DXO Photolab performance benchmarks

[Jacques Cornell]

TL;DR - On my Mac Studio, PhotoLab 8.1 processes DeepPRIME noise reduction at about 9 megapixels per second and DeepPRIME XD2s at 3.4 MP/s. DP XD2s performance was quite consistent, but DP performance varied unpredictably.

The skinny -

Below are results of my test of how long it takes DXO Photolab 8.1 to apply Optical Corrections and DeepPRIME or DeepPRIME XD2s to various Sony RAW files on my 2021 M1 Max Mac Studio running macOS 14.7 with 32GB & 512GB. RAW files were hosted on, and DNG files were exported to, a 1TB 1500MB/s TB3 SSD.

DP XD2s performance was quite consistent with batches of 11 and 44 RAW files of sizes ranging from 24MP to 61MP. OTOH, DP performance varied unpredictably with MP count and batch size.

Test images were from DPR's Studio Scene comparison tool, one of each ISO from 100 to 102,400 (11 images) from each of four cameras (a7III, a7IV, a7RIII & 7RV). Batches of 44 from a given camera were made by duplicating the original 11 files three times. The combined group of 44 comprises 11 files from each camera.

I ran 44-file batches for the a7IV and a7RV because I suspected that the smaller 11-file batches were unrepresentative, and the variable results from the smaller batches seem to bear out this impression. I also set PhotoLab to use 5 cores, then 10 cores, and results were virtually the same.

On average, I get about 3.2 MP/s with DeepPRIME XD2s, scaling processing time linearly with MP count. DeepPRIME was much faster (around 9 MP/s) with the small 24MP batch, the large 33MP batch, and the large combined batch than with the large 61MP batch or the other small batches (around 5.3MP/s). I have no idea why this might be. I ran some of these batches again, but results didn't change.

DxO PL8.1 with macOS 14.7

TIME PER BATCH - TIME PER IMAGE - MEGAPIXELS PER SECOND

24MP (11 files)

• DeepPRIME = 0:32 (3s) (8MP/s)
• DeepPRIME XD2s = 1:24 (7.6s) (3.14MP/s)

33MP (11 files)

• DeepPRIME = 1:12 (6.5s) (5.04MP/s)
• DeepPRIME XD2s = 1:48 (9.8s) (3.36MP/s)

(44 files)

• DeepPRIME = 2:29 (3.4s) (9.74MP/s)
• DeepPRIME XD2s = 7:13 (9.8s) (3.36MP/s)

42MP (11 files)

• DeepPRIME = 1:27 (7.9s) (5.31MP/s)
• DeepPRIME XD2s = 2:24 (13.1s) (3.21MP/s)

61MP (11 files)

• DeepPRIME = 2:05 (11.4s) (5.34MP/s)
• DeepPRIME XD2s = 3:26 (18.1s) (3.26MP/s)

(44 files)

• DeepPRIME = 8:03 (11.0s) (5.56MP/s)
• DeepPRIME XD2s = 13:40 (18.6s) (3.27MP/s)

24+33+42+61MP (44 files)

• DeepPRIME = 3:16 (4.4s) (8.98MP/s)
• DeepPRIME XD2s = 8:30 (11.6s) (3.45MP/s)

BTW, looking over my past tests, I see that PL8 ran DP XD2s about as fast as PL6 ran DP XD, and PL8.1 is about 10% faster with DP and 20% faster with DP XD2s. Nice!

Finally, I've been reading about performance improvements with macOS 15.x and would love to hear real-world reports, especially pertaining to Lightroom and PhotoLab.

--

Event professional for 20+ years, travel & landscape enthusiast for 30+.
http://jacquescornell.photography
http://happening.photos

 

[bakubo]

TL;DR - On my Mac Studio, PhotoLab 8.1 processes DeepPRIME noise reduction at about 9 megapixels per second and DeepPRIME XD2s at 3.4 MP/s. DP XD2s performance was quite consistent, but DP performance varied unpredictably.

The skinny -

Below are results of my test of how long it takes DXO Photolab 8.1 to apply Optical Corrections and DeepPRIME or DeepPRIME XD2s to various Sony RAW files on my 2021 M1 Max Mac Studio running macOS 14.7 with 32GB & 512GB. RAW files were hosted on, and DNG files were exported to, a 1TB 1500MB/s TB3 SSD.

