fredmiranda.com Login
This topic is in the 2023 archive. You can read it and like its posts here. It is read only.

Archive 2023Lightroom Classic AI Denoise (12.3)

  
 
lighthound
Offline
Details
p.7 #1Lightroom Classic AI Denoise (12.3)
RoamingScott wrote:
lighthound wroteIt only took my system 34 days to process a single image. I liked the results but I won't be using this new LR feature as it stands right now.

If anyone has any suggestions on how to improve the speed performance I'm all ears. I'd love to stay inside LR for most of my processing, but as of now, DXO PR3 is my workflow of choice for NR work.

Are there any LR "preferences" tweeks we can make to improve performance?
I'm running this for my processing system.
CPU: AMD Ryzen 9 3900XT 12-Core, 24threads 3.8GHz
GPU: GeForce RTX 2060 6 GB
Memory: 64 GB
Win 10 Pro
A little hard to tell how severe your issue is with the hyperbole, but...

Edit > Preferences > Performance tab > Make sure Use GPU is selected for image processing and export. Update your Nvidia drivers.


Thanks I'll check to see what I have set. It was ridiculously slow. I stopped timing it after about 2 minutes or so. I'd guess it took about 4 minutes.
Apr 24, 2023 at 09:34 AM
Zenon Char
Offline
Details
p.7 #2Lightroom Classic AI Denoise (12.3)
https://community.adobe.com/t5/lightroom-classic-discussions/lightroom-denoise-remarkably-slow/td-p/13739319

Need for speed. Denoise is by far the most advanced of the three Enhance features and makes very intensive use of the GPU. For best performance, use a GPU with a large amount of memory, ideally at least 8 GB. On macOS, prefer an Apple silicon machine with lots of memory. On Windows, use GPUs with ML acceleration hardware, such as NVIDIA RTX with TensorCores. A faster GPU means faster results.

Source

https://blog.adobe.com/en/publish/2023/04/18/denoise-demystified

Apr 24, 2023 at 09:36 AM
RoamingScott
Offline
[X]
Details
p.7 #3Lightroom Classic AI Denoise (12.3)
lighthound wrote:
RoamingScott wrote
lighthound wroteIt only took my system 34 days to process a single image. I liked the results but I won't be using this new LR feature as it stands right now.

If anyone has any suggestions on how to improve the speed performance I'm all ears. I'd love to stay inside LR for most of my processing, but as of now, DXO PR3 is my workflow of choice for NR work.

Are there any LR "preferences" tweeks we can make to improve performance?
I'm running this for my processing system.
CPU: AMD Ryzen 9 3900XT 12-Core, 24threads 3.8GHz
GPU: GeForce RTX 2060 6 GB
Memory: 64 GB
Win 10 Pro
A little hard to tell how severe your issue is with the hyperbole, but...

Edit > Preferences > Performance tab > Make sure Use GPU is selected for image processing and export. Update your Nvidia drivers.
Thanks I'll check to see what I have set. It was ridiculously slow. I stopped timing it after about 2 minutes or so. I'd guess it took about 4 minutes.


I honestly don't know much about Ryzen, although I'm noticing the people with the longest times (anecdotally on FM) seem to have that in common. I have a markedly worse GPU than you, and an ancient i7 and my times are around 60 seconds on my old desktop and half that on my newer M1 mac.
Apr 24, 2023 at 09:42 AM
hoodlum90
Offline
Details
p.7 #4Lightroom Classic AI Denoise (12.3)

RoamingScott wrote:
lighthound wrote
RoamingScott wrote
lighthound wroteIt only took my system 34 days to process a single image. I liked the results but I won't be using this new LR feature as it stands right now.

If anyone has any suggestions on how to improve the speed performance I'm all ears. I'd love to stay inside LR for most of my processing, but as of now, DXO PR3 is my workflow of choice for NR work.

Are there any LR "preferences" tweeks we can make to improve performance?
I'm running this for my processing system.
CPU: AMD Ryzen 9 3900XT 12-Core, 24threads 3.8GHz
GPU: GeForce RTX 2060 6 GB
Memory: 64 GB
Win 10 Pro
A little hard to tell how severe your issue is with the hyperbole, but...

Edit > Preferences > Performance tab > Make sure Use GPU is selected for image processing and export. Update your Nvidia drivers.
Thanks I'll check to see what I have set. It was ridiculously slow. I stopped timing it after about 2 minutes or so. I'd guess it took about 4 minutes.
I honestly don't know much about Ryzen, although I'm noticing the people with the longest times (anecdotally on FM) seem to have that in common. I have a markedly worse GPU than you, and an ancient i7 and my times are around 60 seconds on my old desktop and half that on my newer M1 mac.


