I’ve checked in on the status of TB.
dunno about the status with Intel, but normy M4 doesn't have it, so M4 Pro/Max/Ultra won't either.
theres no m4 out on the market designed to drive a monitor at 5k+ with promotion and uncompressed which would be the primary benefit of tb5 in an m series device.
I get it, but the main target of the CPU core design is now the iPhone. They start with that, scale it down for the Watch, and scale it up for the iPad and Mac. They’re not gonna design a core for the Mac Studio then scale that down. The cost wouldn’t make sense. They might make the Mac Studio more higher-end by increasing the clock, giving it more cores, etc. Generally running it at a higher wattage. But they’re unlikely to launch that first, and they’re unlikely to launch CPU core design changes on it first.
Doesn’t matter, it’s about chip design.
Every Mx Pro, Max and Ultra has been something akin to “the same chip with more cores glued on.” You can see photos and X-rays, the underlying architecture is identical.
Thunderbolt is basically the PCI bus, it’s intrinsic to the architecture. m4 with Thunderbolt 5 would be different enough to justify calling it an M5.