I recently received a new M4 Max MacBook Pro. I had been planning on waiting for the M4 Max Mac Studio but with the expected launch being in the middle of 2025 and the economic uncertainty of tariffs possibly affecting prices by then, I decided it was better to get a system now rather than wait. The reviews of the M4 systems have been great. Even an M4 Pro based system would have been a significant upgrade over the seven year old iMac Pro I had been using.

I hadn’t planned on buying a laptop. I am not really mobile that often where I need a computer, especially a high end computer and when I am, my iPad generally suffices. However, after having the MacBook Pro for a couple of weeks, my attitude there may be changing. My need for a high end iPad may be vastly reduced, especially since the Apple Silicon Macs can run most iPhone and iPad apps.

While I’m going to focus on the astrophotography image processing software PixInsight, much of what I’m going to talk about is generalizable to other usage so it may be worthwhile for those considering an M4 Max system for other workloads.

My big concerns for a system like this (one that would be used for both heavy CPU and GPU loads) was that the cooling required would be loud and still possibly insufficient to prevent thermal throttling. Fortunately, a friend bought an M4 Max based system and he ran some tests for me that showed that fan noise was probably tolerable. It was enough for me to plunk down my credit card and place an order.

I opted for the M4 Max with 16 cores (12 performance and 4 efficiency cores) and 40 GPU cores. PixInsight loves as much CPU as it can get so this was an easy decision. PixInsight also loves as much RAM as it can get so I maxed the system out with 128GB. This was an especially painful choice as the up-charge for the RAM was expensive. I chose a 2TB SSD. I’ve found 1TB a bit tight and would have preferred even more but, again, the Apple Tax on upgrading here was more than I was willing to pay.

I won’t bother with running conventional benchmarks like GeekBench or Cinebench. My results are in line with what others are reporting. Instead I wanted to focus on three things:

1. How does PixInsight perform for a moderate WBPP workflow?
2. How do some common processes perform for post-processing?
3. Do this in all power modes

While doing these tests I monitored fan activity and system temperatures. I also checked the fan noise to quantify how loud it got.

In my case, I have the MacBook Pro on a stand that raises the bottom of the laptop off the table. It also exposes a good portion of the bottom of the laptop to the air. I did this primarily to raise the laptop screen up to the level of the external display but it also might help with cooling. 

For those that are unfamiliar with PixInsight, WBPP or the Weighted Batch Pre-Processing Script is the way most people do all the pre-processing steps on their images from calibration to image integration. Depending on the resolution of the sensor and the number of subexposures, this can be relatively quick to painfully long.

I previously ran the same set of tests on the iMac Pro and an M3 MacBook Air so we have those for comparison. 

The tests were run in three different configurations on the MacBook Pro. In the first configuration, the source data was on a NAS and the destination data was on a Thunderbolt 4 SSD. In the second both source and destination were the Thunderbolt 4 SSD. In the third the source and destination were on the internal SSD.

The network connection for the MacBook Pro was through a Thunderbolt to 10 gigabit ethernet connection while the iMac had built in 10 gigabit ethernet. The NAS also has 10 gigabit ethernet.

The test here are the exact same workflows I used in a previous video comparing the iMac Pro to the M3 MacBook Air. For the WBPP workflow, the iMac Pro handily outperformed the much more modern but thermally constrained MacBook Air. This time around, however, the Apple Silicon turned the tables.

The M4 Max MacBook Air has twice the memory of the iMac Pro and it’s internal SSD is also much faster. The MacBook Pro also has Thunderbolt 5 while the iMac Pro was Thunderbolt 4. I don’t have any Thunderbolt 5 peripherals.

The WBPP process used master darks and had three nights of data, each with it’s own set of lights and flats. The data was from an ASI6200. Each night had around 21 subs. WBPP performed the calibration, cosmetic correction in auto mode, registration, local normalization, integration and 1x drizzle integration followed by an autocrop.

Here are the timings:

Computer	Source	Destination	High	Auto	Low
iMac Pro	NAS	External SSD	1:55:39	N/A	N/A
M4 Max Macbook Pro	NAS	External SSD	0:54:27	0:55:00	1:59:24
M4 Max Macbook Pro	External SSD	External SSD	0:49:10	0:51:47	1:51:22
M4 Max Macbook Pro	Internal SSD	Internal SSD	0:33:26	0:35:10	1:09:20

In the most direct comparison where both used the NAS as the source and the external SSD as the destination, the MacBook Pro was more than twice as fast. When all the data was on the internal SSD, the MacBook Pro was nearly 3.5 times faster.