DP XD2s performance was quite consistent with batches of 11 and 44 RAW files of sizes ranging from 24MP to 61MP. OTOH, DP performance varied unpredictably with MP count and batch size.

Test images were from DPR's Studio Scene comparison tool, one of each ISO from 100 to 102,400 (11 images) from each of four cameras (a7III, a7IV, a7RIII & 7RV). Batches of 44 from a given camera were made by duplicating the original 11 files three times. The combined group of 44 comprises 11 files from each camera.

I ran 44-file batches for the a7IV and a7RV because I suspected that the smaller 11-file batches were unrepresentative, and the variable results from the smaller batches seem to bear out this impression. I also set PhotoLab to use 5 cores, then 10 cores, and results were virtually the same.

On average, I get about 3.2 MP/s with DeepPRIME XD2s, scaling processing time linearly with MP count. DeepPRIME was much faster (around 9 MP/s) with the small 24MP batch, the large 33MP batch, and the large combined batch than with the large 61MP batch or the other small batches (around 5.3MP/s). I have no idea why this might be. I ran some of these batches again, but results didn't change.

DxO PL8.1 with macOS 14.7

TIME PER BATCH - TIME PER IMAGE - MEGAPIXELS PER SECOND

24MP (11 files)

• DeepPRIME = 0:32 (3s) (8MP/s)
• DeepPRIME XD2s = 1:24 (7.6s) (3.14MP/s)

33MP (11 files)

• DeepPRIME = 1:12 (6.5s) (5.04MP/s)
• DeepPRIME XD2s = 1:48 (9.8s) (3.36MP/s)

(44 files)

• DeepPRIME = 2:29 (3.4s) (9.74MP/s)
• DeepPRIME XD2s = 7:13 (9.8s) (3.36MP/s)

42MP (11 files)

• DeepPRIME = 1:27 (7.9s) (5.31MP/s)
• DeepPRIME XD2s = 2:24 (13.1s) (3.21MP/s)

61MP (11 files)

• DeepPRIME = 2:05 (11.4s) (5.34MP/s)
• DeepPRIME XD2s = 3:26 (18.1s) (3.26MP/s)

(44 files)

• DeepPRIME = 8:03 (11.0s) (5.56MP/s)
• DeepPRIME XD2s = 13:40 (18.6s) (3.27MP/s)

24+33+42+61MP (44 files)

• DeepPRIME = 3:16 (4.4s) (8.98MP/s)
• DeepPRIME XD2s = 8:30 (11.6s) (3.45MP/s)

BTW, looking over my past tests, I see that PL8 ran DP XD2s about as fast as PL6 ran DP XD, and PL8.1 is about 10% faster with DP and 20% faster with DP XD2s. Nice!

Very interesting.

Finally, I've been reading about performance improvements with macOS 15.x and would love to hear real-world reports, especially pertaining to Lightroom and PhotoLab.

Yes, I am interested in 15.x and Lightroom performance too.

 

[Jacques Cornell]

I replaced my 24-core GPU M1 Max Mac Studio running macOS 14.7.5 with a used 32-core version running macOS 15.4, looking for a little speed bump and more storage space at minimal cost. Both have 32GB RAM. Nothing dramatic, but here are the results.

Test files are as before, 11 files spanning ISO 100 to 102,400 from four Sony cameras (24MP a7III, 33MP a7IV, 42MP a7RIII and 61MP a7RV), for a total of 44 RAW files.

Lightroom Classic 14.2 - Denoise

• 24-core = 18:15
• 32-core = 14:47

DxO PhotoLab 8.3.0.39 - (optical corrections + DeepPRIME XD2s noise reduction)

24-core

• Auto = 9:12
• ANE = 8:42
• GPU = 10:29

32-core

• Auto = 8:27
• ANE = 8:59
• GPU = 7:56

 

[Jacques Cornell]

I replaced my 24-core GPU M1 Max Mac Studio running macOS 14.7.5 with a used 32-core version running macOS 15.4, looking for a little speed bump and more storage space at minimal cost. Both have 32GB RAM. Nothing dramatic, but here are the results.

