Lightroom Classic export benchmark comparisons, post yours!

BananaDunka · 24 Apr 2023 at 13:21

I forgot to do this when I had the RTX 3080 Ti but from a curiosity point of view and there seems to be little in the way of data online about this, considering this is an important part of photography for all of us I found this quite puzzling. The speed to export using the new 4090 has seemingly significantly improved for large batches of RAW files to 100% JPEG full resolution, which is the typical export I do from weddings/events etc.

I'd already exported a set of 1385 RAWs to JPG from the 30MP 5D mk 4 which took about 3 mins 42 seconds, I could swear the 3080 Ti was taking 12-20 minutes previously but that was with the 25MP 5D Mark 3 RAWs so not comparable I guess. Then today I exported 10x 50MP test DNGs taken with my phone before getting the idea for the thread. I figured I'd pose the question here to see if others can post their CPU, RAM and GPU and then time their export for the same images. I'm quite intrigued to find out if a faster CPU will bolster a faster export time when the GPU is already super fast, or if it's all down to the GPU during exporting assuming the use GPU for Exports option is enabled in Lightroom settings.

Link to the RAWs: https://drive.google.com/file/d/1nv54CPwcUx-PsN_939tNiWPyJjCsgY8J/view?usp=share_link

Steps:
1: Import the 10 DNGs into Lightroom, make sure to generate only standard previews, no other processing.
2: Select all 10 > Export to JPEG at 100% quality, no resizing, no additional processing:

3: Start the timer as soon as the export button is clicked. Stop the timer as soon as this progress bar goes away:

Edit***
Lastly, if you have time, then copy and paste the 10x DNGs into the same folder (Windows will name them accordingly so no issues), then import them to do an export of 100 to test out a memory saturation curiosity.

Post your result!

Mine:
CPU: 12700KF @ stock
RAM: 64GB DDR4 3600 CL18
GPU: RTX 4090 @ stock
Export time for 10x: 9 seconds
Export time for 100x: 42 seconds

Memory/CPU saturation on the 100x export:

GSXRMovistar · 24 Apr 2023 at 15:41

2019 iMac
CPU: 3.6 GHz 8-Core Intel Core i9
RAM: 48 GB 2667 MHz DDR4
GPU: Radeon Pro 580X 8 GB
Export time 10x: 10.2 seconds
Export time 100x: 1:49.02 min/seconds

Raymond Lin · 24 Apr 2023 at 16:13

I'll lose, it's a 10 year old Intel i7...Dual Core.

5mins?

GSXRMovistar · 24 Apr 2023 at 16:18

TBH I was thinking to myself I don't recall exports taking long on my Mac, but with a newer CPU and expensive GPU being just 9 seconds I was expecting my 4 years old system to be a lot slower. For the 1 sec difference I'll be skipping any upgrading just yet!

BananaDunka · 24 Apr 2023 at 16:22

It's interesting for sure, maybe the Mac OS has deeper integration with stuff like this meaning they process faster for exports, is the Mac version using GPU acceleration for exports too?

Worth noting that your Radeon Pro 580X is a productivity centric GPU though, so it makes sense that it would be very good if it too is being utilised for exporting!

What would be very interesting to see is a batch export of 100 RAWs the same way, could I trouble you to test this out @GSXRMovistar? Just copy the 10x DNGs to total 100 copies and import/export as per above? I'll do the same - Just want to see how much of a difference is made when we factor in a big batch that would saturate the RAM, VRAM system. Copy them into subfolders so Windows doesn't moan about the same filenames or something, then make sure LR is set to not ignore duplicates

Raymond Lin · 24 Apr 2023 at 16:34

M1 Mac mini, 16G Ram

19 seconds. I'd thought it'll be faster....

BananaDunka · 24 Apr 2023 at 16:37

I've updated my OP to add the 100x DNGs export, 44 seconds to do 100 and much higher memory saturation than the 10x export, I suspect because 10x is so small it's not that significant, whereas 100x needs to saturate more RAM and VRAM, see how over 15GB of VRAM was using during the 100x export and 27GB of system RAM!

GSXRMovistar · 24 Apr 2023 at 16:38

For the 100 photos it came in at 01:49.02, so pretty linear scaling.

GSXRMovistar · 24 Apr 2023 at 16:40

If I get a chance I may also try on my old MacBook Pro which I know will be a lot slower.

