I may remember this wrong but when the first rumors about the 3mm/M3 surfaced, they suggested a significant performance improvement, i.e. more “dramatic” than the M2 improves on the M1. If Gurman is right, performance gains for the M3 will be negligible. So, would the real advance we can expect from the 3mm process be less heat & more energy efficiency, making a 27” or 30” inch iMac possible as well as longer battery times in the MacBooks?
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.
Apple's M3 Pro, M3 Max, and M3 Ultra Chips Could Offer Even More CPU and GPU Cores
- Thread starter MacRumors
- Start date
- Sort by reaction score
Agreed, and this is just an overblown issue. If you need the SSD performance as it was before, you would NOT be getting a base unit. I can edit 4K video with only 270 MB/s. I can edit 8k video at 550 MB/s. This is in line with just SATA speeds.
I hope the base ram gets bumped up to 16GB after 10 years!
Feel like the iPhone Pro should have an M chip by now, even if just for marketing purposes. Maybe they draw a bit too much power for small battery.
Your history is wrong, but also I think you're just talking about this differently than I was. I was very careful to separate out IPC gains from clock gains. IPC gains will be felt across the board; I think we're not likely to see much in the way of clock gains on the base M3, while there's a decent chance we could see substantial ones on the desktop Macs. More modest gains are quite plausible in the MBP line as well.No whining here, just reality.Historically, one processor generation is roughly 20% or so better in single threaded performance than the last generation for the chips Apple produces. That held true for the M2 over the M1 as well, so there’s a bit of a precedence here. When the M3 comes out with roughly 20% better single threaded performance, there will still be those that think M4 will have 40-50% performance. So, too with the M5 and on and on.
[...]
Right, and the “best chips they can” that meet the thermal and power envelopes they’re looking for, will be roughly 20% more performant than the M2.
In fact, the single-core performance of the M2 is roughly 12% (using GB6, a useful general-purpose yardstick) higher than the equivalent M1. 9+% of that is from clocks, showing that there are very few improvements from M1 to M2. This is a *shocking* deviation from previous years' improvements, which vary widely but were always better than that, usually much better. Shocking, that is, until you understand the TSMC fiasco.
We're going to see likely 20-30% IPC gains in the M3, *before* any clock gains. That's a MUCH bigger bump than the M1->M2.
That's fairly accurate, but they've already taken that step due to AMD. There's no performance left on the table... until their next generation. Meteor Lake is now expected in September or October. It's entirely possible that they won't be able to beat the base M3 single-core even with their most cranked-up KS chip, depending on how much IPC boost Apple manages to pull out, and how much Intel does. It's definitely a race. And Apple can almost *certainly* beat them substantially, if they're willing to boost clocks. If they are, Intel has no chance in the next generation, and probably not even in the following gen.Happily willing to be wrong there, but Intel has shown a very strong desire to NOT lose that crown, by any means necessary.
Note that that 13900KS chip I'm talking about is a desktop-only part, clocking up to 6GHz (though it can't sustain that), that really wants water cooling and draws up to 360W - probably even more if you are really good at the cooling game, I didn't check. The M2 Max is (apparently?) under 50W, and even the most upclocked M3 (one that easily beats the forthcoming KS-type chip) is likely to be under 100W.
What makes Apple constrained to only making laptop SoC's? And besides, the Ultra chip isn't in any laptops.
I mean, the new Mac Pro is a complete joke! What an utter embarrassment.
This is just silly. There's way too much stuff in the Mx chips that is useless on the phone. And yes, they're too big and too power-hungry. If that weren't true, that would show that Apple had made some big mistakes. But they haven't.
The A series chips have been embarrassing the entire mobile chip industry for a decade. Nothing has changed. Marketing doesn't need M chips.
As an oldie who understands computers get faster by Mhz , what does more cores do? Is it the same thing as that core2duo thing where you have 2 processors and that split the work between 2 processors working together instead one fast processor?
how many "cores" are they going to push into these things we are already at 32? so 32 processors?
how many "cores" are they going to push into these things we are already at 32? so 32 processors?