Test files are as before, 11 files spanning ISO 100 to 102,400 from four Sony cameras (24MP a7III, 33MP a7IV, 42MP a7RIII and 61MP a7RV), for a total of 44 RAW files.

Lightroom Classic 14.2 - Denoise

• 24-core = 18:15
• 32-core = 14:47

DxO PhotoLab 8.3.0.39 - (optical corrections + DeepPRIME XD2s noise reduction)

24-core

• Auto = 9:12
• ANE = 8:42
• GPU = 10:29

32-core

• Auto = 8:27
• ANE = 8:59
• GPU = 7:56

FWIW, I ran the same files through DeepPRIME and the new DeepPRIME 3 (included in PureRAW5 and PhotoLab 8.5+) on my newer Mac Studio and got 3:13 and 2:57 respectively using "Auto". So, DP3 seems to be about 8% faster. I haven't tested the quality of DP3 yet.

 

[LukeP_Photo]

I appreciate you taking the time to share the information. I use DxO for noise reduction as well. I can't do an apples to apples comparison to what you did since I do not have Sony cameras. However, this may provide a useful data point. What will be interesting is to see if both Adobe and DxO can get their software to take full advantage of the Apple Neural Engines again without sacrificing image quality. For a while denoising in DxO was quite fast. The main thing you can see with my numbers is that because of my computer I can basically "brute force" my way into quicker times. I do a lot of high iso event coverage, and will start ramping up more video work soon. I was also trying to maximize my purchase for the future with this desktop. The time savings so far have been nice, and it's not just time saved with noise reduction. Export time are also very fast.

Mac Studio M3 Ultra, 80 core GPU, 512gb RAM running Sequoia 15.4.1

DxO PL 8.5.0

I ran 44 Nikon Z9 raw files using optical corrections and Deep Prime XD/XD2s with ISOs ranging from 8,000 to 12,800 ISO. Numbers below is total time for all 44 files in minutes: seconds

Auto: 6:58

Apple Neural Engine: 6:38

M3 GPU: 3:59

 

[Jacques Cornell]

I appreciate you taking the time to share the information. I use DxO for noise reduction as well. I can't do an apples to apples comparison to what you did since I do not have Sony cameras. However, this may provide a useful data point. What will be interesting is to see if both Adobe and DxO can get their software to take full advantage of the Apple Neural Engines again without sacrificing image quality.

With the exception of a brief hiatus that was rectified by an update, DxO has always leveraged ANE for DeepPRIME processing. A DxO rep told me as much. Adobe's history is the converse: it leveraged ANE for only a brief period, then stopped.

For a while denoising in DxO was quite fast.

I get about 5 seconds with DP3 and 15 seconds with XD2s on 61MP a7CR/a7RV RAWs on an M1 Max Mac Studio.

The main thing you can see with my numbers is that because of my computer I can basically "brute force" my way into quicker times. I do a lot of high iso event coverage, and will start ramping up more video work soon. I was also trying to maximize my purchase for the future with this desktop. The time savings so far have been nice, and it's not just time saved with noise reduction. Export time are also very fast.

Mac Studio M3 Ultra, 80 core GPU, 512gb RAM running Sequoia 15.4.1

DxO PL 8.5.0

I ran 44 Nikon Z9 raw files using optical corrections and Deep Prime XD/XD2s with ISOs ranging from 8,000 to 12,800 ISO. Numbers below is total time for all 44 files in minutes: seconds

Auto: 6:58

Apple Neural Engine: 6:38

M3 GPU: 3:59

Interesting. Your GPU time is about half that of my M1 Max with 32 GPU cores and 32GB RAM, which is about what I'd expect. But, your Auto & ANE times are only about 25% shorter despite the Ultra having dual ANEs. FWIW, my times are for 44 RAWs with an average size of 40MP (11 each of 24, 33, 42 and 61MP).

Since I can generally fall back to DP3, which runs 2.5x faster than DP XD2s, my older Studio is still plenty fast enough for my event workflow.

