chucker23n1
macrumors G3
It is. The M6 has the fastest cores.
And the M5 Ultra has the most cores of any M5.
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 Debuts M5 Ultra as Most Powerful Chip Ever
- Thread starter MacRumors
- Start date
- Sort by reaction score
It is. The M6 has the fastest cores.
And the M5 Ultra has the most cores of any M5.
mode11
macrumors 68000
mode11
macrumors 68000
These sorts of comparisons are meaningless. The early versions of anything (e.g. CD players) are super expensive, then quickly come down in price. Current AI-driven pricing madness excepted.
mode11
macrumors 68000
Apple don't manufacture their CPUs (SoCs) - that's TSMC's job.
They could design their own RAM and NAND chips, but what's the point? It would make them even more expensive.
The M6 has fewer Super Cores CPU than the M5 . It is going to shine on hot rodding single thread drag racing . But it still has lots fewer cores than the M5 Pro/Max ( except for NPU cores ) . For multithreaded workloads each thread getting its own core is usually better than time sharing on half as many cores.
The cores are in pure tally faster ; not twice as fast.
Of any M6 also since reportedly there is no M6 other versions coming.
chucker23n1
macrumors G3
Nothing you said negates what you've quoted. Each M6 core in its respective tier is faster than each M5 core in its respective tier.
As for having fewer super cores, that comparison is tricky, because the M5 only has two tiers, whose names were retroactively changed to super/efficiency.
I'm sure
- for higher-end use cases (heavily parallelized), you'll want the Pro/Max/Ultra, just as before
- for almost all use cases, the M6 will beat the M5
- in some use cases (poorly parallelized), the M6 will even beat the M5 Ultra
So to get back to what I was responding to:
For the average person, the M6 is the better SoC than the M5. As it should be.
I don't know what you're responding to here.
Right.
It's not perfect (as I acknowledged), but my using inflation adjustment isn’t meant to claim that maturation of technology and scaling of production aren't real. It’s a way of contextualizing the price in terms of relative purchasing power between then and now. A consumer or business spending $2500 in 1984 is like us spending $8000 now, regardless of all the technology and other stuff behind a product. The fact that new technologies often become cheaper over time doesn’t make that comparison meaningless; it’s why separating purchasing power from technological/production maturation matters.
Also, stating the “new technology was just unusually expensive” explanation only goes so far: Apple’s corporate gross margins were already around 40-50% in the early 1980s[1], versus about 37-40% on products today. That doesn’t tell us the Mac’s specific manufacturing cost at either point, but it does show that Apple was already a high-margin hardware company then and prices then were not just new technology being unusually expensive.
Reference
1. This is in Spanish, but has data and sources: https://www.scribd.com/doc/169830008/Caso-Apple. I used to be able to link directly to a New York Times archive that showed Apple's 1980s gross margins, but that's no longer publicly available as far as I can tell. Otherwise, I'd link to that to back up my claims.
chucker23n1
macrumors G3
Apple hasn't been in the manufacturing business in 20-25 years.
To what end would they do that?
- to uniquely improve RAM compared to the competition? RAM is RAM. There isn't really that much to it, unlike with CPU/GPU/NPU cores.
- to increase competition, driving down prices? Apple isn't a low-cost manufacturer, and even if they were, it would take years to get such a factory going.
mudflap
macrumors 6502a
The Macintosh IIFX would like a word with you,
Macintosh IIfx 4/80: 4 MB memory, 80 MB HDD. US$9,869.
And you still need a video card. 🫣
Details on Wikipedia.
Last edited:
Howard2k
macrumors 604
Have all of the smart people already purchased their computers and now it's just the dumb ones?
If someone cannot differentiate the different CPUs they should find someone to help them or do research. We don't need Apple to dumb everything down to support those people.
I'd like to hear more about the "quad die" architecture of the M5 Ultra and what that actually means. I thought at first it might be the “Extreme” version of the chip that was long rumored, but the CPU and GPU count are only doubled over the Max, not quadrupled. Hmm.
User.node
macrumors newbie
It's similar to what intel was doing with their higher performance/server chips. the high compute chips were released later after the mainstream chips were released. I guess it's a way to let the manufacturing of those nodes mature because stability was required.
A slightly beefed up M5 Ultra and it costs almost $11,000 USD....and still only comes standard with 1tb storage. At a loss for words over the greed of Apple and their storage on products.
That's a feature, not a bug: helps keep less-than-informed customers from shelling out for the Pro/Max/Ultra by "accident"
CWallace
macrumors G5
There is no financial benefit to Apple to do so and absolutely no financial benefit to Apple's customers. Building a FAB costs billions and takes years and needs to operate at full capacity to cover the high operating costs.
With the M5 generation Pro and Max, the CPU and GPU cores are now separate functional blocks ("dies") whereas they were all part of the same monolithic die in prior generations.
The M5 Pro / Max are each composed of two dies. Basically there is a CPU+I/O chiplet and two GPU+Memory chiplets. The Max GPU chiplet is bigger with more GPU cores and also more memory controllers (i.e., more bandwidth and RAM capacity).
" ... M5 Pro and M5 Max introduce the Apple-designed Fusion Architecture, a state-of-the-art design that connects two dies into a single SoC.... The two dies include a powerful new CPU, scalable GPU, Media Engine, unified memory controller, Neural Engine, and Thunderbolt 5 capabilities.
... M5 Max supports up to 128GB of unified memory with higher unified memory bandwidth up to 614GB/s. ..."
