# Topaz Video AI Performance Alpha for Windows Nvidia & AMD GPUs **URL:** \ **Category:** Video AI Beta\ **Created:** [March 2, 2023, 10:29pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961 "2023-03-02T22:29:48Z")\ **Posts on this page:** 20\ **Page:** 1
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 2, 2023, 10:29pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/1 "2023-03-02T22:29:48Z")
Hi All, Here is an early version of app that should improve performance for both Nvidia and AMD GPUs. Please let me know if the performance is slower. Download: [Win](https://downloads.topazlabs.com/deploy/TopazVideoAIAlpha/3.1.8.2.a/TopazVideoAIAlpha-3.1.8.2.a.msi) Low power mode has been removed for windows, this will allow more filters to be chained together and running multiple processes. **Known Issues** : - Proteus Auto will not work, use Proteus with manual settings - Models cannot be run on CPU - Models cannot be run on intel GPU Thank you for testing PS: Since this is an alpha version, it will not interfere with Beta or Release versions on the machine. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 2, 2023, 10:31pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/2 "2023-03-02T22:31:04Z")
---
**Author:** ![TPX](https://community.topazlabs.com/user_avatar/community.topazlabs.com/tpx/32/194727_2.png) [@TPX](https://community.topazlabs.com/u/TPX)\ **Post date:** [March 2, 2023, 11:19pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/3 "2023-03-02T23:19:51Z")
Fist short test. What the … … processed in 18 seconds (10 sec Preview) PAL to 1080p, 33,5 fps, Proteus. fastest b4 this version was 30 sec. 35 fps Artemis low q. Theia does not download (update below), Gaia (HQ) does not load and not download. Gaia GC, 13.4 fps. —\> after running this model Gaia (HQ) did download and work too, and Theia did too. Themis + Proteus = error. Themis + Artemis = error. Stabilisation + Artemis = error. end of short test. Win 11, 7820X, Quadro RTX 5000. The GPU and CPU are still ok, no overheat, both have still room for more performance, but the speed is nice. ---
**Author:** ![alanhusband-677420](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/a/7ba0ec/32.png) [@alanhusband-677420](https://community.topazlabs.com/u/alanhusband-677420)\ **Post date:** [March 2, 2023, 11:46pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/4 "2023-03-02T23:46:00Z")
On the GTX 1060 performance is consistent with last weeks alpha version (~2.4fps on a Proteus 720p to 1080p upscale). ---
**Author:** ![cleuton.4k](https://community.topazlabs.com/user_avatar/community.topazlabs.com/cleuton.4k/32/84448_2.png) [@cleuton.4k](https://community.topazlabs.com/u/cleuton.4k)\ **Post date:** [March 3, 2023, 12:14am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/5 "2023-03-03T00:14:31Z")
gtx 1050 ti 6% slower in all tests as gpu usage is not at 100%. ![171](https://community-cdn.topazlabs.com/original/3X/6/f/6f805d4719701d8e2fe3eb7bbe97207d52c44592.png) ---
**Author:** ![alanhusband-677420](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/a/7ba0ec/32.png) [@alanhusband-677420](https://community.topazlabs.com/u/alanhusband-677420)\ **Post date:** [March 3, 2023, 1:20am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/6 "2023-03-03T01:20:48Z")
My suggestion would be rather than remove the low power mode option completely, instead replace it with a “MAX Power Mode” (that is off by default) for those who just want to run a single filter at 100% gpu usage. Then make sure there is a prominent note by this option that turning it on could introduce instability when running some process combinations. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:29am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/7 "2023-03-03T01:29:11Z")
Might just link the boost to max processes. If it is set to 1, then the boost. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:29am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/8 "2023-03-03T01:29:43Z")
Compared to the release version or previous alpha? ---