 

[Jacques Cornell]

Below are times for my M1 Pro MacBook Pro with 10 CPU cores, 14 GPU cores and 16GB RAM, running macOS 15.4.1 and DxO PhotoLab 8.5.1.

I replaced my 24-core GPU M1 Max Mac Studio running macOS 14.7.5 with a used 32-core version running macOS 15.4, looking for a little speed bump and more storage space at minimal cost. Both have 32GB RAM. Nothing dramatic, but here are the results.

Test files are as before, 11 files spanning ISO 100 to 102,400 from four Sony cameras (24MP a7III, 33MP a7IV, 42MP a7RIII and 61MP a7RV), for a total of 44 RAW files.

Lightroom Classic 14.2 - Denoise

• 14-core = 30:27

• 24-core = 18:15
• 32-core = 14:47

DxO PhotoLab 8.3.0.39 - (optical corrections + DeepPRIME XD2s noise reduction)

14-core

• Auto = 9:00

24-core

• Auto = 9:12
• ANE = 8:42
• GPU = 10:29

32-core

• Auto = 8:27
• ANE = 8:59
• GPU = 7:56

The modest MBP ran surprisingly close to the 24-GPU Studio on DeepPRIME XD2s, and only about 15% behind the 32-GPU Studio. It also built 1:1 previews in LRC as fast as the 24-GPU Studio. And, it ran DxO's less-demanding new DeepPRIME 3 in 1/3 the time needed for DP XD2s - about 4 seconds per image on average, making this unpretentious 3-year-old laptop plenty fast enough for high-volume event work. I'm starting to wonder whether I really need the Studio at all...

 

[Jacques Cornell]

TL;DR - 32 GPU cores are better for my workflow than 24 in the M1 Max Studio but not in the M1 Max MacBook Pro.

Background: The biggest bottlenecks in my high-volume high-ISO event photography workflow are time to import and build 1:1 previews in Lightroom Classic and then process noise reduction when exporting finished JPEGs from DxO PhotoLab. Over the past two years I've made several upgrades intended to shorten these times. What I've found is that upgrading from an M1 Mac mini and M1 MacBook Air yielded substantial improvements, upgrading from an M1 Max Mac Studio with 24 GPU cores to one with 32 GPU cores yielded a 19% improvement in noise reduction performance, and the move between MacBook Pros with those same chips brought no improvement in noise reduction or import+preview generation performance, but a 35% boost in LRC Denoise speed (which I don't need).

So, for my purposes, the move from 24 to 32 GPU cores in the Studio was helpful, but the same move in the MacBook Pro brings no benefit at all.







--
Event professional for 20+ years, travel & landscape enthusiast for 30+, stills-only.

Landscape & Wedding Photographer - Jacques Cornell

Jacques Cornell Photography - landscape, travel, portrait and art reproduction

jacquescornell.photography

Happening Photos - Event Photography in New York City

Event Photography in New York City and Around the World - conferences, sales meetings, fundraisers and more.

happening.photos

 

[Meh007]

Below are times for my M1 Pro MacBook Pro with 10 CPU cores, 14 GPU cores and 16GB RAM, running macOS 15.4.1 and DxO PhotoLab 8.5.1.

I replaced my 24-core GPU M1 Max Mac Studio running macOS 14.7.5 with a used 32-core version running macOS 15.4, looking for a little speed bump and more storage space at minimal cost. Both have 32GB RAM. Nothing dramatic, but here are the results.

Test files are as before, 11 files spanning ISO 100 to 102,400 from four Sony cameras (24MP a7III, 33MP a7IV, 42MP a7RIII and 61MP a7RV), for a total of 44 RAW files.

Lightroom Classic 14.2 - Denoise

• 14-core = 30:27

• 24-core = 18:15
• 32-core = 14:47

DxO PhotoLab 8.3.0.39 - (optical corrections + DeepPRIME XD2s noise reduction)

14-core

• Auto = 9:00

24-core

• Auto = 9:12
• ANE = 8:42
• GPU = 10:29

32-core

• Auto = 8:27
• ANE = 8:59
• GPU = 7:56

The modest MBP ran surprisingly close to the 24-GPU Studio on DeepPRIME XD2s, and only about 15% behind the 32-GPU Studio. It also built 1:1 previews in LRC as fast as the 24-GPU Studio. And, it ran DxO's less-demanding new DeepPRIME 3 in 1/3 the time needed for DP XD2s - about 4 seconds per image on average, making this unpretentious 3-year-old laptop plenty fast enough for high-volume event work. I'm starting to wonder whether I really need the Studio at all...

 

[Jacques Cornell]

Below are times for my M1 Pro MacBook Pro with 10 CPU cores, 14 GPU cores and 16GB RAM, running macOS 15.4.1 and DxO PhotoLab 8.5.1.

I replaced my 24-core GPU M1 Max Mac Studio running macOS 14.7.5 with a used 32-core version running macOS 15.4, looking for a little speed bump and more storage space at minimal cost. Both have 32GB RAM. Nothing dramatic, but here are the results.

Test files are as before, 11 files spanning ISO 100 to 102,400 from four Sony cameras (24MP a7III, 33MP a7IV, 42MP a7RIII and 61MP a7RV), for a total of 44 RAW files.

Lightroom Classic 14.2 - Denoise

• 14-core = 30:27

• 24-core = 18:15
• 32-core = 14:47

DxO PhotoLab 8.3.0.39 - (optical corrections + DeepPRIME XD2s noise reduction)

14-core

• Auto = 9:00

24-core

• Auto = 9:12
• ANE = 8:42
• GPU = 10:29

32-core

• Auto = 8:27
• ANE = 8:59
• GPU = 7:56

The modest MBP ran surprisingly close to the 24-GPU Studio on DeepPRIME XD2s, and only about 15% behind the 32-GPU Studio. It also built 1:1 previews in LRC as fast as the 24-GPU Studio. And, it ran DxO's less-demanding new DeepPRIME 3 in 1/3 the time needed for DP XD2s - about 4 seconds per image on average, making this unpretentious 3-year-old laptop plenty fast enough for high-volume event work. I'm starting to wonder whether I really need the Studio at all...

How many mpx are the images?

Please read the bold text above again.

 

[Jacques Cornell]

The base M4 is almost exactly as fast on DxO DeepPRIME 3 and XD2s as my M1 Max Mac Studio with 32 GPU cores. I just benchmarked with my standard test batch in PhotoLab 9.0.1 with macOS 15.7 on a 13" M4 MacBook Air with 24GB RAM and 512GB SSD.

 

[bakubo]

The base M4 is almost exactly as fast on DxO DeepPRIME 3 and XD2s as my M1 Max Mac Studio with 32 GPU cores. I just benchmarked with my standard test batch in PhotoLab 9.0.1 with macOS 15.7 on a 13" M4 MacBook Air with 24GB RAM and 512GB SSD.

That is pretty amazing. M1 Max 10/32/16 and M4 10/8/16 processing DxO NR at the same speed. I had assumed that the M1 Max 32 GPU cores would still blow away the M4 8 GPU cores.

 

[Jacques Cornell]

The base M4 is almost exactly as fast on DxO DeepPRIME 3 and XD2s as my M1 Max Mac Studio with 32 GPU cores. I just benchmarked with my standard test batch in PhotoLab 9.0.1 with macOS 15.7 on a 13" M4 MacBook Air with 24GB RAM and 512GB SSD.

That is pretty amazing. M1 Max 10/32/16 and M4 10/8/16 processing DxO NR at the same speed. I had assumed that the M1 Max 32 GPU cores would still blow away the M4 8 GPU cores.

M4's upgraded Neural Engine seems to be a big factor here, as DeepPRIME makes good use of it. I'll be interested to test LRC preview building and Denoise next, but have a big job over the next few days, so it'll have to wait.

BTW, M4 MBAs get two more GPU cores with the upgrade to 24GB RAM, so mine is 10/10/16.

 

[ChuckB]

Regrettably, my M1 Macbook Air suffers from Adobe's inability to use the neural engine with Lightroom. My 'upgrade' turned out to be Lenovo Legion 5i laptop with an Nvidia 5060 GPU. Denoise in 10 seconds or so.

