Updating AI presets is slow but not using available PC resources - CPU and GPU only 20%

[paulg66021388]

Hello.

Using Lightroom Classic 13.1 (up to date version on todays date) when I import all the images from a few photoshoots I apply the Adaptive AI preset "polished portrait" to approx 2,000 images it is taking over an hour. Now I understand these things take time, but I have a powerful computer and when I look at my pc performance during this operation there is nothing much going on! CPU is at around 20%, GPU is just going between 1% and about 20% and there's no other bottlenecks visible, RAM, disks etc are all just idling.

I've paid for and built a powerful machine so these kind of tasks should be faster - why is there nothing in my system maxed out? 

Regards

Paul 

[JohanElzenga]

One hour for 2000 images is 3600 seconds for 2000 images, or less than 2 seconds per image. What do you expect?

[paulg66021388]

Hi. I know there is processing to be done for every image and I don't
expect it to be done instantly. It's a great feature and 2 seconds
certainly isn't unreasonable. But I'm asking why my pc isn't working at
full power to get the task done as quickly as possible?

[GoldingD]

While I do not think my following inquiry will help with the develop module application of the preset. 2000 photos is way way too many unless you want to go out to dinner, and hopefully return to see that the mod is accomplished. I do have two observations, and one inquiry..

1. GPU NVIDIA GTX 1660 Super driver 31.0.15.4601 20/10/2023. As you can see that driver (v546.01) is very old. Over at NVIDIA the current driver (Studio) is v 551.23 (WIN 10 or 11).  Why is your driver so old? 

2. GPU NVIDIA GTX 1660 Super, 6 GB. Released back in 2019. Not so powerful any longer, a tad old.

Q1. What do you have in /preferences/performance/ for Camera RAW CACHE limit?

[paulg66021388]

Hi thanks for your reply!

Well the driver is from just over three months ago so I don't know if I'd
agree that it is very old!
I have the nVidia software set to auto update and I update whenever there's
a notification to do so - I've been away the past couple of weeks so hadn't
the Jan 24th update done. Anyway I've updated to the latest now and I don't
see any difference.

Yeah agree the GTX 1660 isn't the most powerful card out there but it meets
and exceeds all Adobe's recommendations for Lightroom. I'm not gaming or
editing video or anything that needs bleeding edge GPU performance. I'm not
aware of any particular feature this card lacks that Adobe requires

Camera Raw Cache is 100 GB and it's on a fast PCIe 4.0 NVMe drive

I wouldn't agree 2000 is too many photos either - a wedding photographer
could easily shoot 2000 in a day and would want to apply a portrait preset
to them all. These Ai features are all about making it easier for us and
being able to apply settings to a large batch is a great way to do it! It
did complete the batch in about 70 minutes which isn't unreasonable. But
even doing smaller batches of just a few images where I'm less likely to
walk away and come back later it would be appreciated if three minutes
could be one minute instead!

But to get back to my original point... I just want to know why my computer
isn't being pushed to its limits when performing a big batch task like
this. When I do an export with a batch of files I see my CPU at 100% for
however long it takes. My GPU acceleration is set to Auto by the way...
I've also noticed that if I go into custom settings and enable all the
options for GPU a batch export takes about 60% longer than when using CPU,
but again I'm not seeing the GPU at 100% it hovers about 20%-30%.

[JohanElzenga]

I think @D Fosse may have already answered your question.

[John R Ellis]

My Macbook Pro M2 Max (2023), which has a very fast CPU and GPU, takes about 0.75 seconds to compute a Subject Mask. So your observed time of about 2 seconds on a much slower GPU seems normal.

While updating a batch of Subject masks, the CPU is at about 35% utilization, and  GPU is about 35%, roughly comparable to what you're observing.

As mentioned by D Fosse, LR has to move the photos pixels from main memory to the GPU's memory (VRAM), do the computation, and then move the result back. That schlepping of pixels takes time too, which isn't measured by Task Manager. Different systems have different CPU and GPU speeds and memory bandwidths, so the utilizations will vary.

[D Fosse]

All that data has to go somewhere. So when it's not doing the math, it's probably shuffling it back and forth to memory and disk via operating system paging.