BananaDunka · 24 Apr 2023 at 16:43

Very interesting, so it sounds like the 8GB of VRAM for you was saturated which is why you saw a linear export time for the 100 batch, whereas my VRAM went up to +17GB. Now I'm wondering if outright VRAM helps more or outright GPU cores - Anyone with a 24GB 3090 or 7900XTX on here?

2StepSteve · 24 Apr 2023 at 22:18

I've recently dumped Windows and gone full blown Mac, due to reasons

.

10 RAW Export

Macbook 14" 2023 (M2 Pro 12 core CPU / 10 core GPU / 1TB SSD / 16GB RAM): 12 seconds

Mac Mini 2023 (M2 Pro 12 core CPU / 10 core GPU / 1TB SSD / 32GB RAM: 11 seconds

The above are literally the same machine besides the RAM difference.

I couldn't stomach paying for 32GB on the Macbook, am happy waiting the extra seconds when exporting on the move.

BananaDunka · 24 Apr 2023 at 22:23

You could say such a small difference is just down to variances and margin of error too tbh, looks like the PU and CPU being the same result in basically the same results :cool:

Any chance of trying the 100 export?

2StepSteve · 24 Apr 2023 at 23:10

Totally, it's definitely within margin of error, too close to call.

In the name of science

.. I tried again with 100 of the samples:

1 minute 8 seconds on both Macbook M2 Pro and Mac Mini M2 Pro.

I would have expected to see a slight difference between the 16GB and 32GB, even a few seconds. As what is interesting is the 32GB Mac Mini peaks at 19.8GB RAM during the export. The Macbook with its 16GB peaks at 14.9GB of usage of LR. On the Macbook process monitor shows 1.4GB of cache being used and 5.5GB of swap during the export.

The SOC on the M processors are a bit insane, cache and swap are keeping up (in this situation!) with the speed of memory. Makes me feel even better for skimping on the RAM for the Macbook

BananaDunka · 25 Apr 2023 at 17:31

Not too bad that

Good to know that the days of MacOS being ahead of PC for creative works are long over, though

Raymond Lin · 26 Apr 2023 at 09:49

Got to say though, that a £600 Mac Mini can get a respectable time vs a £1500 graphic card, not counting the rest of it...is amazing. The upgrades to RAM might be overpriced, but the value for money aspect, it isn't even a debate that is a bargain.

BananaDunka · 26 Apr 2023 at 09:54

I guess it also highlights Adobe's generational updates to the application. They seem to increase how much GPU acceleration is used each major version, like previous to the last major update there was no GPU accelerated exporting, now there is, in another past update GPU acceleration was only in certain things to some degree, now most of the develop module is accelerated as well as the mouse hover over presets so you get 0ms latency on the preview of the preset in the main window when you hover over a preset and then run the mouse over a large list of presets to see what it will look like.

I suspect Adobe will further enhance GPU acceleration to take advantage of more shader cores etc when a GPU has more of them to offer because at the moment the GPU is asleep during these exports, You can see from my screenshot that a 4090 is only 20% utilised during that 100 export batch, can you imagine what the export speed would be if even 50% was utilised?!

I think this is all good news going forwards.

The other thing is highlights is that you don't need the fastest GPU to take advantage of the superior speed of GPU acceleration in Lightroom/Photoshop, as any moderate GPU will run circles around CPU only acceleration!

Captain · 27 Apr 2023 at 07:03

I feel like this isn't fair on my 16 core windows 11 system but here's the results:

CPU: 4.4GHz Ryzen 9 5950X
RAM: 64GB 3600MHz DDR4
GPU: RTX 4090
Export time 10x: 4.1 seconds

I'm confused how to export 100 photos? Do you mean add the files 10 times?

BananaDunka · 27 Apr 2023 at 08:45

Impressive looks like the extra cores are assisting, the 100x is just copying the 10 DNGs in a folder, then pasting 9 times and importing those 100.

Captain · 27 Apr 2023 at 16:45

mrk said:
Impressive looks like the extra cores are assisting, the 100x is just copying the 10 DNGs in a folder, then pasting 9 times and importing those 100.

Tried that, it’s not playing ball

Keeps trying to get me to overwrite the previous files rather than duplicating them!

BananaDunka · 27 Apr 2023 at 17:50

Windows 11? Copies with new names for me no probs: