Topaz AI Software – Does an eGPU make a difference?
HOME
/
ARTICLES
/
GEAR
/
TOPAZ AI SOFTWARE – DOES AN EGPU MAKE A DIFFERENCE?
An external GPU (eGPU) can boost the graphics power of your laptop. We investigate how much faster the Topaz AI photo processing products can be.
DEC 10, 2019
Topaz Labs have recently released some Artificial Intelligence driven image editing tools that can produce some very good results in the areas of Noise Reduction, Sharpening and Image Upsizing. In order to process an image using AI models the software requires some serious processing power to run. When a powerful enough graphics card is present in the Computer, the AI processing can be offloaded to the GPU (specifically an eGPU in this case) to reduce processing times.
Unfortunately for most laptop users, the GPU inside a laptop is usually quite weak when compared to what can be found inside a desktop computer. With the introduction of Thunderbolt 3 connections on modern laptops it is now possible to connect an external graphics card (eGPU). This allows for much stronger GPU performance.
This article will discuss the use of an eGPU to potentially speed up the Topaz Labs AI powered software.
Check Out Topaz Labs AI Products
Artificial Intelligence driven noise reduction, sharpening and resize toolsGo!
I recently took possession of an eGPU in the way of a Razer Core X enclosure and an ASUS ROG Vega 64 Strix graphics card so will be testing those with my 2018 MacBook Pro that I use for my photo editing.
Click below if you want to jump straight to a particular application
DeNoise AI
Sharpen AI Results
Gigapixel AI Results
Conclusion
Test Setup
The Computer
As I am testing my real-world experience here I performed all tests on the hardware that I use in my daily photo processing workflow:
MacBook Pro 2018 with 6-core i7 @ 2.6GHz
Radeon Pro 560x dedicated GPU
16GB RAM
7200rpm External HDD connected via Thunderbolt 3 dock. (All Photo files used were located on this drive)
Dell 27″ 2560×1440 Monitor connected to MacBook Pro via Caldigit TB3+ Dock.
The eGPU setup includes the above, plus:
Razer Core X with Vega 64 Graphics Card connected to the MacBook Pro via Thunderbolt 3. The Screen is connected directly to the graphics card in the eGPU setup.
All of the image processing was done on the attached external display.
The Software
All RAW images chosen for the test were opened in Capture One Pro and adjustments applied before being sent as a 16-bit TIFF file to Topaz Studio 2 where the tests were then performed. I kept the workflow exactly as I would when processing images normally.
The timer was started for each test as soon as either the ‘Update Preview’ or ‘Process Image’ button was pressed and stopped when the progress window was complete.
Interestingly, the Topaz software didn’t automatically select the Radeon 560x when doing these tests without the eGPU attached, instead opting to use the CPU. This turned out to be the correct choice as forcing the MacBook Pro to use the internal 560x slowed down the process by over 200% when compared to using the CPU to process images.
DeNoise AI
Topaz Denoise AI eGPU test.
(Check out DeNoise AI)
To test the Topaz Labs noise reduction software I chose 10 images taken between ISO1600 and ISO6400. I then set the sliders in DeNoise AI to the optimal setting for each image (I rarely rely on the default suggestions). For the preview timings I set the zoom level to 100% and made sure I was looking at the same portion of the image for both the bare MacBook Pro and the MacBook Pro + eGPU options. This ensured that each GPU was doing the same amount of work between the two tests.
Average time taken to update the preview image:
MacBook Pro = 4.85 seconds
MacBook Pro + Vega 64 = 4.55 seconds
Average time taken to fully process the image:
MacBook Pro = 41.57 seconds
MacBook Pro + Vega 64 = 29.91 seconds
Results
From the results above you can see that the time taken to render the preview image does not significantly change with the addition of an eGPU. The Vega 64 eGPU rendered previews around 6% faster than using just the MacBook Pro.
A much larger result can be seen when processing the full image, with an average of about 11 seconds being saved per image (28% improvement).
Sharpen AI
Topaz Sharpen AI eGPU test.
(Check Out Sharpen AI)
For the Sharpen AI test I chose a selection of 10 images containing varying levels of detail taken at various ISO settings. If this noise level of the photo was high enough that I would normally run it through noise reduction, I ran it through DeNoise AI first (as is suggested by Topaz). All of the images I chose were acceptably sharp when photographed (ie, I didn’t test the software on images that I would not have otherwise processed).
Just like in the DeNoise Test, I set the preview zoom level to 100% and made sure I was looking at the same portion of the image for both the internal and external GPU tests. I tested all three modes of sharpening (Sharpen, Stabilise and Focus).
