P: Status of Apple Neural Engine Use?

[jhapeman1]

I wanted to follow up on this topic as it's been radio silence on it for nearly 8 months since AI DeNoise was launched.  In the meantime we've not only had multiple updates by Apple to Ventura, we've also had the release of MacOS Sonoma which is rapidly approaching the release of version 14.2.  

While I have pretty much left Topaz behind with the release of Adobe's DeNoise and my dissatisfaction with earlier releases of PhotoAI, I just checked it again and with their model updates its now back to being the equal of Adobe's Denoise, but it's a lot faster--it takes less than half the time to process a file on my M2 Ultra with Topaz as it does with Adobe DeNoise.  They are using the Apple Neural Engine, regardless of any bugs--and this is confirmed as I can monitor the power draw of the neural engines in the console as I process images.  DxO is also using the neural engines.  Clearly there are ways to work around the bugs, or these companies have already taken advantage of an Apple fix that we don't know about. 

Can we get ANY update from Adobe on this topic?  Is it still broken on the MacOS side?  Is there a timeline for any future use of this on the Mac?  Right now its quite fast, but its clear it could be a lot faster using the ANE (something even Adobe mentioned at release).

[ShootingMountaineer]

Same question over here.  I have started trying to incorporate AI denoise into my timelapse work on an M1 Max but...the time it takes to apply to hundreds of images is maddening.  And occassionally I'll find errors in random frames of the timelapse and have to back and tweak the whole process over again.

[WCollier]

I'd like to know too. I have a Mac Mini M1 2020. I'm not thrilled with Topaz and DxO for various reasons, but it takes me 2-3 minutes to process 1 picture in LR. Although I prefer the results thru LR, I have too many photos to process at one time.

[Brandon34692959shdw]

Very frustrating that adobe and apple have not addressed this issue, especially Apple. They market the heck out of neural engine as the best new thing,  it then make it impossible for it to be used due to bugs. I had to process 400 images using AI Denoise and it took three hours! My understanding is the neural engine could do it in 25-30 minutes.

[jhapeman1]

At this point I put more on Adobe for this as both Topaz and DxO have workarounds for whatever issue is there that lets them use the ANE.  You can monitor it in the console and confirm this.  I stopped using Topaz PhotoAI when Adobe released this becuase they were atually behind Adobe, but when I just looked again at the newest version its massively improved and now is better than Adobe's in some ways, and faster.  

[Brandon34692959shdw]

How much faster is it? Right now it take's Camera RAW 28 seconds to process AI Denoise a 24MP Canon R3 file. I rutinely process 200-300 images at a time for clients.

[Rikk Flohr: Photography]

Effective with CR 16.3, LrC 13.3, and LrD 7.3, the Apple Neural Engine is active.

[jhapeman1]

Awesome news!

[Brandon34692959shdw]

Just updated to Camera Raw 16.3 and can confrim using Denoise AI is now TWICE as fast! It took my machine 4 minutes to process 10 images (24mp Canon R3 files) using the previous version. CR 16.3 only took 2 minutes. Thats a huge improvement! Incase anyone cares, im running a 16in Macbook Pro, M1 Pro with 32GB of ram. Sonoma 14.4.1.

[Rikk Flohr: Photography]

Effective with the Max 2024 October release, the Apple Neural Engine has been disabled. 

[RuuduitdeGraaf]

Hi Rikk, 
That sounds  like a problem or a step in the wrong direction (disabled)
I assume that Denoise AI is much slower now again ?

[Rikk Flohr: Photography]

The speeds will return to 13.2 levels but the results - particularly in the quality issues,  particularly in deep shadow areas should improve substantially. 

The decision to opt for better quality over modest speed improvements was made in favor of image quality. 

One additional detail: 
Check out the Camera Raw Tech Preview detailed here: https://community.adobe.com/t5/camera-raw-discussions/technology-preview-enhance-denoise-super-resol...as it will show the new direction that Enhance will be taking. 

