GPU for Lightroom AI processing (2024)

A DX12 GPU with a minimum of 8 GB of VRAM is recommended for Lightroom. The new AI features use machine learning models which take advantage of dedicated AI accelerators present in current gen GPUs.

Any RTX40xx or RX7000 graphics card fits the criteria, with higher models providing faster AI acceleration.

QuietBob said:

A DX12 GPU with a minimum of 8 GB of VRAM is recommended for Lightroom. The new AI features use machine learning models which take advantage of dedicated AI accelerators present in current gen GPUs.

Any RTX40xx or RX7000 graphics card fits the criteria, with higher models providing faster AI acceleration.

Click to expand...

Great, that confirms what I have been assuming from my reading. Is there an easy way to evaluate each vendors "version" of the RTX40xx models? They vary so much in price. I know some are OC and vary by the number of fans and aesthetics.

I can't provide a GPU tier recommendation for Adobe AI Denoise specifically. Deep learning models in general are very compute intensive and can utilize huge amounts of video memory. Still, this benchmark of a competing AI de-noising algorithm shows little difference between a $250 and $1400 GPU:

If you're looking for information on particular card models, check out the in-depth graphics cards reviews on this site.

Topaz runs much faster than LR Denoise, even with integrated graphics. My Intel NUC 12th gen with i5 CPU and Iris graphics processes Canon CR3 files from the R6 in under 10 seconds - versus LR Denoise which processes the same files in 5+ minutes. I'm guessing this is why there's so little performance difference with Topaz on various machines - it's fast on all of them. However, I find LR Denoise results much better, as Topaz makes people's skin look a bit too artificial and waxy for my taste.

For general info, DxO's PRIME XD noise processing takes about 75 seconds on the NUC. Totally acceptable for me, given the exceptional quality. Still, I'd like to keep up with Lightroom as Denoise did an amazing job on some school science fair pics.

DRODZ said:

Lightroom AI Denoise is taking 5 min to process a 45 MB file. Resource monitor shows memory and cpu below 25%. I am using AMD 5700G (integrated graphics) and it is at 100%.
So I temporarily used a RX 6650 XT (8 GB) and the time dropped to 1 min. I am now in search for a new discrete GPU. I have done a ton of research but cannot find what features are important for AI processing? Is it VRAM, base clock, boost speed, memory speed, core speed, tensor cores, Cuda cores, Directx 12 Ultimate (12_2), RAM bandwidth etc? It is hard to compare these cards when I don't know what specs are important to LR AI and Photoshop. It seems ALL reviews seem focused on Frames per second.
I am sure LR and other apps will be using more AI very soon so I want to future proof as much as possible. I am not looking so much for a GPU recommendation as I am understanding what specs are critically important to this type of processing. I know the RTX 4000 series are not getting great reviews for the price, but it seem that the new architecture (ada lovelace) could be important. Would a GPU that is good for video editing be the type of card I should focus on?
Side note: While DeNoise is processing my screen flickers, this is common from what I've read. comments?
Thanks for any help you can provide.

Click to expand...

I’m on the same page: looking for an optimum future-proof GPU for photo and video editing. I’m still waiting for some other components for my new custom desktop. But here are some data from my recent experiments with AI denoise in ACR on my 2 current systems: 1) Acer laptop swift X 14 late 2021 model with amd ryzen 5 5600u, 16 gb DDR 4 RAM and rtx 3050 mobile (4 gb). 2) old PC with intel core i7 3770, 32gb ddr3 RAM coupled with radeon rx 580 8gb sapphire nitro+. For the experiment I used 10 compressed RAW files of my sony a7r4a with moderate noise (shot just before dawn at 800 iso), each file is 60-ish Mpxls. I opened these 10 files in ACR Photoshop and set ai Denoise to 50%. The results are the following: 1. Laptop - 13 min. (1 min. 18 sec. per photo) 2. PC 20 min 10 sec. (roughly 2 min. per photo).
Bear in mind that rtx 3050 mobile goes neck to neck with rx 580 in terms of common gaming synthetic tests. On the other hand amd ryzen 5 5600u has 15470 scores in passmark while i7 3770 - just 6400 scores. So, CPU also matters for ai denoise. Then I installed gtx 1080 ti on my old PC. 1080 ti is twice as fast as rx 580 (according to technical.city). No wonder that it cut the time exactly in half on my old PC: 1 min. per 60Mpxl raw file. I bought this msi gtx 1080 ti gaming X for just $160 so it is a good bang for the buck. I will do the same experiment on my new PC when I assemble it (based on amd ryzen 9 7950x) and I will let you know the results. I am also considering rtx 3090 which you can buy used in the range of $600 - 700.