In my opinion, DxO is currently the market leader when it comes to reducing noise in RAW photos.

One of the headline new features of DxO PhotoLab 6 and PureRAW 3 was the introduction of a new noise reduction model called DeepPRIME XD.

The ‘XD’ in DeepPRIME XD is what stands for ‘eXtra Detail’ and promises both better noise reduction results and more detailed images.

Lets take a look to see how it fares

What is DeepPRIME?

DeepPRIME and DeepPRIME XD are what DxO has named it’s photo noise reduction technology.

The technology is built into the latest versions of PhotoLab 6 Elite and PureRAW 3, which are their RAW image processors. The software will analyse the RAW images from compatible cameras and provide very good noise reduction during processing.

The Images

For this test I have chosen two bird images taken with a Nikon D500 at high ISO. The D500 is a 20 megapixel crop sensor camera, so has about the same pixel density as a 50 megapixel full frame camera.

The full sized images with noise reduction applied can be seen below:

The Test

I ran these images through DxO PureRAW 3 on my Windows 10 PC that I normally use for image editing. You can see the settings I used below – with the only thing changing between the images being the swap between ‘DeepPRIME’ and ‘DeepPRIME XD’ options for noise reduction.

DeepPRIME vs DeepPRIME XD Speed

Before I get into the image results – just a quick note on the speed differences between the two processing modes.

My PC Specs are as followed (yes, I know my graphics card is old 🙂 )

• Ryzen 3900X CPU (12 core, 24 threads)
• AMD Vega 64 GPU
• 32 GB Ram
• Images processed off a speedy SSD drive

I ran a batch of three images. The two 20 megapixel images above, along with a 45 megapixel landscape photo from my Nikon Z7II. Timings are below:

Mode	Processing Time
DeepPRIME	30.5 sec
DeepPRIME XD	49 sec

Based of that result, with my hardware, DeepPRIME XD takes about 60% longer to process images than the standard DeepPRIME. (Read on for a second test with more images)

At an average of 10 seconds vs 16 seconds per image – speed will probably never be the deciding factor on which mode to use for me.

Another Speed Test

I thought I should try a second speed test with a few more images, so I chose a batch of 10 images from both my 45 megapixel Z7II and another 10 from my 20 megapixel D500.

Camera & Method	Total Time	Time Per Image
Z7II: DeepPRIME	3min 10s	19s
Z7II: DeepPRIME XD	4min 38s	28s
D500: DeepPRIME	1min 19s	8s
D500: DeepPRIME XD	1min 53s	11s

So, DeepPRIME took about 50% longer for the Z7II files, and 40% longer on the D500 ones.

It’s pretty safe to say that on average (on my hardware), it takes around 50% longer to process files with DeepPRIME XD than DeepPRIME.

The Results

So lets get into how the actual images turned out.

The Robin Image

The robin image was taken at ISO 18000 (ouch!). You can see a huge amount of noise in the original.

These are 100% crops so you can reall see the differences between zero noise reduction and the two DeepPRIME modes.

The difference between DeepPRIME and DeepPRIME XD are fairly subtle, but the XD version does show more detail in the features and overall appears to be a sharper image than the regular DeepPRIME version.

Now lets take a look at the background of the image to see how much noise was actually removed:

It’s pretty clear that both DeepPRIME algorithms did a remarkable job when compared to the original image.

DeepPRIME XD has definately removed more noise from the background of the photo compared to regular DeepPRIME – however, it has also introduced some less than desireable artifacts.

[adinserter block=”1″]

Take a look at the tip of the beak, the ‘whiskers’ and the background towards the middle of the picture. Because the algorithm is tuned to grab extra detail, it has struggled with these areas and has ‘enhanced’ some details that it probably shouldn’t have.

The regular DeepPRIME has resulted in a slightly more natural looking background in this case. The XD artifacts wouldn’t be too difficult to clean up in Photoshop, and might be worth the effort if you’re looking for a sharper, more detailed image – but as they stand – regular DeepPRIME did a better job here in my opinion.

The Honeyeater Image

The second image I’ve chosen was shot at a much more reasonable ISO 8000. I would rarely shoot above this as with light levels this low there are usually other issues that would result in a poor outcome.

First, take a look at a more detailed part of the image

Once again, both algorithms have removed a significant amount of noise from the photo.

Also once again – the DeepPRIME XD version has pulled out more detail and provided a sharper image. You might be able to see around the bird’s cheek though… it has introduced a few artifacts where it has gotten a little confused about the detail vs noise choices it has had to make.

[adinserter block=”1″]

Lets take a closer look at the background of the image.

Firstly, I am REALLY impressed that it has been able to remove the noise, but still retain those dust particles that were in the air (the light dots in the top right of the images below). The fact that it could differentiate between the two is very impressive!

Overall, DeepPRIME XD has reduced the noise more than regular DeepPRIME. There is a small area near the bird’s cheek that has retained a little more noise than the rest, and there are some ‘swirly’ parts near the dust particles.

Once again, DeepPRIME has produced a more natual background, but the DeepPRIME XD image is cleaner and wouldn’t take much to fix in PhotoShop if you’re after the extra details it produces.

Summary

I will expand this test a bit later with some more images – but from the couple I have looked at here is is pretty easy to say:

• DeepPRIME XD does pull out more details from the image
• DeepPRIME XD does remove more noise than regular DeepPRIME
• The tradeoff is that it might produce some artifacts that will need to be cleaned up in PhotoShop afterwards

What do I choose?

DeepPRIME XD

Why?

If I’m taking the time to process an image, especially one that requires as much noise reduction as these – I will always want to get the best results possible. I’m almost never in a rush to process my images so the extra time it takes to run the DeepPRIME XD algorithm and do a little cleanup on any problem areas it might create is worth the effort for me.

As I said above – I will do some more testing on a variety of other images from a range of ISOs… but my initial observations are that these artifacts are less of an issue as ISO levels go down (which makes sense).