You are correct, all chips are binned, but many refer to full-die chips as un-binned as a generalization...?
Apple holding the Mn series for the iPhone Ultra...?
yup parallel work
think of washing a pile of dishes, it's 32 people washing the dishes at the same time
Or 31 washing the dishes and 1 moving the lawn at the same time
Huge efficiency of Mx is due to its architecture, not node. This is also why we see ARM in mobile applications, and never x86. But each node step just widens this gap by a lot.
For me having wrists as low over the table as possible is the best, the slight tilt of keyboard (it is like 2mm, much less than in keyboards) is also good. Personally I prefer wedge design.
Good to hear about this. Apple silicon is getting even more powerful!!
No one expects a person who owns an M2 would upgrade to an M3. That would be almost pointless.
The person buying an M3 is likely upgrading from a 3 to 6 year old Mac that would still have an Intel chip inside.
people with m2 max 12/38 going to m3 max 16/40.....not sure if it's worth it.
This will not be sufficient to even catch up to last years nNidia /AMD performance. ☹️
Was hoping for more, now gaming (VR) is coming to Macs.
Was hoping for more, now gaming (VR) is coming to Macs.
M3 follows m2 , pretty radical I must say
What next m4 or is that just too crazy an idea ?
Tbh all this ultra , max , pro stuff is giving me a headache. Any way of simplifying things ?
What next m4 or is that just too crazy an idea ?
Tbh all this ultra , max , pro stuff is giving me a headache. Any way of simplifying things ?
There really isn't a 'KS' Meteor Lake desktop chip coming.
Intel's canceled Meteor Lake-S CPU for LGA-1851 socket has been pictured - VideoCardz.com
Early engineering sample of Intel’s future processor spotted in the wild The very first photograph of the next-gen desktop CPU happens to be of the canceled SKU. Intel Meteor Lake-S LGA-1851 CPU, Source: @wnxod The desktop platform known as Meteor Lake-S might have been scrapped, but it appears...
Meteor Lake is primarily all laptop. From the reports it appears that Intel has tuned the core design to not support heavy duty overclocking so that they could do a better job in the laptop space. ( Which means a ton of tactical and strategic sense to save market share in the laptop space over the intermediate term. Lose most of that and they will be in extremely bad shape. Going to match M-series in efficiency? No. Does it have to? No. Mainly needs to compete with AMD (which is also getting incrementally better). ). Meteor Lake will get the new CPU branding. There is a "Raptor Refresh" coming this Fall (labeled 15th gen I think with last gasp of the old branding. ). And if things don't slide Arrow Lake S in late 2024.
https://videocardz.com/newz/intel-n...o-feature-3mb-cache-per-each-performance-core
As for Intel having "no chance" ... probably should wait until people put their SoC out to get actual measurements before declaring the contest over. Zen5 isn't going to arrive 'empty handed' in single thread performance either next year. Apple is behind on ST. So a substantive chunk of any gains M3 makes is going into the "catch up" category; not "pull ahead". If AMD/Intel move forward that is even more that has to go into the "catch up" category.
Apple should generally stay out of the 4090-killer and Desktop-killer smack talking business. They may get some corner case wins from time to time, but they are going to lose there from time to time too.
If Apple can get 20-30% without clock gains, then I’d say it’s even more likely we won’t see any significant clock gains, likely just staying in the 3.4 to 3.6 range again. But, hey that’s just me. I may be completely surprised, but one thing I won’t be is disappointed
Gurman really didn't say much substantively about performance one way or another here. There is no substantive indication of negligible performance increase/decrease here at all. The clocks , memory hierarchy bandwidth , number of instructions processed and completed per clock , etc all play just as critical , if not more, roles in performance measured.
