Performance notes about version 9(May apply to earlier versions)

DxO PhotoLab Windows topics
1

For the entire year I’ve been using Photolab, I left the settings in the performance tab largely untouched. Like many others, I’ve been unsatisfied with the snappiness of Photolab on my system, especially when it comes to DeepPRIME processing.

To frame what I’m about to say, here are my system specs:
AMD Ryzen 7 5800H (8 cores)
32GB DDR4 3200MHZ SODIMM memory
No dedicated GPU, SSD boot drive and scratch drive.
Sony a6100(Which puts out 24.2MP lossy compressed RAW photos)

Recently, after reading part of the Photolab 9 user manual, I decided to change the AI acceleration from Auto(CPU), to "AMD Radeon™ Graphics), as well as setting OpenCL to On. Attached is a screenshot of my exact performance configuration.

2025-11-23 01_02_59-Preferences
2025-11-23 01_02_59-Preferences814×731 60.3 KB

Previously, with DeepPRIME 3 processing, images took 1 minute and 30 seconds to process, on average, to export. Now, they take 21 seconds, on average.

Previously, with DeepPRIME XD2S processing, images took 2 minutes and 50 seconds, on average, to export. Now, they take 50 seconds, on average.

I’ve also noticed the image browser is much more snappy and loads thumbnails way faster, particularly with images that contain advanced local masking such as AI masks, which went from taking minutes to process(yes my computer is a relative potato), to 30 seconds or so.

In the Customize tab, the good news continues. Rather than adjusting a setting and waiting for the result to appear, the preview is rendered at what could be considered to be a non-slideshow FPS, across multiple different types of customization.

The program itself is also faster to load all modules after launching, and faster to display thumbnails in new folders.

I hope this helps someone out there, because man it’s been a struggle using this software that unironically costs more than my computer. I always figured I simply have a computer not capable of running the software, when in reality while the software perhaps isn’t the snappiest(I don’t have a good frame of reference, not having experienced the program on any computer with a dedicated GPU, let alone one meeting the recommended configuration.), it definitely runs like how you’d expect such high end software to run. Again, based on my limited experience with high end software and computer hardware.(I spent it all on lenses)

2 Likes
2

Ran a short test (on Mac and PL 9.1.1) exporting 16 images, all with the same settings.

As for parallel exports( 1, 2, 3…5) with standard denoising,
average times from 2 runs were 58, 41, 39, 38, 38 seconds.

With DeepPRIME XD/XD2s, export took 250 seconds (5 parallel)
Due to the duration and my not usually using XD/Xd2s, I refrain from testing other settings.

Even though DxO recommends to set parallel export to 2, it might be worth trying with higher values. I usually found the sweet spot to range from 3 to 5, so I set 4 as default.

3

I set my export to 10, I’ve also noticed a sharp improvement in export when Deep Prime is turned off.

4

As far as I can tell, the effectiveness of parallelization greatly depends on the specs of your system. On higher end GPUs, it’ll be more effective, on lower end ones, less so. For my system I find 2 to be the sweet spot with AI denoise, but like @LVS pointed out, parallelization is far more effective without denoising since it’s a task most GPUs don’t need to use their full power/bandwidth to perform.

5

Using i7-14700KF and RTX 4070 my GPU is 90% busy during 2 thread exports with XD2s, so no point increasing that. For DP3, 3 thread exports are about 20% faster than for 2 threads, so not a game changer, while asking for stability problems.

Increasing the number of export threads beyond the default may cause instability of PL to DopCor connection (seen in PL logs, although exports finish ok). Actually, because of current DxO/NVIDIA/MS problems, using 1 export thread may make your exports more stable if AI masks are used, as pointed out in this forum (smaller VRAM footprint).

Note also that this topic is about Windows, for Macs it might be different.

3 Likes
6

On my previous PC, there was an export speed difference between picking the generic graphics (Nvidia - don’t remember the entire line) and selecting the actual GPU by name (Nvidia GeForce GTX 1050 Ti). It wasn’t a big difference, but if you’re optimizing…

Caveat: this was with PL v5 or 6, didn’t bother to see if changing it back for later versions made a difference - and I don’t have v9.