Explore related tutorials & articles
7
Replies
The performance of Denoise has nothing to do with CPU cores. You can double or triple your CPU cores and get no improvement in Denoise timing.
What does improve it is the number of GPU cores. On Macs (it works differently on PCs), doubling or tripling the GPU cores should roughly run Denoise 2x or 3x as fast, respectively.
Other Photoshop features use a mix of components, but Denoise and other AI-related features tend to lean heavily on the GPU.
This is a chart showing Denoise times across various Mac models. You can clearly see how GPU cores map to Denoise performance, with the only variations accounted for by generational improvements (M3 to M4 for example).
Here is a link to the video where those test results came from.
Pro Photo & Video test on Every MAX from M1 MAX
If you are concerned about how Mac components are used (CPU, GPU, memory) you should watch that video, because you should clearly see that different Photoshop and Camera Raw/Lightroom features use different mixes of those components. It is not possible to use all components at maximum all the time, just as you don’t use every ingredient and gadget in the kitchen for every meal.
Also, if you’re using Denoise, you’re in Camera Raw, not Photoshop. If you’re interested in making the most use of CPU cores, that is a little more possible in Camera Raw because it can process images in large batches, such as syncing settings or exporting. If you do a bulk edit or export, and you watch the CPU History or GPU History in macOS Activity Monitor, you can see many CPU and/or GPU cores being used.
Photoshop can only use so many cores because it’s usually editing just one still image document. If you really want to make the most use of all cores in a Max or Ultra, you should be using a video-editing or 3D application.
---
I mean: what do I have to do to run Denoise faster than 10 seconds?
By francescotacconi
You may need to adjust your expectations, and also have a better understanding of the context: Whether you have a Mac or PC, right now, Denoise timings for a 45 to 60 megapixel image are:
“Slow” is more than about a minute per image.
“Fast” means under 15-20 seconds per image. This speed is reached by the fastest Macs and PCs today.
So if you want under 10 seconds, especially for a high-megapixel image, that is actually excellent. Your Mac is doing very well, among the best available Macs or PCs, and further improvement is difficult to achieve.
The chart I posted shows the fastest time as around 13 seconds, with most Macs being far slower.
Your expectations are that Denoise should be almost instant. That has never been the case. It is exceptionally intensive to calculate, so much so that CPU cores are not practical to use (they would be too slow). So AI-friendly GPU cores are used..
Honestly, I don't really care about your videos since I had a Mac Studio Ultra M2 and now I have a Mac Studio Ultra M3, and a stopwatch. So let's forget single and multi-core, does the GPU on the Mac Studio Ultra M2 look the same as the Mac Studio Ultra M3 to you?
By francescotacconi
But to consider the GPU, we can’t forget about multi-core, because, as is well established, Mac Denoise performance depends on the number of GPU cores.
It looks like your current M3 Ultra has 60 GPU cores. How many GPU cores did your M2 Ultra have? The tests suggest that if your M2 had the same number of cores, the M3 will probably be slightly faster only because of the generational improvement, but not dramatically faster.
What the tests show is that, for Denoise, the only Mac that will process Denoise meaningfully faster than a 60-core Apple Silicon GPU is the 80-core GPU option. Because it is 25% more cores, that should provide a roughly 25% further time savings.
Honestly, I don't really care about your videos
By francescotacconi
You should, because they explain how this works. If you don’t want to study the tests, you won’t understand. But if you want to understand, why would you want to ignore the facts about how Denoise is processed?
Also, they are not my videos.
AdChoices