**Author:** ![david.123](https://community.topazlabs.com/user_avatar/community.topazlabs.com/david.123/32/163156_2.png) [@david.123](https://community.topazlabs.com/u/david.123)\ **Post date:** [March 3, 2023, 2:06am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/9 "2023-03-03T02:06:34Z")
Hi Suraj, any idea on when you will integrate FFMpeg v6? There should be some performance to be found with that. ---
**Author:** ![alanhusband-677420](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/a/7ba0ec/32.png) [@alanhusband-677420](https://community.topazlabs.com/u/alanhusband-677420)\ **Post date:** [March 3, 2023, 3:50am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/10 "2023-03-03T03:50:04Z")
Just noticed the colour correction option. Is that working in this alpha? If so, what does it do? ---
**Author:** ![cleuton.4k](https://community.topazlabs.com/user_avatar/community.topazlabs.com/cleuton.4k/32/84448_2.png) [@cleuton.4k](https://community.topazlabs.com/u/cleuton.4k)\ **Post date:** [March 3, 2023, 6:04am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/11 "2023-03-03T06:04:39Z")
compared the two, because for me 3.1.7 and 3.1.7.2 alpha has the same performance. note: since version 3.1.2 videos in 720p had a performance drop of 4% compared to version 3.1.1 (other resolutions like 480p have had a performance increase since then). I believe it’s because of this change mentioned in the changelog "Significant performance improvement for GTX 16xx series GPUs compared to 3.x.x ". ---
**Author:** ![TPX](https://community.topazlabs.com/user_avatar/community.topazlabs.com/tpx/32/194727_2.png) [@TPX](https://community.topazlabs.com/u/TPX)\ **Post date:** [March 3, 2023, 9:07am UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/12 "2023-03-03T09:07:50Z")
So, the AMD system with 3960X & Radeon Pro W6800 does Proteus in 19 - 24 sec (10 sec Preview) with 27 - 29 fps. We did start at Jan 28 with 58 sec for proteus, so performance has been more than doubled since then. 👍 nice job. (but that speedup is only with small videos.) After first test run export is faster (seems like somthing did get cached) The 24 Core TR is still bored but thats ok, the intel 8 Core is almost maxed out but not overloaded. Both systems can still be used while exporting with TVAI. * * * * * * AMD Sys test: 24 core 3060X & W6800. * * * 10 sec preview. * * * 1080p denoise Artemis Low → 3.1.7 = 46 sec Artemis Low → 3.1.8.2 = 35 sec 1080p denoise - RTX 5000 & 7820X Artemis Low → 3.1.7 = 31 sec Artemis Low → 3.1.8.2 = 32 sec * * * 1080p denoise Proteus → 3.1.7 = 49 sec. Proteus → 3.1.8.2 = 37 sec. * * * 1080p → 4K Proteus → 3.1.7 = 1 Min 18 sec. Proteus → 3.1.8.2 = 1 Min 5 sec. * * * 1080p → 4K Artemis low → 3.1.7 = 1 Min 17 sec. Artemis low → 3.1.8.2 = 1 Min 1080p → 4K - RTX 5000 & 7820X Artemis low → 3.1.7 = 1 Min 5 sec. Artemis low → 3.1.8.2 = 57 sec. * * * 1080p → 4K Theia → 3.1.7 = 1 min 38 sec. Theia → 3.1.8.2 = 1 min 18 sec. * * * The bigger the video the lower the speedup. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:37pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/13 "2023-03-03T13:37:04Z")
Make sure memory setting is maxed out. I always use 1080p input for all my testing. The results in almost all cases are around 2x faster. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:38pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/14 "2023-03-03T13:38:18Z")
Should be in a couple of weeks. We generally update from ffmpeg main branch every month, so most of the 6.0 stuff should already be in there. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:39pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/15 "2023-03-03T13:39:34Z")
That is just for performance testing. It’s always on for color correction models and will not be an option during release. ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 1:40pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/16 "2023-03-03T13:40:06Z")
I see the stabilization issue. Will look into it. ---
