Yep
Comparison with the 6-core A19 Pro (17 Pro Max?): https://browser.geekbench.com/ai/v1/compare/335061?baseline=335713
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
Comparison with the 6-core A19 Pro (17 Pro Max?): https://browser.geekbench.com/ai/v1/compare/335061?baseline=335713
Check this out. 19 Pro vs 18 Pro Metal score.
46000 vs 32000. Almost 50% increase and on par with M2 ten core gpu in mac mini.
If regular M4 was based on 18 Pro, means an increase from 32000 score to 57000 (M4 base mac mini), an increase with 80%.
If we see same ball park of increase for M5 based on A19 Pro thats 82000. Roughly same as M2 Pro with 19 gpu cores.
Let’s predict M5 Max, although with them probably using new tech this could be off by alot. But M4 Max 40 cores, 187000 score. That’s 330% above base M4. If M5 Max keeps same and base has 82000, that’s 270.000. Or yeah, almost 50% faster vs M4 Max.
iPhone18,2 - Geekbench
Benchmark results for an iPhone18,2 with an ARM processor.
browser.geekbench.com
46000 vs 32000. Almost 50% increase and on par with M2 ten core gpu in mac mini.
If regular M4 was based on 18 Pro, means an increase from 32000 score to 57000 (M4 base mac mini), an increase with 80%.
If we see same ball park of increase for M5 based on A19 Pro thats 82000. Roughly same as M2 Pro with 19 gpu cores.
Let’s predict M5 Max, although with them probably using new tech this could be off by alot. But M4 Max 40 cores, 187000 score. That’s 330% above base M4. If M5 Max keeps same and base has 82000, that’s 270.000. Or yeah, almost 50% faster vs M4 Max.
Last edited:
Depends , but i can see the nex M Max around 225k mark at least, so around these two level
| AMD Radeon HD W6800X | 225732 | |
| AMD Radeon Navi21 | 224593 |
| Apple M3 Ultra | 228598 |
Thanks! I think that line of extrapolation is the best we can do, without knowing what architectural changes Apple has implemented with SoIC, assuming the rumors are correct and M5 Pro and above will adopt it.
The imagination tends to run wild, so I’ll refrain from specifics because I don’t have any idea what or where the limits are. But I doubt the basic relationships built into Apple’s product line will change. So the M5 Pro will still need to fit into the Mac mini, the M5 Max will still need to fit into the MacBook Pro, and the M5 Ultra will still need to fit into the Mac Studio.
That said, I wonder if the Max could be “the new Ultra,” so to speak, in the sense that it will increase (double?) both CPU and GPU, just without the UltraFusion packaging. The Ultra, then, would no longer be simply double the Max, but would instead be “the new Extreme,” so to speak — so the failed 4C is reborn in the M5 Ultra. TSMC always goes out of its way to emphasize that SoIC retains the capability for UltraFusion packaging.
For what it’s worth, I strongly dislike the “Extreme” name — I think it is appropriate for something like TSMC’s high-power N4X and N3X (where the “X” stands for “Extreme”) nodes, but I think the word sends the wrong message in Apple’s quest for both efficiency and performance. “Ultra” (the word means “Beyond”) embodies the mission perfectly and should remain as the top tier.
Amazing. It seems like they focused on a new GPU architecture this time, right? Otherwise, just the shift from N3E to N3P wouldn’t justify such increases in GPU performance.
What’s strange, is that aside from mentioning the ML/AI accelerators embedded on the GPU, they didn’t mention a new GPU core architecture…
Let’s cross our fingers and expect huge computing gains for the M5. By the way, what if the M5 is clocked higher than the A19 Pro? Would that mean even bigger gains?
What’s strange, is that aside from mentioning the ML/AI accelerators embedded on the GPU, they didn’t mention a new GPU core architecture…
Let’s cross our fingers and expect huge computing gains for the M5. By the way, what if the M5 is clocked higher than the A19 Pro? Would that mean even bigger gains?
My number play have already taken that into account. Same freq increase from M4 to M5 as the increase in freq from 18 Pro to 19 Pro. But yeah, exciting times.
I think the new tech will allow Apple to really mold SoCs exactly how they want them without taking into account what must be a tier based on a binned / cut version of something else. That’s correct, right?
Which bring exciting possibilities. I wouldn’t be surprised if they dropped the Pro and Max monikers and instead had:
* Base, basically like today but with great GPU boost. Could possibly end up in an Apple TV too with console abilities? With support for more demanding games of course. Such an easy way to enter the console market for Apple. Basically macOS games on the TV. Probably with either thirdparty controllers and/or they release a controller themselves.
* CPU-focused chip. For CPU demanding productivity.
* GPU-focused chip. For 3D-related tasks or macOS gaming.
* And ”Extreme” or whatever they call it. A large one with max of both CPU and GPU.
Kind of like they already have today but with clearer lines between chips since the Pro does not have to be a cut Max, and the Ultra does not have to be two fused Max’s.
NO!!!
GB6 ML benchmark can be run on NPU, GPU, or CPU.
1. It gets reported (visually) in much the same way in all three cases. TERRIBLE UI. You ALWAYS have to look very carefully to see which of the three was targeted.
2. Even if you supposedly targeted say ANE, some code may still run on other hardware. The most common case is that the ANE FP32 results are meaningless because ANE doesn't support FP32 and those are run on the GPU. Quantized results for the GPU are also unhelpful, in that they appear to be unpacked to FP16 before being run on the GPU -- this may change after tool and benchmark updates for the A19, which appears (kinda sorta maybe? through the Apple marketing vagueness) to now have some limited support for quantized values.
Basically you have to be very careful in making sure the values come from where you think they come from, and mean what you think they mean.
For ANE the FP16 and quantized values are generally legit.
For GPU the FP32 and FP16 values are generally legit.
For CPU all three values are generally legit.
I think I’ve already said it, and that I’m not the only one, but man, I’m really hyped with the upcoming M5 and I’m wishing they publish any interview with Shrouji and other Apple executives about the new A19 Pro, the new cores, the new architecture…
By the way, I’ll leave you with a bit of food for thought: do you think the only difference between the A19 and the A19 Pro is the amount of unified RAM? Do you think the M5 will be based on the A19 or the A19 Pro? If it’s on the latter… do you think we could see, yet again, another increase in the minimum amount of RAM on the M5 or, at least, bigger increases for each step from the base 16GB?
By the way, I’ll leave you with a bit of food for thought: do you think the only difference between the A19 and the A19 Pro is the amount of unified RAM? Do you think the M5 will be based on the A19 or the A19 Pro? If it’s on the latter… do you think we could see, yet again, another increase in the minimum amount of RAM on the M5 or, at least, bigger increases for each step from the base 16GB?
