M4 Mac Announcements

M4 Pro is 4.49GHz and also 4.51GHz. interesting.
Geekbench seems to attempt measuring CPU frequency rather than trusting the reporting of system tools like powermetrics. This puts them at the mercy of the precision of the timers they use, and whether the CPU actually stays in its top frequency for the full duration of the measurement. This results in significant run-to-run variance.
 
I haven’t heard anyone else talk about the SSDs on the new Macs, so maybe I’m the only one hoping for something around 10 to 15 GB/s. I’ve ordered two Mac minis to experiment with clustering.

TB5 can theoretically transfer* at 120 Gb/s = 15 GB/s, and three ports plus the local drive would mean something shy of 60 GB/s. The best accessory for a Mac mini would simply be another Mac mini (use-case dependent).

However, I’m tempering my expectations, given that one would think they’d have talked about it (although if it depends on the SSD size, they might have decided to not do so). Anyway, my fingers are crossed.

* Interestingly enough in the Intel publications I’ve seen as of a couple weeks ago, only the symmetric 80-80 Gb/s mode was available for networking, while the asymmetric 120-40 Gb/s mode was for display support. So it seems that either the Intel material was incorrect or Apple has added that functionality to TB5— in the Mac mini video, they explicitly show “120 Gb/s transfers”
 
I haven’t heard anyone else talk about the SSDs on the new Macs, so maybe I’m the only one hoping for something around 10 to 15 GB/s. I’ve ordered two Mac minis to experiment with clustering.
10-15GB/s SSD? Do such things exist? Wow. I very much doubt the Mac ssd is that fast if so.
 
10-15GB/s SSD? Do such things exist? Wow. I very much doubt the Mac ssd is that fast if so.
Yep! There are a few manufacturers of PCIe v5 SSDs that have 14+ GB/s read and 12+ GB/s write, and they consume ~3W more than the typical previous PCIe v4 SSDs:




Also, given that TB5 uses PCIe v5, it’d make sense to use it for the SSD as well.

Edit: That last bit is wrong— TB5 supports PCIe 4.0. Oops! Still going to hope for faster SSDs, though.
 
Last edited:
If those are anywhere close to accurate, M4 is just murdering the top end of Raptor Lake (i9 or Ultra 9 or WE). I wonder what the P/W comparison will look like.
M4 Pro beating M2 Ultra, including on multicore. Imagine what M6 will do.
 
Which one? 16,7 or 16,11?

Pretty sure 16,11 is the mini, btw. I did a bit of a deep dive into the identifiers after making a foolish prediction at the other place (I’ll follow up). Either way the M4 Pro CPU results are awesome.
I was thinking that 16,7 was the mini, but you’re probably right.
 
M4 Max
Great cpu score, ok gpu score (I thought it would be higher).
1730464348289.png

1730464374128.png
 
M4 Max
Great cpu score, ok gpu score (I thought it would be higher).
View attachment 32419
View attachment 32420
In a Mac Studio this will replace my iMac with 10700K and Radeon Pro 5700XT.
If I could get the cpu of the M4 Pro with 48 gpu cores instead I would.
But I really do look forward to it. Really hoping for an upgraded studio display by the time it happens too. Otherwise I’m not quite sure what I’ll do for display replacing the iMac
 
In a Mac Studio this will replace my iMac with 10700K and Radeon Pro 5700XT.
If I could get the cpu of the M4 Pro with 48 gpu cores instead I would.
But I really do look forward to it. Really hoping for an upgraded studio display by the time it happens too. Otherwise I’m not quite sure what I’ll do for display replacing the iMac
Oh God what I would give for a tandem oled studio display.
 
[
If I could get the cpu of the M4 Pro with 48 gpu cores instead I would.

I believe many people would like tailored performance devices. If Apple could just go with the interconnect scheme and build chips with customized CPU/GPU SoC combinations, they could offer users the machine they really want, sidestep the reticle limit and possibly optimize yields/bins. It seems like the sensible approach: burn interconnected SoCs on the wafer and do cut-and-reconnect operations for outliers.
 
[

I believe many people would like tailored performance devices. If Apple could just go with the interconnect scheme and build chips with customized CPU/GPU SoC combinations, they could offer users the machine they really want, sidestep the reticle limit and possibly optimize yields/bins. It seems like the sensible approach: burn interconnected SoCs on the wafer and do cut-and-reconnect operations for outliers.
I think they’ll get there. I also think the reason they haven’t gotten there yet is not that the interconnect technology won’t let them, but that there are just a shit-ton of wires that would have to be routed between die, and they all switch simultaneously because they are correlated bus signals, and that generates tremendous cross-coupling. There are ways around it (staggered wire swizzling, shielding with power rails, etc., but Apple tends to wait until they can really nail the solution.). There’s also the potential for an inductance problem. But I think we’ll see something along these lines around M6.
 
In a Mac Studio this will replace my iMac with 10700K and Radeon Pro 5700XT.
If I could get the cpu of the M4 Pro with 48 gpu cores instead I would.
But I really do look forward to it. Really hoping for an upgraded studio display by the time it happens too. Otherwise I’m not quite sure what I’ll do for display replacing the iMac
This looks promising: Asus ProArt Display PA27JCV.
 
Back
Top