In high power mode, Apple allows the fans to run up to full speed. They become noticeably loud. Sitting about 24 inches away from the MacBook Pro the fan noise was about 50dB. The noise floor of the room was around 33dB at its quietest. At 50dB the fans were louder than I’d want to endure for a long period of time but it wasn’t horrible. The tools I had to monitor things weren’t granular enough to say for sure but it appeared the system would still thermally throttle when all the CPU cores were going at 100% for a long period of time. If I’m right that means the cooling system can’t keep up when the CPU cores are under high load for several minutes. But, it’s possible I’m interpreting this wrong. The system may make decisions about changing the clock frequency that are dependent on more than just temperature.

In auto power mode the task ran about 6% longer. The system appeared to limit the CPU temperature to about 107C and also limited the fans to about 50%. At 50% the fan noise was about 40dB. Definitely still noticeable but not nearly as intrusive. In this mode thermal throttling was definitely happening to prevent the temperatures from going any higher. The impact of this was much less than I had anticipated. It was measurable but before hand I had though it would have a much greater impact. If noise is a concern then the ‘auto’ mode is a good compromise between performance and noise.

There is another possible interpretation on why the difference between auto and high power modes is so small. It may not be that auto does so well but that high power doesn’t do as much as it might because the cooling capacity of the laptop holds it back. I don’t have enough insight here to say which interpretation is correct. If I had to put money on it, I’d choose the insufficient cooling capacity as the likeliest. We should find out when the M4 Max Mac Studio is released. It should have a much beefier cooling system. Unfortunately, benchmarks like Geekbench don’t run long enough to really stress the MacBook thermally so it may not look like there is a difference but long running workloads like WBPP might run faster on the Mac Studio. 

In low power mode the system appears to primarily use the efficiency cores. The fans either never ran or ran so slowly they weren’t audible at 24 inches. The performance impact was rather large. Basically, in low power mode, the MacBook Pro performed like the seven year old iMac Pro. However, it was doing this with four efficiency cores compared to the 10 power hungry Xeon cores of the iMac Pro.

I ran all of these tests with power connected. I won’t normally be running PixInsight under battery power so that wasn’t something I was concerned about but you can be sure that the WBPP process would have a noticeable impact on battery life. It doesn’t keep the CPU maxed out for the entire test but there are definitely large chunks of time where that is true.

Here are some observations about the test:

PixInsight cares both about processor speed and I/O bandwidth. That’s not a surprise but the extent to which I/O bandwidth impacts pre-processing performance was a bit surprising. The internal SSD is much faster than the external SSD and the external SSD is much faster than the NAS. Given that Thunderbolt 5 SSD’s should have a similar performance to the internal SSD, this argues strongly for using them with PixInsight, especially when you consider how much Apple charges for internal SSD storage.

In high power mode, the fan noise does become loud but if you aren’t going to be in the room when it runs, it does run a bit faster. I was surprised at how well the auto mode did at balancing noise and performance. 

In all of the MacBook Pro tests the PixInsight swap drive was on the internal SSD. I tried various combinations of the internal and external SSD but using a single swap folder turned in the best results on the PixInsight Benchmark. On the iMac Pro I had two swap folders, one on the internal and one on the external.

My worries about the thermal behavior under load and the fan noise were both not a problem. The system performed wonderfully and the fan noise in high power is loud and I would want to sit in the room with that noise for 30 minutes or more but the fan ramps down quickly when load reduces and even at its worst it wasn’t as obnoxious as I feared. In auto mode it was noticeable but not objectionable. 

I only ran the low power tests for completeness. I can’t imaging ever running PixInsight in low power mode but from an academic perspective it is interesting to see how well low power mode compares to the old system. Even in low power mode, the MacBook Pro trounces the MacBook Air. The MacBook Air is really held back by the lack of cooling. However, it wasn’t designed for workloads like PixInsight so it isn’t really a surprise.

In an ideal world, I would prefer to run these tests at least three times each to see the variability in result, but even at this level it gives a good sense of how the system performs for WBPP.

