Topaz GPU benchmarking at DP Review

A small group at DPR is benchmarking the time to process a test image using primarily Topaz Sharpen AI.

The steps for testing are as follows:

  • Go here to retrieve the test image
  • Download the car console shot found under sample raw images, nikon_d850_24.nef
  • Open the file directly in the Topaz software. Make sure Allowed Graphics Memory Consumption is set to High if your video card will support it. You may need to close and reopen the software after making this change, I don’t have documentation on this.
  • Manually select the Stabilize Mode.
  • Manually set your Sharpness slider to 22 and Noise Suppression to 24 (I’ve found some variability with Auto settings)
  • Choose dng for output, source folder
  • Start the timer when you hit Save and stop it when the Processing indicator disappears.
  • Report the result along with system specs (CPU and GPU at a minimum).

Most of the results are for Sharpen AI although there has been at least one test using Denoise. The settings used for Denoise were as follows:

  • Remove Noise 20
  • Enhance Sharpness 20
  • Recover Original Detail 10
  • Color Noise Reduction 0.10

The page with the first tabulated results is here

Please contribute is you wish, the more results the better.

Your link to the results page is for Nvidia GPU using Denoise. You asked for sharpen AI and no mention of the GPU brand. This must be the wrong page.

It’s the right page. The thread referenced started out as a general discussion of GPU performance in Topaz apps. Even though Denoise is referenced in the thread title, Sharpen soon entered the conversation and the benchmarking is designed for either app with settings given to be entered manually. It just so happens that almost all of the participants (except one or two) used Sharpen for their testing and so the Sharpen results were placed in table form for reference. There isn’t enough Denoise data to bother with yet.

With regard to GPU brand, every card tested was listed by name (but not vendor name, e.g. EVGA) and model number so I’m not sure what you mean by this. So far, it’s an all Nvidia show, no AMD/Radeon owners have replied—maybe because Nvidia was mentioned in the thread title, but I’m sure results from the other camp would be welcome.

I find very little sharpen replies but I posted my results for my Sapphire Radeon RX5600XT GPU. I used Sharpen and Denoise and the Nikon 850_24 original file. In both cases I saved as a DNG file.
My results were 1:55 (clock) 1:51 Sharpen app timer. Used your settings.
0:31 (clock) 0:27 Denoise app timer. Used auto and Denoise mode

Hello Everyone,

I have been reading through this thread after building a new computer and would like to try and add some insight into how much the GPU affects processing times. The common thread I have seen is that in order for the benchmark results to be accurate, it needs to be run on the same computer with different cards.

My current build:
AMD R9 3960x w/ large air cooler
MSI sTRX4 Creator Motherboard
128gb Team Force Vulcan DDR4 with XMP 2 enabled
Samsung 980 NVME drive
EVGA 750ti (yes it is old, there is a shortage and I couldn’t snag a new card when I built the PC)

Since I am guying a new GPU, I thought it may be interesting to compare results on the exact same system with only the GPU changing. That being said, I have not been able to download the photo. When I click download image, it just gives me a preview. Saving it from there gives me a small jpeg.

What am I doing wrong?

Is this the link you’re using?

https://img.photographyblog.com/reviews/nikon_d850/photos/nikon_d850_24.nef

If you are able to download the raw (nef) file, note that the parameters (terminology) in Sharpen have changed since that DPR thread was active. So Stabilize is now Motion Blur (or so it appears). I don’t know if the slider values are 1:1 vs the older versions either, so maybe try to replicate the settings as best you can and note all of your settings. This way maybe we can establish a new baseline for the newer software.

That was not the link I was using, i was able to download this time.

Specs: Application & Version: Topaz Sharpen AI Version 3.0.2

Operating System: Windows 10 Version 2009

Graphics Hardware: NVIDIA GeForce GTX 750 Ti/PCIe/SSE2

OpenGL Driver: 3.3.0 NVIDIA 466.27

CPU RAM: 130967 MB

Video RAM: 2048 MB

Preview Limit: 2954 Pixels

I am running the game ready driver

Preview: 9 seconds 4% CPU usage and 1% GPU
Saving/processing: Remove/Blur: 22 Suppress Noise 24
11’50" GPU 100%, CPU 9% 3.6gb of system RAM used, 64% of Vram used

Processing using CPU: 1’23" 40% utilization, temps around 80 degrees, 18 gb of system RAM utilized. CPU clock 4.15 ghz on all 24 cores

GPU: GTX 750Ti 2GB GDDR5, memory bit width 128 bit base clock 1020mhx, boost 1085mhz, 640 CUDA cores. GPU ran at about 47 degrees with fan at 100%, it throttled with auto settings since the fan wouldn’t go above 40%. HW Monitor had the GPU clock at 1392mhz for most of the processing.

Update:

I got my hands on a Gigabyte 6700xt. I have the card overclocked to 2856mhz for the GPU clock and 2150 for memory. Power limit is at 115%. Processing times for the same image at the same settings is 12 seconds. and gpu usage was 98%.

On a side note, denoise would not use the GPU more than 2% and defaulted to using the processor.

I tried this on test on my MacBook Pro M1 with 16 GB of RAM. It took 15 seconds.