Yeah, while a gap remains, they do seem to be catching up.
M4 Mac Announcements
- Thread starter casperes1996
- Start date
Nycturne
Elite Member
- Joined
- Nov 12, 2021
- Posts
- 1,392
You have to get in too close to see it IMO. The effect on text is minor enough that my glasses prescription (chromatic aberrations) is the bigger issue for me.
It's a typo in their support article. I submitted feedback so they should fix it soon.It seems there is a slight error in the list:
Identify your Mac mini model - Apple Support
Learn all the ways to identify your Mac mini model.support.apple.com
Found M4 and M5 Mac identifiers
[NOTE: I’ve cleaned up my mess here, but for the sake of posterity, I’ve retained the “Mac16,15” information even though it is an Apple support error and should be Mac16,11. Use the link below to see if it has been corrected.] Okay, the "Identify your ..." pages are up for the new models, and...forums.macrumors.com
The Mac Mini is 16,15 and 16,10. 16,15 did not originally appear in the list. If the original list is incomplete, then indeed Gurman is likely right that the M4 generation is going into every model.
Regarding the list being incomplete, that's certainly still a possibility. That list came from a leak in the first 15.1 beta, way back in July.
I think I'm going to forego my base Mac mini M4 Pro teardown and return it. I was mostly interested in analyzing the PCBs under the assumption that the mainboard was split into two (which I now know it isn't), and there doesn't seem to be much interest in other parts (aside from the SSD and perhaps TB5-related stuff). The minis have garnered enough attention that I'd be surprised if there wasn't a decent teardown of an M4 Pro model within 24hrs. If there isn't, I'll reconsider.
Anyway, I've got two higher specced units that should arrive in a couple weeks, which have a useful amount of RAM for me (unlike the base model).
Edit: Just got a notification that they'll be a week early, just 5 days from now!
Anyway, I've got two higher specced units that should arrive in a couple weeks, which have a useful amount of RAM for me (unlike the base model).
Edit: Just got a notification that they'll be a week early, just 5 days from now!
Last edited:
Nycturne
Elite Member
- Joined
- Nov 12, 2021
- Posts
- 1,392
Ran the two laptops through my Messier 33 data from about a month ago. The setup does calibration and integration of images captured in a session, using some pre-processed calibration data to help speed things up, similar to what I'd be doing normally. The working data set is 58GB, most of that produced by the process itself, so there's quite a bit of I/O involved. The M1 Max does it in about 22:47 (~23 minutes) and the M4 Pro does it in about 12:50 (~13 minutes).
Interestingly, this was almost exactly the improvement I expected based on the early Geekbench 6 results. Not bad. This will let me process larger multi-night projects much faster.
M1 Max:
M4 Pro:
The gains from Xcode aren't as impressive, but my project is already fast enough on the M1 Max. It went from 10-15sec down to 6-8sec for a full clean build. This is for a13kloc 37kloc scale project.
Geekbench 6 CPU + Metal for grins:
Interestingly, this was almost exactly the improvement I expected based on the early Geekbench 6 results. Not bad. This will let me process larger multi-night projects much faster.
M1 Max:
M4 Pro:
The gains from Xcode aren't as impressive, but my project is already fast enough on the M1 Max. It went from 10-15sec down to 6-8sec for a full clean build. This is for a
Geekbench 6 CPU + Metal for grins:
Last edited:
Thanks for posting that. Can you ballpark the relative importance of CPU vs GPU performance for your M33 processing? Is that process MC or SC on the CPU?
And how important is RAM size? If you had enough to hold the entire 58 GB dataset, would that make a big difference (like if you were inverting a matrix of that size), or is it not a process that needs to work on the entire dataset at once? I suppose you could test this by adjusting the dataset to be small enough to fit entirely in RAM, and seeing if there is a big jump in speed.
Last edited:
Nycturne
Elite Member
- Joined
- Nov 12, 2021
- Posts
- 1,392
PixInsight itself doesn't use the GPU at all in this calibration script, but it is heavily multithreaded using one thread per core. The dataset held in RAM is a fixed subset, a few GB. Just needs enough buffers to keep the threads fed as they process the files in parallel. My surprise is that despite the process constantly reading and writing files, that it scaled almost perfectly with the CPU improvements. Tells me it's not as I/O bound as I'd think.
