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 Apple Silicon Chip Reportedly Enters Mass Production
- Thread starter MacRumors
- Start date
- Sort by reaction score
I remember when 32Nm chips used in Macs were state of the art.
Yes.
I'd go so far as to suggest there will be $12,000+ Apple Silicon Mac Pros within the next year or two, and that they will sell well for the category.
The vast majority of users never "upgrade" when a new chip comes out. They only buy a new computer when they need a new computer and then buy whatever is on the market at that time.
OK, there are a VERY few users who are always performance-constrained and have never owned a computer that was fast enough that they did not have to sit around and wait for it. But most people are not AI researchers or artest working with multi-gigapixel images.
People here on this forum are not a good sample of Apple users, they (we) are the nuts who just want the latest stuff even if we have no technical need for it.
Actually I kind of miss the old days when I used to start a compile job and then get up and find some coffee, talk with co-workers, and then go back and check progress. But the 1970s and early 80s are gone and we are much more productive today.
Today I use an M2-pro 16/515GB but connected to a NAS with 10gig ethernet, basically unlimited storage (there is 11TB of free space in it right now). The use is robotics software and I have some still/video cameras I like to use but as a hoby. I can see a need to upgrade. I want to be able to run a large virtual machine with Linux while still using the Mac and also an 8B parameter AI model, all at the same time. I can almost do that now but it is damn slow if I do it all at once. So I'm looking at an "M8 Ultra with 24+ cores and 64GB RAM. But today the main bottleneck in productivity is my limited brain power, not my computer.
There is a bit of room for thermal, but it isn't massive. 10% if they go with the same power - which would be hard to sell.
If there are further AI improvements those have to come with either another groundbreaking change of the architecture (because M4 already is crazy good at that ppw wise) or memory. The latter would point to them using SoIC for some kind of on stacked memory not unlike some of AMDs Epyc chips use, but if so then possibly "under" the actual die instead of "on" the die - for thermal reasons.
But that would be relatively expensive, hence I would be rather surprised if they did it for the standard M Series variant. Remember: AMD prices in that ONE small SoIC die on the 9900X3D vs the 9900X as about 300$ extra. Doesn't mean it's 300$ more expensive, not even close. But it means it certainly isn't cheap - and that's on N4P, meaning refined 5nm. On the other hand, yes, THAT would really fix a lot of problems the M4 has with AI stuff, and frankly also general GPU tasks, and it would allow for more power to go towards actually computing stuff and less towards moving around bits.
The reason I don't see them stacking either GPU or CPU or even an NPU on top or even under the SoC, period: For technical reasons you can't do the same nodes stacked on top of each other. And since all architecture since M3 is N3 that would mean they would have to put some compute logic in 5nm structures. And even for Apple backporting either of those logic structures to 5nm isn't "cheap", not even talking about losing the advantages gained with 3nm.
Which could make hella sense, but would be hella ambitios and hence again I don't see it on standard M5: Put I/O (serdes stuff like TB), accellerators (de/encoders), DRAM interface in a 5nm slab at the bottom, and stack a 3nm slab with GPU, NPU and CPU on top - and have both slabs use the additional space gained for big caches. Like the Mall Cache Mx already has, just (much) bigger.
If they actually put that into the standard M5 ... that would run circles even around the M4. The question is: how are they gonna put that into a 600 dollar computer then and make any profit? ^^
iPad OS needs an overhaul. The devices are brilliant but it’s silly to run a phone OS on them.
Apple's chips are getting so good that you don't need to upgrade every generation. Even folks with the M1 are still hanging onto their devices.
This is not a dealbreaker. Most people aren't going to notice a difference.
This will be a nice addition. This is similar to AMD's 3D V-Cache.
Apple needs to sort out iPadOS. It will be more powerful but you can only do so much with what is available right now.
This reminds me of another rumour that says the high end M5 chips will use separate CPUs and GPUs for server use. It will be interesting to see if this plays out or not. I can't see it happening. The article can be found here.
I’m curious, what software are you running, since you’re noticing such a huge speedup?
Going from my M1 Mac mini to my M4 (non-Pro) Mac mini, there was a small but significant speed improvement for browser based activities, but it was less noticeable for my office apps, since they don’t tax the CPU very hard.
Yep. Having said that, my M1 Mac Studio, which I love, is showing its age. I bought the base M4 Mac Mini to tide me over until the new Mac Studio is ready and it blows my Mac Studio away. My Mac Studio overheats and slows down while the $600 Mac Mini chugs along without a sweat.
I’m curious, what software are you running, since you’re noticing such a huge speedup?
My friend also noticed a significant speedup from his base Mac Studio to M4 Mac mini. He’s a web designer. He is not going to bother with a new Mac Studio. The M4 is more than fine for his design work.
I also noticed a speedup from the M1 Mac mini to M4 Mac mini, but it was less dramatic for me since I run mainly office type applications.
Graphic software mostly, lately, a lot of Blender. I think if you only do basic office stuff, you really never have to upgrade.
Interesting. Somebody had me run a volume shader benchmark on my M4 Mac mini and it was very, very slow.
It performs much better on Windows with the 4070.
I guess the Blender stuff you do is less dependent upon GPU?
I don't think we can make that assumption yet.
Apple themselves have given no indication of what the future, if any, is for the Mac Studio.
I imagine something like the new Mac Mini, but perhaps an inch taller (making it more like a cube.)
Need the extra space on top for the heat sink (either aluminum or copper as with the Ultra SoC.)
Apple has indicated they are working on servers for their own farms, but I expect those to fit in standard racks.