**Author:** ![TPX](https://community.topazlabs.com/user_avatar/community.topazlabs.com/tpx/32/194727_2.png) [@TPX](https://community.topazlabs.com/u/TPX)\ **Post date:** [March 3, 2023, 2:03pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/17 "2023-03-03T14:03:22Z")
I did test at 50% and 100% and it didnt make any difference. Do you test it with ui or without? Whats the difference when we run the software? ---
**Author:** ![suraj](https://community.topazlabs.com/letter_avatar_proxy/v4/letter/s/ecccb3/32.png) [@suraj](https://community.topazlabs.com/u/suraj)\ **Post date:** [March 3, 2023, 4:16pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/18 "2023-03-03T16:16:39Z")
With GUI, could be because of Low Power Mode vs no low power mode ---
**Author:** ![TPX](https://community.topazlabs.com/user_avatar/community.topazlabs.com/tpx/32/194727_2.png) [@TPX](https://community.topazlabs.com/u/TPX)\ **Post date:** [March 3, 2023, 5:31pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/19 "2023-03-03T17:31:56Z")
Could the low power mode stick at some point and stay forever until i switch something in the reg? ---
**Author:** ![mikmod1](https://community.topazlabs.com/user_avatar/community.topazlabs.com/mikmod1/32/239469_2.png) [@mikmod1](https://community.topazlabs.com/u/mikmod1)\ **Post date:** [March 3, 2023, 6:02pm UTC](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961/20 "2023-03-03T18:02:08Z")
With 360p → 720p (that destroyed D&D cartoon FTW!) I see quite an increase, from around ~34fps in latest v3.1.7 stable, to ~45fps in this alpha. That’s on RTX 3090, 100% memory set, on i9 7940X CPU (fairly _old_ one) and 64GB of RAM. Preview parameters: ffmpeg “-hide\_banner” “-nostdin” “-y” “-nostats” “-t” “15” “-ss” “103.736” “-i” “W:/Video/Dungeons & Dragons Episode 19 The Last Illusion.mp4” “-flush\_packets” “1” “-sws\_flags” “spline+accurate\_rnd+full\_chroma\_int” “-color\_trc” “2” “-colorspace” “2” “-color\_primaries” “2” “-filter\_complex” “tvai\_fi=model=chr-2:slowmo=1:rdt=0.01:fps=60:device=0:vram=1:instances=0,tvai\_up=model=prob-3:scale=0:w=960:h=720:preblur=0.06:noise=0.14:details=0.33:halo=0.4:blur=0.23:compression=0.86:kcolor=1:device=0:vram=1:instances=0,scale=w=960:h=720:flags=lanczos:threads=0,split=2[out1][out2]” “-map” “[out1]” “-c:v” “tiff” “W:/Temp/436928168/temp/\_tvai\_249556639/%06d.tiff” “-map” “[out2]” “-c:v” “hevc\_nvenc” “-profile:v” “main10” “-preset” “medium” “-pix\_fmt” “p010le” “-b\_ref\_mode” “disabled” “-b:v” “0” “-map\_metadata” “0” “-movflags” "use\_metadata\_tags+write\_colr " “-map\_metadata:s:v” “0:s:v” “-map\_metadata:s:a” “0:s:a” “-map” “0:a” “-c:a” “copy” “-metadata” “videoai=Slowmo 100% and framerate changed to 60 using chr-2 ignoring duplicate frames. Enhanced using prob-3 with recover details at 33, dehalo at 40, reduce noise at 14, sharpen at 23, revert compression at 86, and anti-alias/deblur at 6. Changed resolution to 960x720” “W:/Temp/436928168/previews/Dungeons & Dragons Episode 19 The Last Illusion\_3\_chr2\_prob3.mov” However, the Chronos frames interpolation makes gray frames here: ![2023-03-03](https://community-cdn.topazlabs.com/original/3X/c/2/c2dcc027afd1d9b8848d79553e0226085fb5300f.gif) GIF doesn’t play all 60fps, but the gray frames are visible every time, where calculated image content otherwise would be. It’s Proteus manual, the “estimate” button, as expected, doesn’t work either, even after dropping the necessary model for Alpha. 🙂 Since it’s a preview, the creation/finalization of the 15s preview video file is fairly slow (TIFF files after all, from a HDD), it “feels” slower, than in current stable v3.1.7. ~~I hope the gray frames aren’t the cause of the increased performance…~~ 🙂 Edit: They’re not - with Apollo model I see ~30 fps in v3.1.7 vs ~35fps, and Artemis with Apollo - ~30fps vs. ~37 fps in this Alpha. And no gray frames. 🙂 Color Correction _maybe_ takes 1 fps out of the performance, if anything, I’d say the hit is negligible. Gray frames issue is only with both Chronos models, with and without Replace Duplicate Frames option. [Next page](https://community.topazlabs.com/t/topaz-video-ai-performance-alpha-for-windows-nvidia-amd-gpus/40961.md?page=2)