It is my guess. Because DXO made an effort to fix this issue in the last update and I think if they actually were capable to fix it, they would do it. The current solution looks half-baked, so my guess is it’s not that DXO lazy, but there are some limitations from Apple’s side too.
Adobe could be using a completely different approach or even technology, I don’t know any tech detail on that.
3 Likes
According to Adobe, they actually use the GPU for denoising as there’s “an issue on the Apple side” with the ANE (search for a reply and confirmation from Rikk Flohr).
3 Likes
As far as I’m concerned, the bug is fixed with PL6.6 running on macOS 13.4 (Release Candidate)
3 Likes
I can confirm that the PL 6.6.0 build 50 together with final release macOS 13.4 (22F66) have corrected the purple tone for me.
I also tried the previous PL 6.5.1 built 49 and then the purple color tone was back.
4 Likes
Using 6.6 and processing RAW files from Olympus OM-D E-M1, I am seeing no difference between images processed with the Neural Engine and GPU using an M1 Mac Mini and Ventura 13.3.1. I used Prime XD noise reduction and optical corrections and the Neural engine was much faster at 13 seconds vs. 41 seconds for the GPU.
4 Likes
Thanks, that’s good news - Lightroom will be much faster when Apple fix the issue.
1 Like
Glad ver6.6 finally resolves the neural engine issue. Below are time comparisons between my two Macs doing default corrections and denoise on 306 (21MB) RAWs from my Sony RX10m4.
M1 MacBook Air (16GB, 8cpu/8gpu):
DeepPrime w/GPU = 54min 20sec (10.7 sec/image)
DeepPrime w/NE = 18min 48sec (3.7 sec/image)
DeepPrimeXD w/GPU = 3hrs 30min (41.2 sec/image)
DeepPrimeXD w/NE = 47min 08sec (9.4 sec/image
M1 Max Mac Studio (32GB, 10cpu/24gpu):
DeepPrime w/GPU = 18min 48sec (3.7 sec/image)
DeepPrime w/NE = 13min 59sec (2.7 sec/image)
DeepPrimeXD w/GPU = 1hr 22min (16.1 sec/image)
DeepPrimeXD w/NE = 43min 26sec (8.5 sec/image)
3 Likes
Hi @NedFlanders , I have a question regarding your two Macs.
When viewing images using your two Macs, do you ever link them up to the same monitor. And if so, how do they compare when initially rendering the raw image on the screen.
I find that my Mac Mini renders a 24mp raw image pretty much instantly on a 1920x1080 screen but there is a delay for it to fully render when I use a 4k screen for editing. I’m curious if the extra GPU cores in the Mac Studio alleviate this.
Hi @CHPhoto, let me make sure I understand the definition of “render”. Are you referring to when we view photos in the Photo Library and Customize tabs? I notice zero lag and no difference between my Mac Studio versus M1 MacBook Air when advancing through each photo in a collection. The image displays pretty instantly with both Macs. I can hit the right-arrow as fast as I can and everything renders immediately, with maybe a slight black frame between images, but it’s extremely fast like a blink of the eye. I have both Macs connected to my 43" 4K TV that serves as my desk monitor. The Mac Studio is connected with HDMI and the MacBook Air uses a CalDigit SOHO dock with HDMI.
When I had my base M1 Mac Mini hooked up to the same 4K TV everything felt really responsive and equally fast. That is why I asked about “render” to make sure we’re both understanding one another. Rendering has never even been an afterthought to me on three of my M1 Macs. Is your Mac Mini Intel-based or on Apple Silicon?
Thanks for the reply, and you ask a good question:
When you navigate to the next image, at the top right of the image it says ‘Correction preview’ and a little swirly circle appears too. After a short delay, the default lens corrections and WB corrections (and any other corrections you may have on the image) are applied and you visually see a difference between the initial (unaltered) raw file and the current state of the edit.
That’s what I was referring to when I typed ‘render’.