Apple debuts M5 Pro and M5 Max to supercharge the most demanding pro workflows
Apple today announced M5 Pro and M5 Max, the world’s most advanced chips for pro laptops, powering the new MacBook Pro.
www.apple.com
They share costs by sharing the same CPU chiplet. And the Max was growing larger and larger so the yields were probably going down a bit. It is a similar way of 'skinning the cat' that they did on the M1 Pro Max where they basically shared a design subset for CPU "half" and the Max was just a bigger die with more stuff seemlessly added to the design by "doubling' the GPU+memory subsystem. Only Apple is using chiplets and 3 dies instead of just 2 die designs.
So to get to "two Max" you now need four chiplets. Two CPU chiplets and Two GPU chiplets.
" ... Additionally, M5 Ultra features a massive amount of high-bandwidth unified memory, up to 512GB, and delivers a staggering 1.2TB/s of unified memory bandwidth that is 50 percent higher than M3 Ultra. T ..."
https://www.apple.com/newsroom/2026...for-a-big-leap-in-performance-and-ai-compute/
2 * 614GB/s is approximately. 1.2TBs ( 1,200GB/s ). Two max class GPU chiplets would delivery exactly that.
Sadly Apple isn't showing any package pictures anymore on the press releases. Will have to wait for teardowns and x-rays to get more details on how physically oriented in this connection.
Simiilarly the Fusion conneciton bandwidth lines up with what they didn't say in the M5 Pro/Max press release. Seems likley to be same connector used to make those is being used to make this Ultra. ( just perhaps something inbetween all four to make it a 4-way instead of point-to-point connection )
Doesn't negate, but now you have to explain some tricky corner cases where it isn't quite true. The inattention to detail (core counts ) of the original quote you responded to wasn't being addressed. The poster said "anything with M5 name" which would essentially mean the M6 would have to whip the M5 max to 'earn' a '6'. A 150-180mm^2 die whipping an aggregate probably 500-600mm^2 on every possible ''faster' isn't likely to happen.
That is actually what has happened all along through the M1-M6 sequence. The 'n+1' designation came because the design and/or implementation was new of lots of stuff besides the CPU core subsystem. How the system emphais balance has shifted between generations also.
ohbrilliance
macrumors 65816
Has Apple really more than doubled the single core performance in two generations or does “1.25x higher single-threaded performance” mean a 25% increase?
25% increase. If take the M1 'score' and multiply by 1.25 you get new score. If multiply by 1 then get same score. 🙂
for example. Single core Gbench 7.
Mac Benchmarks - Geekbench
browser.geekbench.com
M1 Mini -- 2191
M4 Mini -- 3277 ( 1.49 )
M5 MBP -- 3637 ( 1.65 ) [ and 1.10 relative previous ]
so all benchmarks aren't quite 25%. Apple is just cherry picking the best one they can find.
Last edited:
Nobody like Nvidia?
NVIDIA Spectrum-X Ethernet Networking Platform
The World’s First Ethernet Platform Built for AI.
www.nvidia.com
"... NVIDIA ConnectX-9 SuperNICs deliver ultralow-latency, 800Gb/s networking to maximize efficiency in gigascale AI infrastructure. Built for NVIDIA Spectrum-X Ethernet, ConnectX-9 SuperNICs accelerate data movement, optimize RoCE performance, and enable consistent and predictable networking for the most demanding AI workloads. ..."
NVIDIA Launches BlueField-4: The Processor Powering the Operating System of AI Factories
New NVIDIA BlueField DPUs with 800Gb/s throughput, NVIDIA ConnectX-9 SuperNICs and NVIDIA DOCA microservices deliver breakthrough acceleration for AI data storage, networking and security.
blogs.nvidia.com
Spark DGX has Connect-X 7 adapter built-in.
" ..Hi @iner, DGX Spark does not support infinband mode and there are no plans to add support in the future. ..."
Infiniband?
I’m curious if anyone has tried to use these with an Infiniband switch?
forums.developer.nvidia.com
Yes they have Inifinband solutions for 'midrange' also, but "nobody" would mean every single possible implementation.
ohbrilliance
macrumors 65816
Thanks! I read 1.25x higher to mean “125%” higher, i.e. more than twice. It’s poorly worded.25% increase. If take the M1 'score' and multiply by 1.25 you get new score. If multiply by 1 then get same score. 🙂
for example. Single core Gbench 7.
Mac Benchmarks - Geekbench
M1 Mini -- 2191
M4 Mini -- 3277 ( 1.49 )
M5 MBP -- 3637 ( 1.65 ) [ and 1.10 relative previous ]
so all benchmarks aren't quite 25%. Apple is just cherry picking the best one they can find.
neotint
macrumors regular
The angle I see is that Apple has been moving towards more vertical integration, so the full cycle is to get back into manufacture not just design. So that they can better do what the do to RAM and disk like they did with CPU, which is kind of the direction the SOC paradigm is evolving so far.
If the big pendulum is shifting back from timesharing systems (the "cloud") to local compute, and prices are already rising why not make a better complete package? Just like dependency on Intel was an achilles heel, so is dependency on the few RAM and drive makers. Time to roll your own on these too, Apple.
Also seems aligned to this, Apple's new CEO being hardware deep, that's non-trivial to me. Sure "Apple Hyper Memory" (TM) wouldn't come out until 5 or 10 years but a shift this big takes time. They certainly have the capex for it. The direction is clear, the cloud era of which big AI is a part is ready to snap, crackle, pop. The mac can only be a "bicycle for the mind" if it's independent and a thicker client. Thin client is dying, again.
Last edited: