DeNoise AI v2.4.1 FAR slower than 2.3.6

Windows 10 Pro, I7, O/C 5820, 32gb memory, Samsung 860 SSD image storage, AMD 5700 XT 8GB GPU.

2.4.1 is EXTREMELY slow processing an image. So was 2.4.0, so going back to that is pointless.

In case you only read this far before replying - IT’S HAD CALIBRATE RUN, REPEATEDLY!

For reference, when I was running 2.3.6 it randomly, after some amount of use would stop updating the previews and the only fix was a completely removal of all topaz and reinstall. BUT WHEN IT RAN, it wasn’t wretchedly slow:
Start Denoise AI 2.3.6 STANDALONE. Go into prefs, have GPU on, usage HIGH, press Calibrate, open a small Nikon D500 (20 megapixel) image. Let it calibrate as it created previews (it got re-installed so often I got good at this). Change zoom and let it recreate previews. Process a Nikon .NEF using Denoise, generating a .dng output. Processing time from 0 – 100% approximately 12 seconds. Time from when it reached 100% to when it returned to the preview screen approximately 15-18 seconds. Total time from hitting the button to back in preview, around 33 seconds.

Do EXACTLY the same thing with Denoise 2.4.1. GPU on, usage HIGH, press Calibrate, previews, zoom, previews, process .NEF. EIGHTY SECONDS to get from 0 to 100% and STILL another 15 seconds to get back to the preview screen.

Back in Preview, go BACK into prefs and press Calibrate. EXIT from Denoise in case the calibrate doesn’t actually work in previews even though it SAYS IT’S CALIBRATING. Restart Denoise, repeat process – previews, process .NEF to .dng output. ABSOLUTELY NO CHANGE. Still takes 90 seconds to process a small image. It’s now been “CALIBRATED” IN Denoise and after a restart.

Repeat 8 or 10 more times, trying to calibrate IN Denoise and by restarting. NOTHING changes.

This morning I had a friend who uses the same software do a comparison with his far less capable system. Using Denoise AI 2.4.1 on his system - an i3 (yes, that’s a 3) processor, 16GB of memory and an AMD 5500XT, processing an image from a Nikon D810 (36 megapixel, so LARGER than what I’m processing). With Denoise running standalone he loaded a Nikon .NEF file, selected Denoise, and processed, saving as a .dng file. We both timed the process. Processing time from 0 - 100% was around 14 seconds and time from hitting 100% to displaying back on the preview screen was 11-12, so the WHOLE process cycle was 25 seconds. We did this several times, restarting in between and it was consistent. A THIRD FASTER than the 30-33 seconds I got with 2.3.6 and astronomically faster than I’m getting with 2.4.1.

I restarted Denoise on my system, opened the preferences and TURNED OFF the graphics processor, and EXITED from Denoise. I restarted it standalone with the GPU off, Vino grayed out, NO CALIBRATION (if I Calibrate it turns the GPU back on - doesn’t WORK, but it turns it on), opened an image, and processed the image. This time it took approximately 65 seconds (down from 80) to process from 0 - 100% and approximately FIVE seconds one it reached 100% before it returned to the preview screen.

It appears to me that Denoise 2.4.1 is NOT making use of my AMD 5700XT GPU (which has significantly sped up other processing), and that “Saving” the OUTPUT is SIGNIFICANTLY faster with the GPU turned off…

ALL my processing was done on a newly rebooted system, with NO other significant processing happening. Nothing from Adobe or anything else to compete except tools like Outlook at a normal browser. ALL images are stored on a Samsung 860 SSD. To VERIFY there was no problem with the storage, I copied a subset of images to a SEPARATE SSD on a DIFFERENT I/O controller and reran the tests. NO DIFFERENCE. Total processing in 2.4.1 is extremely slow, and the time from 100% back to the preview screen is equally slow.

I would suspect a problem with Denoise recognizing my GPU, but it DOESN’T have that problem with 2.3.4 or 2.3.6, so I can’t imagine that 2.4.? would.

Calibration not working? I can only THINK of 2 ways to do the calibration - inside Denoise using the previews, or by selecting Calibrate, shutting DOWN Denoise and restarting it. Anybody know of some other method? Something else? Cosmic rays? Anybody have any useful ideas?

Please raise a Support Request at the main website as they may be able to recommend how to speed up processing. The minimum requirements for AMD is Radeon 5870 or greater so maybe there are tuning settings to help:

It’s been in process since about 60 seconds after putting 2.4.0 on the box. At least, so far, 2.4 hasn’t stopped updating. It’s just a total dog on this system.

Well, whatever the problem is, is apparently going to STAY a problem…
2.3.6, from memory, ran a noise reduction ins around 30 seconds. 2.4.0 - 2.4.2 take around 80-90. Support says “We’ve changed nothing that would cause this.”

At this point, with 2.4.?, it processes faster with the GPU turned off. It’s still pretty dreadful in the 68 or so second range, but still BETTER than with the GPU on and (supposedly) calibrated.

So version X takes N seconds to run. Version X+1 (or X that actually got released) takes N*3 seconds. But nothing is different…

Unfortunately, what I was told was “That 80+ second time to process is NOT outside our acceptable time, so there’s no problem.”

So for the sake of comparison

Nikon D700 69MB (uncompressed) 48 bit NEF raw image
Denoise AI 2.4.2 run freestanding on Denoise AI Auto settings, saving to DNG file.

Times from pressing Save to return of preview image.

Including Calibration: 22.0 secs
Open Vino on: 15.1 secs
Discrete GPU - Medium memory: 13.25 secs

Windows 10 Pro 64 bit version 20H2, build 19042.746
Intel i7-10700 Gen 10 CPU @ 2.90GHz, 32GB Memory
GeForce GTX 1660 6GB Graphics card, v461.40 driver
NvMe SSD’s.

And?

Unfortunately, on mine, with the GPU off its about 68 seconds. STILL faster than the total dog at 80 seconds with the GPU on and calibrated but a WHOLE lot slower than 2.4.0.b3, which does it in about 30-33 seconds.
And for the record I’ve got a friend with an ancient i3 with an AMD 5500XT and it’ll do a D810 image in the same 22 seconds as your 10700. All of which is great, but not relevant.

I need MY system to run 2.4.? as fast as it runs the 2.4.0 beta.

This whole thing is making my head hurt…

I HAVE a version of Topaz Denoise that WORKS… I’ve been told, BECAME the 2.4.0 that hit the street. THAT VERSION DOES NOT WORK. It was replaced with 2.4.1 and 2.4.2. They BOTH WORK JUST AS WELL AS 2.4.0. As in they SUCK.

Support says “We din’t change NOTHIN’ between b3 and now”. A works. B doesn’t. In the 30+ years I’ve been a software engineer and database designer, when that happens, the FIRST question is “What changed”… “We din’t change nothin’ isn’t an answer.”

BUT, just for giggles… Took the system and RAN 2.4.2 standalone.
Went into prefs and turned OFF the GPU.
I WANTED to turn on OpenVINO but I can’t. It’s greyed out. So, I clicked Calibrate and NOW I CAN select OpenVINO. Ran the image… 25 SECONDS… EXACTLY ONE TIME. Prefs are now BACK to using the GPU and the next time is 80+ seconds. It’s absolutely repeatable. Change the prefs, hit Calibrate, run the image 24-25 seconds. ONCE…

Anybody have a way to FORCE this thing to let me KEEP OpenVINO turned on?