My machine is M1 Mac Mini with 8Gb memory connected to a 4k screen.
I typically have a delay of a fraction of a second for the ‘Correction preview’ to complete. It’s not major, but I feel it does become a longer delay with greater amounts of alterations applied to the image (including local adjustments). And when editing a big set of images, if that ‘Correction preview’ time can be reduced then that is something I’d liek to explore.
I’m trying to work out what aspect of the computer can speed that part of editing up. Is it CPU, GPU, installed memory or combination of all? Basically, where is money best spent 
@CHPhoto, ah I don’t know why I never noticed that pause before photos become sharper or clearer. I guess I’m still new to the game.
You are correct, I can see there’s a 2-3 second delay during the Correction Preview as it renders the image. Ironically, the delay is shorter on my 16GB M1 MacBook Air (2 seconds) versus my M1 Max Mac Studio (consistently closer to 3 seconds). Now that’s very odd. Both machines on the same version 6.6 build 50 of PL6, and both set to Apple Neural Engine (8 parallel streams for MBA and 10 parallel streams for Mac Studio). Final export is noticeably faster on the Mac Studio but I can’t explain why my 32GB Mac Studio renders a step slower than the MacBook Air.
On the MacBook Air PL6 is only using 1 GB memory whereas I see PL6 use as much as 12GB RAM on the Mac Studio, yet the Air is a little snappier on the render. Sorry that doesn’t really help your upgrade decision.
I first noticed something going on with my MacBook Pro — I just plugged in a 4k monitor and did not give it a second thought. That machine slowed WAY down and I originally attributed the slow down to the older GPU in it. It was worth it to me to unplug the display to process images on that machine as it was otherwise too slow.
I moved to a M1 Studio earlier this year and have been reading up on monitors. I found this article: Mac external displays for designers and developers this seems to be worth your time. There’s a good chance that you are spending GPU cycles scaling instead of processing. I’m not getting 2 second process times - I’m getting 8 second ones. To be honest, 8 seconds is blistering fast compared to what I was used to, but I know that it’s the monitor that’s likely causing this slight slow down. Later this year (probably around Black Friday), I’ll get a proper 5k display for this machine and will perform more benchmarks then.
In the meantime - I’ll note that on this machine (with maxed out RAM and GPU cores), flipping through images in PL6.6 is a breeze. They render very quickly - almost as fast as I can hit the ‘next’ button. It becomes a pleasure to wade through a few thousand shots of a show to pick out the top 200-300 winners … (and from there, only about 4-5 will ‘make it’).
Great link.
I use dual 4K 31 inch monitors with a MacBook Pro 16 Max. No slowdown here. What I do with my 4K monitors though is run them at 1920x1080. I would prefer 2560 x 1440 scaling but that quadruples the amount of memory and calculation required to render the screen. 1920 x 1080 is far more efficient as it’s simple doubling of pixels, which is Apple’s own default Retina render on its own screens.
27 inch 5K renders interface far too small for my well over 40 eyes and there are no 31 inch 5K monitors yet. Pre-Retina I used HP LP3065 2560 x 1600 pixel monitors. Those were wonderful and I still use one (18 years later) at work. My second one finally grew too dim a couple of years ago.
I’ve thought of doing that… but my heart has resisted going to the lower resolution.
With the Studio, I am able to zip through a few thousand images pretty quickly — it renders sufficiently for me to be able to determine focus/composition/etc., so I can sort through the images quickly. I couldn’t do that on my MacBook, even with no external screen attached.
I’m about to get new lenses for my over 50 yr old eyes (I’ve developed early cataracts), so I’m holding off on making a monitor purchase (technically, the 4k screen I’m using at the moment is a loaner). I’ll go to the Mac store to see if I can read a 5k screen with my new eyes in about two months (I’ll need time for them to settle in).
I note that version 6.6 does give me the default of the neural engine, finally. Also, Is there a reason why they limit the number of simultaneous processing to 10? I have more than enough engines and memory to handle far more than that on this machine…