Upgrading the GPU will have the biggest impact. I have a 10 year old i7 system where I recently replaced the GPU with a used NVidia 2060. My 20mp OM-1 files take 16s to Denoise.
Apr 24, 2023 at 10:17 AM
RoamingScott
Offline
[X]
Details
p.7 #5Lightroom Classic AI Denoise (12.3)
hoodlum90 wrote:
RoamingScott wrote
lighthound wrote
RoamingScott wrote
lighthound wroteIt only took my system 34 days to process a single image. I liked the results but I won't be using this new LR feature as it stands right now.

If anyone has any suggestions on how to improve the speed performance I'm all ears. I'd love to stay inside LR for most of my processing, but as of now, DXO PR3 is my workflow of choice for NR work.

Are there any LR "preferences" tweeks we can make to improve performance?
I'm running this for my processing system.
CPU: AMD Ryzen 9 3900XT 12-Core, 24threads 3.8GHz
GPU: GeForce RTX 2060 6 GB
Memory: 64 GB
Win 10 Pro
A little hard to tell how severe your issue is with the hyperbole, but...

Edit > Preferences > Performance tab > Make sure Use GPU is selected for image processing and export. Update your Nvidia drivers.
Thanks I'll check to see what I have set. It was ridiculously slow. I stopped timing it after about 2 minutes or so. I'd guess it took about 4 minutes.
I honestly don't know much about Ryzen, although I'm noticing the people with the longest times (anecdotally on FM) seem to have that in common. I have a markedly worse GPU than you, and an ancient i7 and my times are around 60 seconds on my old desktop and half that on my newer M1 mac.
Upgrading the GPU will have the biggest impact. I have a 10 year old i7 system where I recently replaced the GPU with a used NVidia 2060. My 20mp OM-1 files take 16s to Denoise.


I'm going to replace my editing desktop with a Mac Studio eventually, I'm way past caring about upgrading PCs every 5 years. I've been blown away with Apple silicon...in my evaluation, my upgrade money is far better invested in that than a GPU. Just my personal opinion based on observations. I'm also dealing with files way bigger than 20mp m4/3. Doing a lot of 4k/120 editing these days as well which Mac just destroys.
Apr 24, 2023 at 10:21 AM
Zenon Char
Offline
Details
p.7 #6Lightroom Classic AI Denoise (12.3)
I did a test. Denoise removes colour noise at 1. Opening a file with the sharpening default of 40 and no sharpening. Which file will Denoise have to work harder on?










Apr 24, 2023 at 11:36 AM
Zenon Char
Offline
Details
p.7 #7Lightroom Classic AI Denoise (12.3)
Also if you go into Denoise with the default and come back out you are technically sharpening it twice increasing chance of more artefacts?
Apr 24, 2023 at 11:50 AM
jhapeman
Offline
Details
p.7 #8Lightroom Classic AI Denoise (12.3)
eyal wrote:
Can anyone with an M1/M2 MacBook Pro comment on how these new tools are pushing CPU/GPU use? Wondering if I need to consider 64gb RAM in my next laptop when I upgrade my underspec'ed M1 Pro I got with a lowly 16gb RAM.


I was in the jungle all last week photographing birds, so not only was I unable to do any testing, I was also able to get a ton of images that will *need* the denoising.

I did just do a couple of quick tests on some very noisy images to compare speed and both my Mac Studio Ultra 128GB and a 24GB Nvidia RTX 4080 take the same time on those files, wishing a tenth of a second, so pretty much margin of error--from the moment I push the "Denoise" button until the DNG is saved takes about 20 seconds. A lot of the time right now seems to be saving/creation of the DNG file.
Apr 24, 2023 at 12:45 PM
mcbroomf
Offline
Details
p.7 #9Lightroom Classic AI Denoise (12.3)
jhapeman wrote:
eyal wroteCan anyone with an M1/M2 MacBook Pro comment on how these new tools are pushing CPU/GPU use? Wondering if I need to consider 64gb RAM in my next laptop when I upgrade my underspec'ed M1 Pro I got with a lowly 16gb RAM.
I was in the jungle all last week photographing birds, so not only was I unable to do any testing, I was also able to get a ton of images that will *need* the denoising.

I did just do a couple of quick tests on some very noisy images to compare speed and both my Mac Studio Ultra 128GB and a 24GB Nvidia RTX 4080 take the same time on those files, wishing a tenth of a second, so pretty much margin of error--from the moment I push the "Denoise" button until the DNG is saved takes about 20 seconds.
...Show more →

That's the Studio with the 64 core GPU Ultra right Jeff?

It would be interesting to see how the 48 core GPU version (with 64GB I think) does given how much is being pushed on the GPU.