Now, let’s turn our attention to some post-processing activities. For these tests, I looked at Gradient Correction, BlurXTerminator, StarXterminator, NosieXterminator, HDRMultiscaleTransform and LocalHistogramEqualization. While PixInsight doesn’t use the GPU itself, the XTerminator processes do. These are the same tests I ran for the iMac Pro vs. MacBook Air tests so we can compare to those. 

Here are the timings:

Process	High	Auto	Low	iMac Pro	Macbook Air
Gradient Correction	5	5	6	10	5
BlurXterminator	14	15	32	49	53
StarXterminator	16	16	34	45	56
NoiseXterminator	4	4	7	16	12
HDRMT	32	32	60	44	67
LHE	65	64	132	73	18

The difference been high and auto modes is essentially zero. None of the processes ran long enough to engage the fans except for HDRMT and LHE and even those were not running long enough to get the fans up to full speed. I rounded the times to the nearest second here because there is enough variability from run to run that excessive precision would be misleading. It Is odd that LHE was consistently slightly faster in auto mode but I suspect that if run enough times both auto and high power modes would average out to the same time.

The Xterminator tools make good use of the 40 GPU cores. It is amazing to run these tools in ASI6200 sized files and get results in such a short period of time without having to resort to a power hungry GPU like the NVIDIA 4090 on a Windows PC. Granted, the 4090 would still be faster but this kind of performance in a laptop form factor is pretty amazing.

I was bit surprised that the performance gains for HDRMT and LHE, especially LHE were not as high as they were for other CPU-only processes. I don’t know what these processes are doing internally. They are multi threaded and both processes kept all 16 cores busy. My guess would be there is thread synchronization overhead that is holding these processes back. I still don’t understand the anomalous 18 second result for LHE on the MacBook Air. I’d treat it with several grains of salt.

PixInsight is still compiled for Intel processors and it is running under Rosetta2 on the Apple Silicon Macs. While Rosetta2 is remarkably efficient, I do hope that we will see a version compiled for native Apple Silicon. I don’t think the results will improve dramatically but we should be able to expect some improvements, especially the first time a process runs as it won’t have to be translated before being run.

After a couple of weeks with the MacBook Pro, I’m impressed. It is by far the most capable computer I have ever owned. It’s also the most expensive computer I have ever owned, though the iMac Pro wasn’t much cheaper (perhaps more expensive factoring in inflation). That old iMac Pro served me well for longer than I had anticipated, but I’m looking forward to the faster processing that the new system brings.

If, like me, you prefer to run on a Mac, but unlike me, you aren’t insane enough to spend for an M4 Max MacBook Pro with 128GB then there are some good options that can get you most of the way there. Going with 64GB RAM cuts the price by $800. PixInsight on a 64GB system is still very capable. If you don’t want a laptop then the M4 Pro Mac mini with 64GB RAM is just over half the cost of the laptop I configured. For normal PixInsight processes it would get about 80% of the performance of the M4 Max but has only half the GPU cores so for GPU heavy workloads it wouldn’t do as well. For PixInsight there aren’t very many processes that use the GPU and they are all third party add ons but eventually that seems likely to change though I’ve no idea when that will be.

If, you were planning to wait for the M4 Mac Studio, there is a possibility that that could become a much more expensive proposition. Or not. My crystal ball is not very good. I decided that getting something now was the best bet for me. If the tariffs don’t happen or Apple manages to side step them somehow then I spent more for a system because it was in a laptop form factor than I would have for an equivalently spec’d desktop. My old system was old enough and becoming flaky enough that it seemed better to act now instead of waiting but if your system isn’t as old as mine was then it may make sense to wait and see what happens.

Whichever route you go, the M4 chips are pretty remarkable. The M4 Pro is probably the price to performance value in terms of CPU performance. If GPU performance factors high on your list then the M4 Max is quite capable, especially for video editing. I understand why the memory on these systems is not user accessible. It’s not just Apple being able to gouge for memory upgrades (though they do). If you need the memory then you have no choice but to configure it at build time. However, with Thunderbolt 5, an external SSD should (based on what I’ve seen around the net) be quite competitive with internal SSD performance and even with the high initial cost of Thunderbolt 5 peripherals it should still be less expensive than buying internal storage from Apple although not quite as convenient.

I hope this was helpful and hopefully for the next video we’ll have some actual astronomy instead of astrophotography adjacent material to see!

Until next time, clear skies!