MacBooks - Advice sought

DxO PhotoLab macOS topics
1

A MacBook question more than a PhotoLab question, but based on use of PL8 and posted here as I have found this forum to offer sensible, non-partisan advice (is that sufficient flattery?).

Headline:

Using Deep Prime on Olympus high-iso images an M1 MAX is slower than an M3 Pro. This is not what I expected when I read, in multiple online forum, that the M1 MAX continues to out-shine Mn Pro machines.

Detail:

MacBook Pro 18,2, M1 Max, 16”, 64GB, 4TB, Cores: 8 performance, 2 efficiency, 32 graphic
MacBook Pro 15,6, M3, 14”, 18GB, 2TB, Cores: 6 performance, 6 efficiency, 18 graphic

Both machines have the latest, same OS and the latest, same version of PL8 installed. Both machines were run-up, allowed to ‘settle’ and then used with only default background apps running.

PL8: Deep Prime Acceleration: Auto selection (Apple Neural Engine)

4 x Olympus OM-D E-M10 Mk II RAW files taken at 12,800ISO. Exported to dng in to the same folder as the images (temp folder under User).
M1 = 1 min 04 sec
M3 = 31 sec

As it used to say in my student exam papers … Discuss!

[At the root of this is a decision whether to pension off the M1 while it still has some retail value and get a current model that might outlast its user!! It is not about whether the M1 Max is fast enough.]

Thanks for your time.

Clive

2

The neural engine got a large boost in performance with M3, perhaps this explains it. What’s the performance like using the GPU and not the neural engine on the M1 Max?

3

On my M2 Max, the performance of DeepPrime XD2S with Neural Engine is significantly worse than with GPU (Egypte Benchmark) 23s v. 17s!

4

Yeah, it’s not obvious unless you follow the M-series chip news fairly closely.

First, every M-series chip from the first to the latest has had exactly 16 Neural Engine cores. The exception to this being the “Ultra” versions. These have only ever been available in Mac Pro and Mac Studio.

So, given the vast majority of M-series Mac users have 16 cores, we then move to the generational performance improvements of those cores. M1-M2 was modest. M2-M3 was significant. M3-M4 was, I believe, somewhere in between. From memory, the increase between M1 and M3 was 60%.

Now, as @Stephan says, the GPU is a different thing altogether. It is certainly worth benchmarking NPU (default, I believe) versus GPU on any machine to see which wins. The only caveat to this is I believe the NPU processing is far more efficient. So if you’re in an office with mains power and don’t care about fans, the GPU may give you the edge on higher spec Macs. On my M4 Pro, the GPU cores are modest in number and don’t win. More importantly, running it on the NPU cores leaves the laptop completely silent. It would also, I believe, kill the battery less if I were not plugged in.

5

On My ancient M1 Max the GPU beats the NPU for the noise reduction by miles. I never hear the fans using the GPU, must not be working it hard enough :grinning_face_with_smiling_eyes: and I think the power consumption is still hardly anything, especially compared to using a PC with an Nvidia GPU or something like that in.

1 Like
6

Thank you to the four of you that have given me views and thoughts. What started as a comparison between two machines, both set within PL8 to ‘auto’, has developed into also comparing modes within a machine. Previously coming from the simpler Windows world, I have never got my head around the additional Neural Engine, nor the implications of one chip with one memory with multiple functions. Anyways …

M1
CPU = 1m 47s
GPU Apple M1 Max = 29s / 24s / 21s / 21s (successive runs)
Apple Neural Engine = 56s
Auto = 1m 01s

M3
CPU = 1m 53s
GPU Apple M3 Pro = 41s / 36s / 33s / 32s
Apple Neural Engine = 41s / 42s / 32s / 32s
Auto = 41s / 31s

Summary

M1 Max is the better machine when its GPU is used.
M3 Pro Neural is better than M1 Max Neural.

So, for both machines, leaving PL8 at ‘auto’ is a wrong choice and I have now set it at GPU.

I cannot explain why there should be different results on the same settings, on the same images on successive runs.

I’m not sure where this leave me other than it would probably be a mistake to replace a Max with a Pro and expect anything other than small improvements. Though that would certainly provide me with greater OS longevity of course.

Regards

Clive

2 Likes
7

I’ve observed this behaviour and presume that caching might contribute to the difference. I’ve also timed several runs and I don’t remember anything special.

1 Like
8

I bought a M1 Max to be get high performance with DxO and I was desapointed to see that DxO was more performing on neural engine, so the Max version was unusefull!

Besides, DxO chooses automatically neural engine.

9

That’s not what my simple test showed: M1 GPU 29s vs Neural 56s. Maybe we are doing different tests or maybe different versions of PL (mine is ‘8’).

Yes, hence my comment about choosing the mode that suits the user and work-flow.

I did wonder that Playpus, and that would explain the drop between run 1 & 2 then similar or identical times for 2/3/4. In my mind I have discounted successive runs and am using just the first time.

From your comments I think the Neural M1 56s and M3 41s is to be expected then. Is this a fair statement?

Then … the M1 with 32 Graphic and the M3 with 18 graphic suggests the M3 is going to be worse – and it is. Again, is this accurate? And what might I expect with a M3/4 Max with 32 Graphic – surely better than the M1?

Thanks again to you all for the thoughts and observations.

Clive

1 Like
10

Like the Neural Engine cores, the GPU cores on an M3 will be faster than those on an M2, etc. The question is by how much? But whatever that difference, I think the number of GPU cores would make the bigger difference. 18 cores on M3 versus 18 cores on M4 I would expect the M4 to win, perhaps not by much. 18 on the M4 versus 32 on the M3 I would expect the M3 to win comfortably.

I’m personally not looking for outright speed, but I sure did notice the difference from M1 to M3 (and now M4). I just leave it on the Neural Engine setting because I enjoy the silence.

If I have lots of exports I don’t sit around watching it anyway.