Sharpen Mode
Average time taken to update the preview image:
MacBook Pro = 4.7 seconds
MacBook Pro + Vega 64 = 4.7 seconds
Average time taken to fully process the image:
MacBook Pro = 42.1 seconds
MacBook Pro + Vega 64 = 33.5 seconds
Results
In Sharpen mode the preview update time was exactly the same between the bare MacBook Pro and the MacBook + eGPU.
When processing the image proper the eGPU setup was on average 20% faster than the bare MacBook Pro.
Stabilise Mode
Average time taken to update the preview image:
MacBook Pro = 14.1 seconds
MacBook Pro + Vega 64 = 14.7 seconds
Average time taken to fully process the image:
MacBook Pro = 145.5 seconds
MacBook Pro + Vega 64 = 134.4 seconds
Results
Preview updates with the eGPU actually took slightly longer (increasing by 4%) than with the bare MacBook Pro. This was quite a surprise.
Actually processing the images was 7% faster with the eGPU than with the MacBook Pro alone.
The worse preview results and only slightly better processing times surprised me as the Stabilise mode is more computationally intensive than the Sharpen mode tested above.
Focus Mode
Average time taken to update the preview image:
MacBook Pro = 14.8 seconds
MacBook Pro + Vega 64 = 14.8 seconds
Average time taken to fully process the image:
MacBook Pro = 133.5 seconds
MacBook Pro + Vega 64 = 127.2 seconds
Results
The time to update the preview with both the base MacBook Pro and MacBook + eGPU was exactly the same.
Full processing times were very slightly improved when using the eGPU with a 5% improvement in speed.
As with the ‘Stabilise’ mode tests, these results are surprisingly close.
Gigapixel AI
Topaz Gigapixel AI eGPU test.
(Check out Gigapixel AI)
Test 1:
I print most of my ‘keepers’ at 13×19″. Epson printers work best when fed files at 360ppi so I decided to upsize my test images to this size (4680×6840 pixels). For this test I chose 10 images that had been cropped to varying sizes.
Average time taken to update the preview image:
MacBook Pro = 11 seconds
MacBook Pro + Vega 64 = 6.9 seconds
Average time taken to fully process the image:
MacBook Pro = 452 seconds
MacBook Pro + Vega 64 = 202 seconds
Results
Preview times with the eGPU were significantly (38%) faster at 7 seconds vs 11 with just the MacBook.
When processing images there was a huge difference between the base MacBook Pro vs the MacBook + eGPU. The eGPU took just 45% of the time to process an image when compared to the base MacBook Pro!
Test 2:
As my normal print size doesn’t really push the limits of the Gigapixel AI Software, I took the same set of images and set them all to the maximum allowed enlargement size and repeated the test.
Average time taken to update the preview image:
MacBook Pro = 1.5 seconds
MacBook Pro + Vega 64 = 1.5 seconds
Average time taken to fully process the image:
MacBook Pro = 426 seconds
MacBook Pro + Vega 64 = 249 seconds
Results
When dealing with such huge enlargements the preview window is showing so few pixels that the difference between the MacBook and the eGPU is negligible.
When processing the full image the difference is significant in favour of the eGPU setup. When engaging the eGPU, processing times are just 60% of those with just the MacBook Pro.
Observations
There were a couple of things that were noticeable throughout the testing that doesn’t fit into any of the results sections:
Firstly, when processing images using the MacBook’s CPU, (ie. with the eGPU not attached), the CPU temperatures quickly climbed to their maximum levels and the fans were very loud at full speed. With the eGPU doing the processing the MacBook fans did not come on at all and the fans in the eGPU enclosure were practically silent.
The time taken to process each image was very similar for images of the same size. It seems that the content of the image and settings applied to each image matter little in processing times – the resolution of the input image is what affects the processing time.
Processing times were not affected by having the laptop screen open vs closed.
Conclusion
From the tests above it is clear that by harnessing the power of an eGPU on a 2018 MacBook Pro there are…. mixed results when using Topaz Labs AI powered software.
Gigapixel AI: A very large improvement with processing over twice as fast.
DeNoise AI: A modest 25% reduction in processing times
Sharpen AI: Between 5 and 10% improvement for the Stabilise and Focus modes and 20% for the Sharpen mode.
One very positive advantage though is that the computer will run MUCH cooler and quieter when the eGPU does the work instead of the CPU. A cooler CPU is always a good thing when it comes to extending the life of a PC as extreme heat for extended periods of time can damage critical components. Not only that – but it is much more pleasant to work in an environment without massive amounts of laptop fan noise.
Whether the performance gains are worth the cost of an eGPU is hard to say. If you do any kind of video, rendering or gaming on your Laptop then an eGPU will probably be a no-brainer. However if you are just using it to process images with Topaz Software… the performance gains are fairly small with everything except Gigapixel AI.
If you have any questions, comments or want me to try any other tests please don’t hesitate to leave a comment below, or get in touch.