Thanks
Apr 24, 2023 at 12:58 PM
jhapeman
Offline
Details
p.7 #10Lightroom Classic AI Denoise (12.3)
I'm a tad bit late to this show as I was in the jungles of Costa Rica all last week when this came out. I got an email from a friend that let me know it was available so the one night where I had relatively faster Internet I downloaded it to my M2 Max MacBook Pro. I was with two other photographers and we all are current Topaz users, and we all agreed that the new LR Denoise is easily just as good as Topaz if not better.

Honestly, having done a lot of extensive testing of DxO, ON1 and Topaz in the last year, I'd give Adobe the edge overall for producing a very natural-looking result without too much smoothing or artifacts being created. I'm very pleasantly surprised. This doest not feel like the first effort it is--it's incredibly polished.

Could it be faster? Maybe. But a quick test on a couple machines on hand here show it's just as fast as Topaz overall--if I use the Topaz DeNoise AI standalone, it takes about 22 seconds to open render a TIFF, open it, process, and then save the file back to LRC. If I just process in LRC with their new DeNoise, it takes...about 22 seconds to go start to finish. Using the new Topaz PhotoAI, it takes even longer, about 35 seconds, and it's very hard to get rid of the odd sharpening artifacts that application seems to create. FWIW, I haven't even tried the newest version of DxO as with the limited testing I did before Adobe came out with this, I found the results to be worse than Topaz and the application to be even slower yet, never mind the odd subfolder file handling approach they use.

Based on what I'm seeing, my renewals to Topaz are over. So much nicer to have it all in one place, and the results IMO are superior.
Apr 24, 2023 at 01:08 PM
jhapeman
Offline
Details
p.7 #11Lightroom Classic AI Denoise (12.3)
mcbroomf wrote:
jhapeman wrote
eyal wroteCan anyone with an M1/M2 MacBook Pro comment on how these new tools are pushing CPU/GPU use? Wondering if I need to consider 64gb RAM in my next laptop when I upgrade my underspec'ed M1 Pro I got with a lowly 16gb RAM.
I was in the jungle all last week photographing birds, so not only was I unable to do any testing, I was also able to get a ton of images that will *need* the denoising.

I did just do a couple of quick tests on some very noisy images to compare speed and both my Mac Studio Ultra 128GB and a 24GB Nvidia RTX 4080 take the same time on those files, wishing a tenth of a second, so pretty much margin of error--from the moment I push the "Denoise" button until the DNG is saved takes about 20 seconds. A lot of the time right now seems to be saving/creation of the DNG file.
That's the Studio with the 64 core GPU Ultra right Jeff?

It would be interesting to see how the 48 core GPU version (with 64GB I think) does given how much is being pushed on the GPU.

Thanks


Here's the thing: It's not the GPU from what I can see. It certainly hits that, but Adobe explicitly states that it uses the Apple Neural cores, and the Tensor cores in the Nvidia GPUs. I think this is why people are seeing some very slow performance on older Nvidia GPUs--they lack the tensor cores needed to accelerate this processing. Right now the MacBook Pros, Mac Minis and Studios all have the same number of Neural Engine cores--16. I'll verify tonight but my gut feeling is that they all have similar processing times. I did see LRC grab about 4GB more of RAM when processing the images, so as I've said before (and as Adobe now recommends), more RAM is better. I still maintain the 32GB on Apple Silicon should be the minimum a photographer should consider, and 64GB if you can afford it.

BTW, as noted above, the Studio is keeping pace with an Nvidia RTX 4080 GPU with 24GB of RAM which is pretty impressive.

Apr 24, 2023 at 01:15 PM
RoamingScott
Offline
[X]
Details
p.7 #12Lightroom Classic AI Denoise (12.3)
jhapeman wrote:
Right now the MacBook Pros, Mac Minis and Studios all have the same number of Neural Engine cores--16.


Unless you have silly money and go M1 Ultra with the Studio, then you get 32 NE cores. I don't know anyone with that rig, personally

I can understand wanting at least 32GB of ram on Apple, but I'll reiterate that my 16gb M1 MBP leaves me wanting for nothing during intense processing. It's amazing how an "underpowered" M1 machine flies compared to its Intel step-brother.
Apr 24, 2023 at 01:20 PM
jhapeman
Offline
Details
p.7 #13Lightroom Classic AI Denoise (12.3)
RoamingScott wrote:
jhapeman wroteRight now the MacBook Pros, Mac Minis and Studios all have the same number of Neural Engine cores--16.
Unless you have silly money and go M1 Ultra with the Studio, then you get 32 NE cores. I don't know anyone with that rig, personally

I can understand wanting at least 32GB of ram on Apple, but I'll reiterate that my 16gb M1 MBP leaves me wanting for nothing during intense processing. It's amazing how an "underpowered" M1 machine flies compared to its Intel step-brother.


