In case anyone is curious, I’ve been comparing the performance of Photolab 9.6 Build 43 on my M1 MAX Macbook Pro (64 GB RAM) vs my M5 MAX Macbook Pro (64 GB RAM).
First I locked in Photolab’s AI Acceleration preference setting on both machines to GPU. (Using “Auto Selection” or the “Apple Neural Engine” setting is significantly slower than GPU when exporting images. There are reasons why you might want to use the slower “Apple Neural Engine” such as saving battery but I’m configuring for speed.) I also set the “Maximum Number of simultaneously processed images” to 3.
I used the same 350 high ISO RAW images for all the tests.
M1 MAX: Export 350 High ISO images 1 hour, 27 minutes, and 54 seconds 15.07 seconds per image
M5 MAX: Export 350 High ISO images 23 minutes, and 40 seconds 4.06 seconds per image
The rate of 4.06 seconds per image is approximately 3.71 times (≈271.18%) faster than the rate of 15.07 seconds per image.
Setting AI Acceleration to Apple Neural Engine instead of GPU resulted in export performance of 7.86 seconds per image on the M5 MAX.
All images were exported using Deep Prime XD3.
Apple M5 Max
Total Number of Cores: 18 (6 Super and 12 Performance)
GPU - Total Number of Cores: 40
Apple M1 Max
Total Number of Cores: 10 (8 Performance and 2 Efficiency)
GPU - Total Number of Cores: 32
There are no big surprises here. But if you’re thinking of upgrading an older M1 Mac and if you rely on Photolab I think you’ll be happy.
Important: If you are going to perform your own tests, don’t forget to quit and re-open Photolab after changing the AI Acceleration preference. Otherwise the change won’t take effect.
Merci pour votre retour
De mon coté, sur Mac Studio M4 max, c’est deux fois plus rapide en GPU qu’en Neural Engine.
Par contre, sur un MacBook Air M2, c’est le neural Engine qui est plus rapide que le GPU de cet ordinateur.
le choix du réglage est donc à faire selon la machine
Are you saying its better to use GPU rather than neural engine if on mains supply? Interested in this because I have an M4 macbook Pro (24GB RAM) and find PL9 laggy when having previews checked in settings - I also have Pure Raw 6 to do batch processing of images so I can get around the denoise preview problem for now by preprocessing the denoise in PR6 but would love to solve this problem - would appreciate an opinion.
Hi Steve: I’m not saying it’s “better” but it may be a lot faster. The only way to know for sure is to test with some high ISO images and Apple Neural Engine and then again using GPU only. For me it’s a lot faster. But as ‘belnea’ states above it really depends on what hardware you are using.
Important: If you are going to perform your own tests, don’t forget to quit and re-open Photolab after changing the AI Acceleration preference. Otherwise the change won’t take effect.
Thank you belnea. Your point is well taken. It all depends on what hardware you’re using and the best setting is whatever you decide after doing your own tests.
Hi @DaveBurbank , thanks for sharing your test results across the two generations of Apple computer processors.
Regarding your experience with the GPU offering the fastest export time. The Max processors have a large number of GPU cores, typically double that of the Pro and more than double of the base ‘M’ versions of equivalent chip. In my tests with a regular base ‘M’ the Neural Processor was undoubtedly the fastest option.
I suspect when going from the Pro to Max chips the increase in GPU core count makes the GPU the fastest option. But this may vary from generation to generation too as Apple may have prioritised NPU or GPU performance improvements… As you say, test your own system to find out what’s best
Out of curiosity, the files from which camera did you use for your test?
Thank you for that thoughtful reply. I appreciate your explanation of how different flavors of the Apple M series chips will yield different results in terms of whether GPU or Apple Neural Engine will be the best choice.
My high ISO test folder contains 350 Sony RAW images from the Sony A1, A7r5, a9 Mark II, A6700, and A7CR.
The ISO breakdown was as follows:
ISO 6400 109 images
ISO 8000 162 images
ISO 10000 22 images
ISO 12800 57 images
Thanks for the reply. I think @CHPhoto explains well how your results will be different than mine.
30west
(Gary (Fujifilm xt-5, Mac Studio M2 max))
10
I guess this was helpful in terms of raising the issue of these settings. I tested 300 photos with GPU; Apple Neural, and “auto”. Got exact same time for all 3 settings. MacBook Pro M4 48 gb ram (max for this model). Since there are apparently different optimizations for different configurations, it begs the question of whether DxO’s “auto” option is able to determine the optimal solution based on the system it detects.
Great test, Dave! And thanks for sharing. I’m mostly a Windows guy, but my wife has an M1 Macbook Pro, and I’ve been wondering if M5 is better enough than M1 to be worth springing for, in photo processing. I think M1 vs M2, or M4 vs M5, it’s hard to justify spending a lot of money. But at some point: yes it is!
Any chance you could run the test on the M1 Max with Apple Neural Engine? It’s the only test missing in your quadrant (M1 GPU, M5 GPU, M1 Neural Engine, M5 Neural Engine). Neural Engine is usually faster on lower configurations of the M processors.
On second thought, the M1 Neural Engine should be about three to four times slower than the M4 Neural Engine.
My own tests in the day when I had an M1 Max had GPU and Neural Engine about equal.
Apple Silicon Generation
Neural Engine Performance
Speedup vs. Previous Generation
M1
11 TOPS
Base
M2
15.8 TOPS
+43%
M3
18 TOPS
+14%
M4
38 TOPS
+111%
M5
60 TOPS (133 TOPS combined)
+58%
The quick solution to less powerful Apple hardware is to use DeepPrime 3 (not XD). It’s about three times faster and often looks better (less invented detail). DeepPrime XD in PhotoLab 9 subjectively seems to generate far fewer artifacts than in PhotoLab 8.
I will be happy to do this. Probably later today (US Eastern Time). I plan on keeping the M1 MAX so I can assist in the future if there’s a need.
p.s. I am really enjoying Photolab 9. (Other than a few workflow things that drive me crazy.) I have extensive history with Capture One Pro and Lightroom/Photoshop/Camera RAW) but I’m trying hard to make Photolab my go to solution to day-to-day image processing.
Hi Alec:
I ran the above test again today on the M1 MAX with “Apple Neural Engine” selected. I also ran it again with “GPU M1 MAX” selected. Here are the results:
PhotoLab 9.7.0 Build 44
M1 MAX 64GB RAM
GPU M1 MAX
350 images 1h 26m 55s
14.9 seconds per image
Apple Neural Engine
350 images 1h 55m 4s
19.7 seconds per image
That looks about right. I normally used the Apple Neural Engine on M1 Max as that meant that the Neural Engine which otherwise doesn’t get much work is busy and the GPU is available for either further PhotoLab work or for other work, even while exports are going on.
