Mac16,1 - Geekbench
Benchmark results for a Mac16,1 with an Apple M4 processor.
browser.geekbench.com
M4 Mac with 16 GB RAM (not 12 GB). This could be a MacBook Pro though.
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.
M4+ Chip Generation - Speculation Megathread [MERGED]
- Thread starter GoetzPhil
- Start date
- Sort by reaction score
M4 Mac with 16 GB RAM (not 12 GB). This could be a MacBook Pro though.
This Mac M4 score is a bit faster multi-core than a liquid nitrogen cooled iPad Pro M4:
So, it would make sense that it's a MacBook Pro which has a fan. In fact, this Russian YouTube video claims that score was uploaded by them with a 14" M4 MacBook Pro.
If this is genuine, the OpenCL figure is slightly astonishing: a little over 25% higher than the 10 core GPU M3 14” MBP.
All of the scores within the OpenCL result are appreciably higher, too, so it’s across the board, too, not just the consequence of one or two optimisations (which would, in any case, tend to be ironed out by the fact that GB uses a geometric average, I think).
(The M4 iPad Metal figures aren’t so much higher, but it’s still a nice increase.)
Last edited:
Of course someone would be so incompetent to test OpenCL that has effectively been deprecated since 10.14.
Only Metal scores matter these days when it comes to macOS.
Looks like M4 will be a beast in Pro, Max and Ultra. It’s possible Ultra could be threadripper level in multi core.
They’d still have the annual early-bird launch, like M4 iPad Pro in 2024 and M2 MacBook Air in 2022, to create buzz. The only thing they’d lose would be the March Studio/Pro event, admittedly a big loss if they are also using that to focus on new displays and other expensive Studio/Pro hardware.
I believe they will do that (have an annual March event focused on the Studio/Pro ecosystem), but doing so means creating an artificial delay for the Max Studio. I’d like to see them split the difference and:
[1] Kill the Pro Mini.
[2] Launch a Pro Studio and Max Studio along with the MacBook Pro family, saving the Ultra Studio/Pro for March.
Will this happen? Probably not. But you never know, Apple has displayed some interesting agility lately…
Last edited:
Unlikely. At least for the 64-core versions. I'd love to see it, and I'm sure they could demolish Threadripper if that was their goal, but they've given no indication that it is.
The only way I can see that happening is if they build a very high core count chip for their PCC DCs, and decide they might as well release it in a Pro as well. Like I said, it would be very cool, but I doubt they will. (At least over the next two years, which is the limit to how far I'm willing to guess.)
If Apple are building their own ‘AI’ data centre - and are not farming out a massive h/w order AMD’s way (courtesy of the MI350) - (I speculate) that they will have to develop some sort of SoC package solution with a memory bandwidth in excess of 4TB/s.
www.nextplatform.com
If putting a huge amount of (HBM3E) memory on top of a M4 SoC (as AMD utilise in the TSMC package) allows the performance of the Apple data centre chip to scale linearly and also lowers power utilisation (pJ/bit) - would Apple feel compelled to go that route (especially if it also helps with their carbon neutral mission).
What are the chances that Apple would offer this sort of chip to its professional users? A true HPC that doesn’t feel like a compromise when compared to last years X86 HPC with a 2 year old Nvidia keeping the room toasty.
The First AI Benchmarks Pitting AMD Against Nvidia
Rated horsepower for a compute engine is an interesting intellectual exercise, but it is where the rubber hits the road that really matters. We finally
www.nextplatform.com
If putting a huge amount of (HBM3E) memory on top of a M4 SoC (as AMD utilise in the TSMC package) allows the performance of the Apple data centre chip to scale linearly and also lowers power utilisation (pJ/bit) - would Apple feel compelled to go that route (especially if it also helps with their carbon neutral mission).
What are the chances that Apple would offer this sort of chip to its professional users? A true HPC that doesn’t feel like a compromise when compared to last years X86 HPC with a 2 year old Nvidia keeping the room toasty.
Last edited:
In fairness, he did say "threadripper level" rather than "threadripper destroying" and taking a look at the M3 Max and doubling it in various multithreaded benchmarks, especially CB R24, and yeah the theoretical M3 Ultra CPU would've been equivalent to a low end threadripper - but also true that it wouldn't have been anywhere close to the high end chips. But still, depending on which threadripper, it's very plausible.
Also while I know Gurman has said that he expects the M4 to be used in every device, he has been wrong before, especially this far out with the Studio not expected until next year - again according to him. Further, he also has predicted a "Hidra" desktop die. Lastly, that user did uncover the upcoming listed Mac IDs and two indicated they were M5 not M4 generation - could be the desktops. In other words, it's possible that next year's Studio Macs with Hidra dies will be M5s. This is rampant speculation on my part and should not be taken too seriously, but it also fits some of the available data* so it's not completely implausible.
*by saying Gurman is right about two things and wrong about one, but that he's wrong about that one does have some evidence in the form of IDs
If Apple are building their own ‘AI’ data centre - and are not farming out a massive h/w order AMD’s way (courtesy of the MI350) - (I speculate) that will have to developing some sort of SoC package solution with a memory bandwidth in excess of 4TB/s.
