Wow, that seems like a rockstar score to me. What's the closest to something like that in a PC laptop?
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.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
Wow, that seems like a rockstar score to me. What's the closest to something like that in a PC laptop?
i9-13980HX? But that’s basically slightly down clocked desktop CPU stuck into a mobile enclosure. Those things draw insane amounts of power.
Yes, I would tend to agree with you on the Ethernet. I still have a load of dongles to give me GbE on my Macs. Most of the time modern wifi is good enough, but sometimes you need to have the best network speed possible (e.g. access to a fast NAS, or peer to peer data transfers) or just need to connect to something that doesn’t have working wifi (e.g. for router setup).
Yeah I hear plenty of jet noise at work - don't need it from my laptop too.
I enjoy dabbling in CUDA on my desktop PC and always thought it'd be fun to do the same while on the road, but I also want a premium laptop, not some gaming monstrosity. Razer has a i9-13950HX in their 16-inch Blade, and while its GB6 score isn't bad, it still leaves a lot on the table to the M3 Max. If I step down to the similar sized Blade 14, the Ryzen 9 7940HS also looks to be a solid performer, but is even further behind. Now of course those machines can be equipped with stronger GPUs, but I'm happy running my garbage CUDA code on just about anything, so I'd rather have a more performant CPU. Perhaps I should just ditch my CUDA shenanigans and instead learn Metal.If you’re correct and they’re making two distinct versions of the Max, wouldn’t they lose their capacity to bin the Max altogether?
They've got a broader set of differences between the binned and full Max now. Which you would assume just allows them to bin / recover more chips than they previously did.
This may also reset the value proposition when it comes to the pricing between full and binned Ultras, where the current step-up for an un-binned Ultra looks disproportionally steep compared to an un-binned Max.
Last edited:
We're going to be doing similar, but mostly because the prices in 2021 to go from 8GB to 16GB were unreasonable considering all other costs (and in one case I was just taking prebuilt specs for some low-requirement desktops). These days most business spec machines of the classes I would order come with 16 out of the box.
That also opens up iOS devices and Vision Pro to you which could be rewarding.
The need of two IDs for the binned vs full M3 Max likely had to do with the loss of memory channels, so packaging and assembly have to treat them differently. They are still the same die during fab.
I suspect there are two physical Max dies which share the same design. The smaller Max would be a chop of the full Max, with lower section of GPUs and two memory controllers removed. Apple previously used this for Pro/Max differentiation (with Pro being chopped Max), but M3 Pro has its own unique design. The M3 family must be very expensive to tapeout
I think it is no different than what GPU manufacturers do to salvage dies and cut down the memory channels, ALUs etc for the lower performance bracket.
It's probably a yield issue this generation.
The memory configurations (and upgrade prices for those) are all over the map with the lower binned M3 Max going from 36GB straight to 96GB (+$800) and the full die M3 Max going from 48GB to 64GB (+$200) and 128GB (+$1000).
It's probably a yield issue this generation.
The memory configurations (and upgrade prices for those) are all over the map with the lower binned M3 Max going from 36GB straight to 96GB (+$800) and the full die M3 Max going from 48GB to 64GB (+$200) and 128GB (+$1000).
Wait, are you saying that you think Apple taped out two different M3 Max chips?! I find that highly unlikely based on cost alone. The yields on the can-only-be-full M3 Max could be extremely costly. I don’t see the production issue of having just one M3 Max chip that is then binned for M3 Max little.
Edit: I’m talking about that I find it highly unlikely that they run M3 Max little and M3 Max big on different wafers as different chips. Maybe you implied something else.
Last edited:
That's what I was implying, yes, but I had another look at the die shot and I see that what I was thinking makes no sense. There is no clean way to chop off 10 GPU cores and 192-bit worth of memory interface from that die. So yeah, it's the same die.
MacBook Pro (16-inch) Mac15,11 M3 Max with 14-core CPU appeared in Geekbench.
Multi-core score 19025.
Seeing the "low" single-core score (3056) it might still be doing background tasks but that is a decent showing.
Multi-core score 19025.
Seeing the "low" single-core score (3056) it might still be doing background tasks but that is a decent showing.
Or reviewers / influencers have limited time on their hands and start with the most interesting specs, the low-end and the absolute high-end.
The binned M3 Max isn't too bad considering it's losing 17% of it's performance cores but scores only approximately 11% less in the multicore score.
I would need to see real-world application performance before seeing whether the non-binned version is worth it, both for CPU and GPU performance. I quite like the idea of 48GB RAM (I currently have 32GB and sometimes hit some limits), so the expected extra $500 for 2 more p-cores, 10 more GPU cores, and the extra 16GB isn't awful value.
It means Apple didn’t offer the option of reviewing the M3 Pro. That’s all we need to know in respect to how “well” it could have scored.
That's your assumption; there's no reason to believe it's that actual truth.
Like I said, it could well be (and IMHO probably is) a deliberate chop from a common set of Max mask sets.
I thought so too, but looking at the Apple-provided die photos I don't see any way to chop off two controllers and 10 GPU cores with a single cut. The bottom part kind of looks like it's behind a chip line, but there are 13 GPU cores in there.
Why do you need a "clean" way to perform the chop?
The goal of the chop is to save some area. A chop does that even if it doesn't occur along a clean line.
As long as the system is designed to power-gate/isolate the affected areas, the presence of half an IP block on the chip doesn't matter.
Of course none of us know what is going on (maybe we'll eventually see? Maybe not?)
But the design is trying to balance many different issues. Minimizing the cost/area of each variant they want to ship, minimizing mask cost, maximizing yield, etc. Some things have to be compromised, and "making the packaged chip look maximally beautiful" is an obvious first choice in the compromising.
One of the reason I believe it is a binned chip rather than two separate chips is the fact that if it wasn’t binned it would mean Apple is only selling perfect dies for the max. Given how big it is, I find that unlikely.
Maybe they're saving binned Max chips for iMacs or Minis or something 🤷
Pretty sure binned chips have never gotten their own CPID before. CPID (a.k.a. ApChipID) is used to differentiate firmware.
Last edited:
I had a look at the DeviceTrees, and the differences seem pretty minor-- the M3 Max "lite" is missing a DART (IOMMU) entry related to the Power Management Processor. I also did disassembled some of the firmware, and they're nearly identical.
So, maybe the extra CPID is actually only due to the memory channel difference (for which we haven't before seen binning). Or maybe it has something to do with the forthcoming Ultra(s)? Hmm.
So, maybe the extra CPID is actually only due to the memory channel difference (for which we haven't before seen binning). Or maybe it has something to do with the forthcoming Ultra(s)? Hmm.