Apple silicon performance expectations

I had been running a hackintosh using an i9-9900k CPU and a Radeon 580x GPU. With this I was getting about 25 second per image with DeepPrimeXD. Today I switched to an M2 Pro Mac Mini with 12 cpu cores and 19 GPU. I expected a modest improvement based on benchmarks I had seen but actually it’s a bit slower at about 30 seconds per image. I tried various settings (auto, neural engine and gpu) and they all produced about the same result. Is this expected or is something not right?

Hard to say without further detail…

  • Details of files used (brand name, sensor Megapixels, ISO settings, file size…)
  • Edits applied to each image (screenshot?)
  • DPR used stand-alone or as a plugin (what app?)

The main difference with Apple Silicon is power. Yes, they’re quite fast, but you could run that workload on a laptop and not ruin battery life. I’m not even sure you can run that Intel spec in a laptop. If you could, you’d have a lot of noise and heat to deal with.

The proposition of Apple Silicon is most people can do just fine with the MacBook Air… the lowest model they make. I run PhotoLab (and Lightroom and Safari and all sorts of utilities) on an M1 MacBook Pro — the original model — and I have no complaints. Well… except I wish the Neural Engine issue with DeepPRIME/XD would be sorted, which would make exports a little quicker.

Files are irrelevant since the same files were used in both cases but for reference the camera is an RX10IV. The RAW files are about 20MB and the ISO ranges from 100-6400. ISO does not appear to impact processing time on either computer.

I assume by edits you mean settings. Lens correction applied and normal sharpening. But these setting don’t appear to affect processing time either.

DPR was used as stand-alone.

But I think I found the issue. I was relying on the on screen countdown. But when I applied a stop watch I got the expected benefit. Actual run time on the M2 Pro is about 19 seconds. That’s about 30% faster which is in line with benchmarks.

Laptops are not relevant here. Both computers are desktops

I’m saying laptops are relevant to Apple Silicon, which is not designed to monster enormous desktop Intel setups.

If you had a laptop, you would benefit in different ways.

I don’t know what you are talking about. For a given processor desktops will always have a performance advantage over laptops if for nothing else heat management. Apple Silicon SOCs are not more relevant to laptops. And regardless it has nothing to do with my question.

True for Intel. Not really true for Apple Silicon. But… that doesn’t address your question either.

What does, perhaps, address your question… benchmarks are artificial. The performance you get with any application will depend on that application. Like for like, you know the answer on performance because you measured it.

Okay, case closed then.

My problem with PR3 performance on Apple Silicon is that it throttles if not plugged into the wall. I just started my first batch on PR3 on site while on battery, and it only did about 60 pictures in about two and a half hours. I plugged the charger back in, and it did 20 pictures in about 15-20 minutes. After a while of being plugged in, the rate of it slowed down again but not as slow as on battery power. What in heck is going on? There is not much change in my pictures. Most of them should be equally challenging to process. The way this should be is a few minutes of quick processing with a slowdown and quick bottoming down in performance. Battery or AC power should not even matter. It should max out the CPU/GPU on either. The settings menu is greyed out while processing photos, so I can’t go into settings to figure out if there is a setting to not throttle down on battery power.

M1 Pro Macbook Pro with 16GB
Files are CR2s from a Canon 80D and weigh around 28MB each
About 85% of them are event photos in controlled lighting. They’re fairly consistent in settings and content.

What kind of noise reduction did you choose?

  • Never ever use PRIME, because it’s designed for exclusive use of the CPU and will provide the longest possible processing times. Use HQ, DeepPIME or DeepPRIME XD instead.

Is your DPR set to use the CPU only?

  • Switch the setting to either GPU or ANE, observing that the later provides best processing times but can occasionally produce colour casts like described in this post regarding PhotoLab…depending on the version of macOS used. Not sure if that effect also affects DPR though.

Which version of macOS is installed on your Mac?

All true, but after many uses “we understood” from ISO 800 to 100000 makes sense to use only DEEP PRIME XD. The recent patch that disabled PRIME and the older algorithm is right. But performance is important, using the Neural Engine of M1 / M2 would halve (and beyond) the process time for each image, leading to a real gain. It would be important for the Color Shift present and reported as Bug currently to be corrected quickly. Now Lightroom and its noise reduction take an average of 15 seconds per photo, with PR3 it takes 30/40.