All he outlined is the number of cores. Core count is at best an indirect indicator of performance. Also outlined more memory ( which tends to be helpful in performance if taking 'hits' because data you need is still on disk. Micro data sized , drag racing tech-porn benchmarking aside. )
The problem has been all along of these early hype over N3 has been that shrinkage is going to bring big jumps in core counts and big core count jumps would be "more power" . All the while ignoring that not everything was actually shrinking. I/O ( signal communications off die ) and SRAM ( cache) were not going to radically shrink. Dramatically increasing the core counts without also substantively increasing the aggregate cache amount isn't going to buy you much. So the mania about huge count increases was always a bit disconnected from the reality.
Apple has a 'unified' SoC die with several types of computational cores on it ( CPU , NPU , GPU , etc.). N3 brings increased density ( so bigger transistor budget). Well have multiple consumers that all want a much bigger budget. Most are dragging along caches which really aren't going to shrink ( if the total die size isn't going up that means the caches are even more constrained on budget increases). If all the types of cores get bigger budgets then the 'extra' budget will rapidly disappear as the increase gets divided multiple ways.
Furthermore Apple's designs are abnormally cache 'heavy' ( bigger caches ). That buys thems wins on performance/watt in many cases , but it isn't a 'free ride'. If the cache doesn't shrink then it has an increased liability also.
Throw on top the M2 generation bloated out the die size and N3 are even more expensive wafers... Apple is extremely likely looking to go back to M1 die sizes ( or smaller). So the transistor budget is compressed that way also. ( The transistors are smaller but the total die allocation is also smaller. Smaller stuff in a smaller area also tempers some effective HUGE increase in core counts. )
N3 is going to help deliver substantively better performance. But its characterization in rumors as magical panacea that was going to explode core counts larger and be better in every dimension/metric possible is way overblown.
the M2 Max /Ultra could likely fit in a iMac Pro chassis with few major modifications. Apple didn't/doesn't need N3 to do a large screen iMac at all. Apple likely isn't doing a large screen iMac largely because they don't see the need to.
The 24" iMac waiting on a M3 likely was far more likely because Apple didn't want to put money/effort into a M2 one. Fewer updates ( and more time in comatose state. Like Mini 2014 -> 2018 , MP 2013 -> 2019 , etc. Only the Mini isn't in comatose resource allocation 'penalty box' anymore. Apple has shifted priorities. ) .
Unit sales ... what most people buy. (i.e., money. and a desire to get a targeted return on their investement).
The vast bulk of Mac sales are to laptops. So Apple's SoC , which are only going into Macs, are focused on what they mostly sell.
Not just Apple-thing though. The vast bulk of Windows sales are to laptops also. It is just a bigger pond. Apple has around 10% of worldwide market and Windows 90%. 5% of 90% is 4.5% (approximately 5% ) . 5% of 1% is 0.05% ( approximately 0% ). A very small niche of a small niche is as Steve Jobs once put it "Nobody buys it".
Those two are not entirely decoupled. The ultra is a "not so great" chiplet design mainly because it is hobbled to being optimized to being a monolithic laptop die. It is only pressed into a 'side hustle' of pretending to be a chiplet as a tacked on 'add-on'.
The Intel Mac Pro was extremely dependent upon Xeon SP ( server ) sales of that base silicon to be viable. There are no Lunix/Windows Server unit sales to carry the bulk of the volume for the M-series based Mac Pro. Apple going with their own home grown chip means they are completely detached from the Arm server market ( Amazon and Ampere are doing reasonably well there. No real 'hole' for Apple to fill there even if they were in the server business; which they are/were not even in the Intel era when it was even easier to do. ) .
The new Mac Pro isn't a joke. However, it is not looking to the past for inspiration either. Hypermodularity isn't a focus. Old application and driver architectures from 10 years ago are not either. Apple is extremely focused on optimizing the Apple GPU stack. Period. That has impact across just about all of their product lines ( not just Macs but even in Mac space ... the vast bulk of all software is deployed were Apple GPU is extremely dominant; laptops and the lower 'half' desktop space. ). There is a reality of who is actually paying for the work. ( those systems are where the revenue basically is. Money 'talks'. )
The Mac Pro isn't going to cover the same workload space as it covered before. But it also isn't a 'joke' small space either. It really wasn't relatively all that large to begin with anyway.