Ahh yeah, good point, but you know me--I have two Ultras. I'll have to see how the Max compares to the Ultra as I do more testing. It seems like the bottleneck I'm seeing is the DNG file being written/created. Not sure what all is happening there, but it's the slow step on my Studio Ultra and PC at work.
Apr 24, 2023 at 01:32 PM
mcbroomf
Offline
Details
p.7 #14Lightroom Classic AI Denoise (12.3)
Yes, the Ultra chips have 32 Neural cores, the Max has 16. You also drop CPU cores and GPU cores when going down from the Ultra to the Max.

What were the files from, Sony A1?

Apr 24, 2023 at 02:06 PM
Zenon Char
Offline
Details
p.7 #15Lightroom Classic AI Denoise (12.3)
Zenon Char wrote:
I did a test. Denoise removes colour noise at 1. Opening a file with the sharpening default of 40 and no sharpening. Which file will Denoise have to work harder on?


It may be a bug.

https://www.lightroomqueen.com/community/threads/workflow-using-adobe-denoise.47607/
Apr 25, 2023 at 06:04 AM
hoodlum90
Offline
Details
p.7 #16Lightroom Classic AI Denoise (12.3)
You will want an RTX card from any of the 30 or 40 RTX series or the 2060 and above from Nvidia or the Apple M1/M2, as Denoise is using the Neural engine in these GPUs to improve performance.

https://blog.adobe.com/en/publish/2023/04/18/denoise-demystified

Finally, we built our machine learning models to take full advantage of the latest platform technologies, including NVIDIA’s TensorCores and the Apple Neural Engine. Using these technologies enables our models to run faster on modern hardware.
Apr 25, 2023 at 07:15 AM
jhapeman
Offline
Details
p.7 #17Lightroom Classic AI Denoise (12.3)
So for those who care...the additional neural engines in the Studio Ultra models really do make a difference in overall speed. My 16" M2 Max has 16 neural engines vs. the 32 in the Studio Ultra, but they also run at a faster clock speed. On the Studio Ultra the total process from hitting the Denoise button until the DNG is saved on average takes 21 seconds; on the M2 Max it takes on average 26 seconds.

As a reminder my Nvidia RTX 4080 takes the same 21s to process the same files. The only consumer level card that would probably perform better would be an Nvidia RTX 4090, but those are $2000 video cards.

BTW, it's hard to tell when Adobe is doing the actual denoise calculations, but I think it's part of the whole DNG creation process. It's definitely different on my PC and Studio Ultra--the Studio will load an image preview of the Denise in less than 2 seconds; it takes the PC with the Nvidia card about 6 seconds to do that, but ultimately the total time to finish is very similar for both.
Apr 25, 2023 at 07:21 AM
mcbroomf
Offline
Details
p.7 #18Lightroom Classic AI Denoise (12.3)
It looks like an M1 Studio/Ultra is a good buy for a desktop for this AI tech if needed. No point in waiting for an M2 version (next year according to some rumors) as the M2 chips have the same # of Neural Cores as the M1. Go Ultra though for twice as many.
Apr 25, 2023 at 08:19 AM
jhapeman
Offline
Details
p.7 #19Lightroom Classic AI Denoise (12.3)
mcbroomf wrote:
It looks like an M1 Studio/Ultra is a good buy for a desktop for this AI tech if needed. No point in waiting for an M2 version (next year according to some rumors) as the M2 chips have the same # of Neural Cores as the M1. Go Ultra though for twice as many.


Yes and no; I personally suspect that we will see an updated Studio with an M2 Ultra, and that should be quite a bit faster. My belief is that the new Mac Pro will use an M3 Ultra/Extreme and some add-in cards, and quite possibly one of them will be an Apple Neural Engine accelerator, similar to what we saw with the specialized Afterburner cards in 2019. Of course, that's all guessing, but I feel like it's a pretty good guess.

If I had to upgrade today and had the money for the Studio Ultra, I would go that route. If I could wait 3-6 months, I would do so, as I think we'll see an interesting announcement or two in June at WWDC.
Apr 25, 2023 at 08:24 AM
mcbroomf
Offline
Details
p.7 #20Lightroom Classic AI Denoise (12.3)
That's interesting. I can wait (but not too long I think), certainly for WWDC though. I could also go for an M1 Studio if I see good mark downs before a new version is released.
Apr 25, 2023 at 08:29 AM
FM ForumsPost-processing & PrintingJoin Upload & Sell
  
 

You are not logged inLog inRegister