 

[bakubo]

The base M4 is almost exactly as fast on DxO DeepPRIME 3 and XD2s as my M1 Max Mac Studio with 32 GPU cores. I just benchmarked with my standard test batch in PhotoLab 9.0.1 with macOS 15.7 on a 13" M4 MacBook Air with 24GB RAM and 512GB SSD.

That is pretty amazing. M1 Max 10/32/16 and M4 10/8/16 processing DxO NR at the same speed. I had assumed that the M1 Max 32 GPU cores would still blow away the M4 8 GPU cores.

M4's upgraded Neural Engine seems to be a big factor here, as DeepPRIME makes good use of it. I'll be interested to test LRC preview building and Denoise next, but have a big job over the next few days, so it'll have to wait.

BTW, M4 MBAs get two more GPU cores with the upgrade to 24GB RAM, so mine is 10/10/16.

Ahh, okay, M1 Max 10/32/16 vs. M4 10/10/16. It is still surprising, but I think you are right that the Neural Engine is being used a lot along with the GPU. Both have a 16 core NE, but the M4 NE is faster.

 

[Jacques Cornell]

Regrettably, my M1 Macbook Air suffers from Adobe's inability to use the neural engine with Lightroom. My 'upgrade' turned out to be Lenovo Legion 5i laptop with an Nvidia 5060 GPU. Denoise in 10 seconds or so.

How many megapixels? Performance scales linearly with pixel count. FWIW, my M4 MBA does DP3 on files with an average size of 40MP in 3.8 seconds. DP XD2s takes 12.2 seconds. My M1 Max Mac Studio, M1 Pro MacBook Pro, and M4 MacBook Air all perform the same on this test. Switching to DxO PhotoLab might have been a cheaper upgrade with similar results.

 

[ChuckB]

Regrettably, my M1 Macbook Air suffers from Adobe's inability to use the neural engine with Lightroom. My 'upgrade' turned out to be Lenovo Legion 5i laptop with an Nvidia 5060 GPU. Denoise in 10 seconds or so.

How many megapixels? Performance scales linearly with pixel count. FWIW, my M4 MBA does DP3 on files with an average size of 40MP in 3.8 seconds. DP XD2s takes 12.2 seconds. My M1 Max Mac Studio, M1 Pro MacBook Pro, and M4 MacBook Air all perform the same on this test. Switching to DxO PhotoLab might have been a cheaper upgrade with similar results.

Small ones - 20MB. Reason I went with Windows was I can not get used to the Mac OS. I don't know where things are, etc etc. Upgrading to another Mac would have been just as expensive, if not more so, and I needed a Windows machine for all the other stuff I do anyway, so I bagged the Mac. M1 for sale cheap. The Lenovo is plenty fast for me, using small files and not a lot of them. If LR is slower than DP3 the difference is not important and I don't want to learn a new program.

 

[Jacques Cornell]

Regrettably, my M1 Macbook Air suffers from Adobe's inability to use the neural engine with Lightroom. My 'upgrade' turned out to be Lenovo Legion 5i laptop with an Nvidia 5060 GPU. Denoise in 10 seconds or so.

How many megapixels? Performance scales linearly with pixel count. FWIW, my M4 MBA does DP3 on files with an average size of 40MP in 3.8 seconds. DP XD2s takes 12.2 seconds. My M1 Max Mac Studio, M1 Pro MacBook Pro, and M4 MacBook Air all perform the same on this test. Switching to DxO PhotoLab might have been a cheaper upgrade with similar results.

Small ones - 20MB. Reason I went with Windows was I can not get used to the Mac OS. I don't know where things are, etc etc. Upgrading to another Mac would have been just as expensive, if not more so, and I needed a Windows machine for all the other stuff I do anyway, so I bagged the Mac. M1 for sale cheap. The Lenovo is plenty fast for me, using small files and not a lot of them. If LR is slower than DP3 the difference is not important and I don't want to learn a new program.

Since your files are half the size of my test files, you can cut my times above in half - 2s for DP3 and 6s for DP XD2s. An old M1 MacBook Air would be about half as fast, so back to my original numbers. For 20MP files on an M1 MBA, my guess is DP3 would be substantially faster than Denoise on your new Lenovo.