[RuuduitdeGraaf]

Thanks for clarifying, and read the link your provide. 
Interesting that the in-between DNG file is not needed anymore. Will try out soon.
Is this applicable for LR as wel as ACR? 

[Rikk Flohr: Photography]

CR today. LR in the future.

[RuuduitdeGraaf]

Found it in CR. Thanks. Screenshot for anybody curious how it is implemented (windows)
(after turning on new AI features and settings panel in AC preferences - Technology previews 

no further questions 

[FMPVince67]

isn't it possible to use Apple's meural emgine with better precision ? strange that Topaz and Dxo can work with it but not Adobe. The speed is much better. That said, the priority has to be quality. Maybe it's time to feedback Apple about this if there's bug or limitations.

[bdmckeown]

Please clarify. Is this decision to dramatically slow denoise processing speed a temporary one? I am a sports and dance photographer and I work with literally thousands of photos at a time. After the silicon speed increase, I was able to apply AI denoise to dozens of photos in one sitting without sacrificing the delivery speeds that my clients demand. Now I'm left wondering if I need to abandon Adobe altogether to find an application that considers the professional obligations of working photographers in addition to hobbyists. I sincerely hope this speed decrease is temporary. 

[Brandon34692959shdw]

This is not an acceptable solution. The user should decide how they want to process their photos. WithOUT the Nerual Engine processing times are almost twice as long. I routinley batch process hundreds of 25,000 ISO images. 

[bdmckeown]

I recently updated Lightroom Classic to version 14. Since updating, it seems that the AI noise reduction feature is much slower than it was with the previous version. I'm using a Macbook Pro M1. In an earlier version, Lightroom released an update that allowed Apple Silicon processors to work with the AI denoise function. When that update was released, the time it took to apply AI denoise to a RAW photo was reduced by half and it essentially saved my career. After updating to V14 it's back to its pre-silicon update slow speed. Has LR's ability to use Apple's processor been disabled again in version 14? If so, why?

[Jakub Chrudina]

Is this a temporary disable of Neural engine support and are you working on a solution? Speed ​​on Apple MacBook Air M1 16/256 is now tragic! The noise is now practically unusable for me. Also, there is Topaz. Interesting, he doesn't have these problems...

[m cs16279208]

Hmm... major update to ver 14.0 and the Denoise processing time is now worse. This is NOT an IMPROVEMENT feature and instead is a step backwards.  Adobe's reply that the quality is better is not IMO justification for the reversal.  Many LrC users can easily see this function being performed by third party apps with significantly shorter time frames. Yet Adobe engineers have to date shown the incapability of doing the same.  Who ever is managing this group needs to put some fire to thier feet or step aside for someone who can. 

[Jakub Chrudina]

I agree! This is a major mistake by Adobe. I've never had quality issues with denoising using the Neural Engine. Why did Adobe turn off the Neural Engine support completely, rather than leaving the option for the user to turn it on?! By default, Neural Engine support could be turned off, with a user option to turn it on. I know why. It would be too logical a step for Adobe.

[RuuduitdeGraaf]

I'm not sure. I've sinds day one liked Adobe's Denoise AI because I'm absolutely sure it decreases the noise significantly while increasing the sharpeness without any signs of oversharpening or artifacts! It always looks very natural.
The other tools do a fantastic job, but depending on the settings could introduce oversharpened or too smooth surfaces (plastic)
I prefer Adobe Denoise anytime over the other tools, and I prefer quality over speed. 

Please take this as my personal opinion. It might depend on your camera type and editing skills.   

[Jakub Chrudina]

I agree, it may depend on the specific RAW according to the camera brand. However, in my case (Olympus OM-1 I) the quality is exactly the same when using the Neural Engine or when it was disabled. But instead of about 26s, the process takes 46-48s! The output quality is the same! Why I can't as a user have the right to choose whether I want to use the Neural Engine or not?