Gaming not = VR. Lots of folks spin that to be true , but it really isn't.
But besides that apple is more focused on AR ( mixed reality) than on entirely virtual ones. So again even more so not = gaming.
Are there going to be games? Yes. Is it Apple's #1 primary objective? No.
Is Nvidia/AMD dominating in untethered VR ? Nope. So in that area Apple is behind the curve how?
As for the Mac deployments. Apple has been telling folks at WWDC to use their rendering foundation libraries for several years now. Very similar to how the spent more then several years pointing at ARKit ( and SwiftUI) before the VisionPro arrived (and all the apps that put in the effort to tap those libraries pretty quickly have reaped rewards). Pretty good chance that Apple spent more effort into making 'bigger/better' GPU cores than in trying to goose the 'core count' higher. So some perf/watt efficient hardware acceleration of their rending libaries constructs is likely in this update. Is this gong to be a 4090-killer. No. Is Apple GPU M3 generation going to be competitive with affordable 40x0 models. We'll see. Similar for AMD 7600-7800 range.
Is it going to take a bigger 'hammer' to old (or unaligned with their libraries) code? Probably not. But carefully optimized stuff the jury should be out until actually see what they are delivering ( including the memory type/speed/bandwidth. That is a major relevant portion also that is absent from the 'rumor' that spawned this thread. )
P.S. for what AMD , Nvidia , and Apple label as 'Cores' is so substantially different talking at the level of what Apple's GPU core count isn't very deep. At least deep enough to make decent judgements .
You mean to say future technology is going to be more powerful than that of todays? A...mazing.
I would love an Escape key on the Magic Keyboard! Function keys would be excellent, but please...an Escape key.
Hah, I'd completely forgotten about that and raptor refresh. Shows how little time I've been spending in Intel World lately.
Raptor Refresh will have almost no performance improvement over Raptor, though the higher performance will be a bit more broadly available (?). Even clocking up to the rumored 6.2GHz, with IPC gains of 1-3%, I expect we'll see <5% improvement over the 13900KS. Not until Arrow Lake will they do substantially better, and that's quite a ways out in 2024, well after the M3 ships - though perhaps not the M3 Pro/Max.
I stand by what I said, which was that they had no chance *IF* Apple also boosts clocks substantially. I don't know that they will, though I think there's a good chance they will for the desktops. I think it really depends on whether they spend the effort on giving their cores enough headroom to hit clocks over 4GHz. As to whether they actually will do that? Who knows. There's a tension between power efficiency and the capability to run at higher speeds, but they've been heading in that direction slowly anyway. We'll see.
As for why there's no chance... the math was in my previous post. It's pretty obvious. It all comes down to will, and choices. What is Apple willing to do? I think we'll know soon.
I would tend to agree about the 4090. In the near future Apple's GPUs may be as good or better than nVidia's, by some or many measures, but the people who care about that the most are generally using code not optimized for Apple, so theoretical/optimal cases don't matter to them. The one plausible exception to that I can see right now is AI - if Apple can ship large-memory chips cheaper than nVidia, it may be attractive to some people to try to use them for AI. (Or even if not cheaper, because of supply constraint issues.) Oh, also, the whole VR/AR thing opens a can of worms; I don't have enough data to talk about that though.
As for CPU comparisons, maybe not. If Apple decides to take the single-core performance crown, I don't think Intel or AMD can do anything about it for the next couple of years. Though they would close the gap a lot by the end of 2024, maybe early 2025. (Intel still has to prove they can ship Arrow Lake; new process and chiplets...) AMD has perhaps a clearer path forward, as they're on TSMC entirely, but it seems extremely unlikely they could catch up to an M3 running at 4.5GHz with Zen 5. They're behind Intel now for single-core, though ahead of the M2.
Again, there's a lot of hypotheticals here, but in a couple months or so many of them should resolve to known facts.