This process is to reduce the noise as much as possible on captured data. So there's a few steps: calibration of the frames to account for the characteristics of the specific sensor and optics (yes, this is important), registration to align all the frames to each other, and then integration to sum all the data together to reduce shot/photon noise. But every step produces an intermediate set of files to avoid keeping an unbounded data set in memory (I had 65 frames in this data set, but if people shoot short exposures, 1000 frames is not unheard of). Integration is the only point where all the different files are fed into a single buffer, but even that should be doable multithreaded.
EDIT: The GPU perf is in the same ballpark as the M1 Max, so I'm not expecting improvements in GPU-bound tasks. But I don't push the GPU much with my CAD work. I'm just glad that's a sidegrade rather than a downgrade.
Last edited:
I went nano without a thought. I work next to a window all day at the office and the reflections suck.
Additionally as it is an etching and not a coating, I doubt we will see more stain gate with it. I feel more confident of its durability.
We shall see. It arrives this week.
That’s actually not bad? Try build a desktop based around a 4070 with 16 core cpu for less than 2 grand?
Have there been issues with the AR coating in the past? Just did a quick search and didn't come up with much. Not to go too astray of the topic, but my M1 MacBook Pro is in need of a cleaning, and I'm planning to be as careful as possible (invert laptop, water mist, and microfiber cloth). Any tips?
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 365
challenge acccepted!
https://pcpartpicker.com/list/C4Drxg. All top specs, added 10G NIC and got the 4070 Super for less than $2K, plus if AMD supports Zen6 on AM5 then you can just upgrade the CPU without buying whole a new PC.
Assembly cost?
Shipping?
It’s also not mini sized.
Sure you’re close…. But it’s close in price, and you haven’t gotten anywhere near close when other metrics are taken into account like your time.
Not recently but my 2015 anti reflective coating coming off has left me concerned.
You need a case and power supply
Oh and any OEM is going to be more expensive than building your own. Having said that the 9950X is better than the Max (except single core). Also Apple’s won’t beat a desktop 4070 Super in the majority of consumer workloads, but it will obviously crush it though whenever VRAM is at a premium. So as a workstation, you could argue the hypothetical Apple Studio is still competing.
This is the problem with PC builds:
So, you've got your RTX 4070 Super (PCIe 4.0 x16) in the PCIEX16 slot, and your X520-DA1 NIC (PCIe 2.0 x8) in the PCIEX4 slot (that looks like x16, but only supports x4). So I guess the NIC will only run at half speed? Or will it work at all?
And then there's SSD expansion-- M2B_CPU, M2C_CPU, and M2D_SB are all disabled because of PCIe cards, leaving only M2A_CPU (fast) and a PCIe 3.0 x2 (hella slow).
Basically, for $2k+ (need a case and PS) it might work, but maybe not at full capability, or maybe not at all. Click those "buy" links and let me know!
Last edited:
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 365
yeah lol
Its about the same ballpark and this PC does come with 1TB SSD which is a $200 extra with Apple over the base Studio. The Studio will be smaller and will use less power. The PC is upgradeable and offers internal expansion of SSDs etc.
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 365
The buy links are ghosted for some. Yeah, for most folks the Studio is easier to configure
It'll work, just slow. PCIe has auto width and speed negotiation so a x16 gen4 card is quite happy to negotiate all the way down to x1 gen1 if that's what it takes to make the link work OK. (Ask me how I know!)
That’s what it’s supposed to do, but my experience has been such assumptions bring out quirks in the motherboard, bios, and drivers, not to mention SFP+ NICs and transceivers have quirks of their own. Sometimes it’s solvable in software/firmware (with time spent), and sometimes you have to spend more $$$ to get things working.
Interesting. I would've expected the link speed negotiation to all be in hardware state machines (the PCIe LTSSM) at each end of the link. That said,the world is filled with weird implementations of things...
(The issues I ran into were a FPGA connected into a host PC through a perhaps excessively long PCIe extender cable. Sometimes the signal integrity with this cable wasn't good enough to run full width / speed and it would renegotiate down to x1, then my FPGA design would break because it really expected to be able to push data as fast as the full link would allow.)