The First AI Benchmarks Pitting AMD Against Nvidia
Rated horsepower for a compute engine is an interesting intellectual exercise, but it is where the rubber hits the road that really matters. We finally
If putting a huge amount of (HBM3E) memory on top of a M4 SoC (as AMD utilise in the TSMC package) allows the performance of the Apple data centre chip to scale linearly and also lowers power utilisation (pJ/bit) - would Apple feel compelled to go that route (especially if it also helps with their carbon neutral mission).
What are the chances that Apple would offer this sort of chip to its professional users? A true HPC that doesn’t feel like a compromise when compared to last years X86 HPC with a 2 year old Nvidia keeping the room toasty.
I strongly suspect that any solution Apple would choose to go with will be dual use for themselves and their consumer market. I don't think Apple will want to build custom "big iron" only meant for their own data centers that they can't get economy of scale on by also selling to everyone else. Supposedly that's what they are already doing for the M2 Ultra. And is likely to continue for the M4/M5 Ultra. As for memory, while Apple could in theory go with HBM, they seem to get good mileage out of LPDDR and even Nvidia in their big iron Super chip uses both. How much memory bandwidth Apple feels the needs to put in will depend on what they think these chips will be used for both by them and their users as well as cost of course. As I wrote above, we also don't know exactly what Apple's plans for the upcoming Ultra are in terms of relationship to their other SOC like the Max (i.e. the "Hidra" rumor). So that degree of freedom makes predictions about what Apple will and won't do exceedingly difficult.
My understanding of how PCC works is that they are precluded from using anything but their own hardware. They *are* making it, and the real question is, is it useful to customers, and if so will they actually sell it? (Assuming they're not just using standard Ultras, that is.)
As Feregrini says to Wired: Unlike these consumer devices, though, PCC servers are as bare-bones as possible.
Considering about what has been said about it, I was daydreaming that we may see benefits of PCC achievements in a form of office-ready-farm-computing devices at some point.
Last edited:
My head-canon still imagines sub-basements filled with ASi blade servers crunching Apple Intelligence stuff, covering full loops of the Mothership, with an entire level dedicated to hosting the consciousness of Steve Jobs...
Thanks for the link — I had missed that article/interview. Federighi confirms what @Confused-User said:
“Building Apple silicon servers in the data center when we didn’t have any before, building a custom OS to run in the data center was huge. Creating the trust model where your device will refuse to issue a request to a server unless the signature of all the software the server is running has been published to a transparency log was certainly one of the most unique elements of the solution—and totally critical to the trust model.”
If the PCC is written in server side Swift - perhaps Apple will put some money behind building up the Xcode Server frameworks that IBM (and others) stepped away from?
I suspect that there will be renewed interest in Apple to further develop Swift to make it more performant at scale - and/or find ways to get LLM processing efficiency to move closer to C++.
If PCC is not leveraging VM (my speculation if everything is memory / secure enclave resident) - could this bring some much larger memory configurations back to the Mac lineup ?
Data Centre : (speculation) 512GB shared over 128 cores : Four SoCs (1 x NPU + 32 P-cores) is probably more performant than clusters of 256GB / 64 P-cores / 2 NPU in an Ultra configuration especially if unified memory latency is a fraction of NUMA system latency.
Last edited:
I suspect they will be using the GPU for these things. At that scale it is the most parallel solution Apple has to offer.
Personally I find it frustrating that the Studio release so late.
I’d like to buy an M4 Ultra Studio but would buy an M4 Pro Mini instead if that was released sooner (and not upgrade to M4 Ultra). I can’t be alone. I used to upgrade every 6 or 7 years, but my current model is M1 Max.
I’d like to buy an M4 Ultra Studio but would buy an M4 Pro Mini instead if that was released sooner (and not upgrade to M4 Ultra). I can’t be alone. I used to upgrade every 6 or 7 years, but my current model is M1 Max.
indeed. I much prefer a powerful desktop with IO and proper cooling. I just been forced to highend m3max for the time being in a laptop factor. Don’t use the screen and often the fans rev up bad. (Do AI “sketches” (that are deployed on linux/nvidia for proper training )in pytorch and some unrelated work in 3d). i get by for now but would by a m4ultra in a heartbeat
This is my position too.
If the M4 Pro Mini is released next month and proves to be as good as suggested by the M4 benchmarks, then I would probably not wait for the M4 Max Studio as an upgrade to my M1 Mini and M1 Max MBP.
My requirement is to have a capable desktop that I can leave connected to a large collection of network volumes and SSDs, to avoid unplugging the laptop every time I want to go move it. If the M4 Pro Mini has enough power for audio/video/photo editing, I would be happy to stop waiting.
same here !
Despite being broke, i might nevertheless opt for a mini, just with lower specs, then jump on the M4 Max, even hoping it will be an M5, or at least a different M4 take, and keep the "then maybe allready old M4" mini as my 2nd Computer.
I need desperately SC speed ! the more the better ! no way around that.
But: 40% more SC vs. my M2pro Mini, thats allready awesome !
What i might NOT do, is to upspec a M4mini to more than: 16GB/1TB. Comes then to close to a max.
But: i´d really want to have 32GB of RAM, and a 2TB SSD for sure.
While i could live with 16GB. A 1TB SSD would be hard for me to deal with.
Guess, to buy 2 times is my only way out of the dilemma.
That’s why I would prefer to have 24 GB memory options for both the M4 Mac mini and the M4 Pro Mac mini. I don’t need 32 GB so I don’t want to pay for that, but I would prefer to level up from 16 GB if the price is right.
BTW, depending on what you use your drive storage for, perhaps you could run the Mac mini with an external USB-C SSD. I have my Photos library on one.
Similar situation here;
Got a base M1 Max Studio because that was the only one available at the time, while "sufficient" but almost immediately regretted on the 32GB and 512GB, both were limiting my workflow at certain points.
M2 Max came and not deemed enough of an upgrade so I held.
M3 Max came but only in laptop form. Well I got a 16" M3 Max base (36GB 1TB) but that was for portable use, not to replace the Studio (mine is sitting inside a network cabinet, connected to crap load of I/Os, especially the 10GbE).
So I really quite need the M4 Max Studio ASAP, ideally before January as I will work remotely for 2 months, that I have to remote desktop into this Mac Studio from across the globe. The M4 Pro mini may fit the bill but we will see. According to rumors, if the base M4 chip can already do 32GB, it is hopeful that the M4 Pro can go more than the 36GB on M3 Pro. We will see.
Got a base M1 Max Studio because that was the only one available at the time, while "sufficient" but almost immediately regretted on the 32GB and 512GB, both were limiting my workflow at certain points.
M2 Max came and not deemed enough of an upgrade so I held.
M3 Max came but only in laptop form. Well I got a 16" M3 Max base (36GB 1TB) but that was for portable use, not to replace the Studio (mine is sitting inside a network cabinet, connected to crap load of I/Os, especially the 10GbE).
So I really quite need the M4 Max Studio ASAP, ideally before January as I will work remotely for 2 months, that I have to remote desktop into this Mac Studio from across the globe. The M4 Pro mini may fit the bill but we will see. According to rumors, if the base M4 chip can already do 32GB, it is hopeful that the M4 Pro can go more than the 36GB on M3 Pro. We will see.
I’d be surprised if we get a new Mac Studio before spring.
Also, I don’t recall seeing rumours saying the base M4 will do 32 GB. Maybe it will, and it would make sense to do that, but where did you see that actual rumour?
I expect the RAM options on each tier of Apple Silicon will depend on the compatibility of different sized RAM modules on each SoC package, and the artifiicial limits that Apple places on their product segments.
As we know, the base SoC has two RAM modules on the SoC (and 8 memory controllers), which to date have been:
2 x 4GB = 8GB
2 x 8GB = 16GB
2 x 12GB = 24GB
The Pro SoC on the M3 Pro has three RAM modules (and 12 memory controllers), so we have options for:
3 x 6GB = 18GB
3 x 12GB = 36GB
The Max SoC has 4 RAM modules (and 32 memory controllers), with options for:
4 x 8GB = 32GB
3 x 12GB = 36GB
4 x 12GB or 3 x 16GB = 48GB
4 x 16GB = 64GB
4 x 24GB or 3 x 32GB = 96GB
4 x 32GB = 128GB
The Ultra is just double the Max.
I understand that each memory controllers is capable of accessing 4GB of RAM, so technically, the base SoC could access 32GB (not offered) and the Pro SoC could access 48GB (not offered) - so these appear to be artificial limits to push heavy RAM users to the Max SoC.
Presumably all of the RAM module options I’ve listed exist with the possible exception of the 24GB (i.e. 8GB, 12GB, 16GB, 32GB). Whether they are different types of modules for the same capacity for different SoCs, I do not know.
I expect the M4 Pro to stick to the same arrangement, but hope it starts at 24GB (3 x 8GB) rather than 18GB (3 x 6GB), in order to differentiate it from the (rumored) new baseline of 16GB for all other Macs.
I am also banking on the "fact" that the base M4 MBP starting at 16GB, meaning the M4 Pro base staying at 18GB will be deemed too close. With the extra TB4 port on the M4 base, the M4 Pro will need more than a few differentiators to justify the price difference.
As with the M4 Pro Mac mini, there have been conflicting rumors saying the small form factor is only for the M4, while the M4 Pro stays in the current old chassis. This was when the rumors were saying the M4 Pro M4 Max needed more time and the October releases are only M4 Macs. But now even Gurman says the 14" 16" are coming so the M4 Pro mini may as well also launch. Whether the M4 Pro mini comes with a smaller chassis can be a deal breaker, if it means poorer cooling system than the eventual Mac Studio with M4 Max.
I just hope Apple could speed up their Mac update schedule, the M4 has been in the iPad Pros for months now, and we are not going to see M4 Max Studio until next June WWDC?