I have not tried bigger upscale than 854×480 to 1280×720. Maybe in your case it is a real bottleneck due to the final resolution of VAE decoding.
- Changing overlap from 21 to 8 seems risky to me.
When using other tools, like ltx outpaint values near 8 where not enough. It would work most of the time, but rarely I could see some bad hard frame cahnges.
Have you checked 8 frames overlap on many videos? What was the target resulotion? - VAE Decode size - yes it is critical that it will not excee 15.6GB. In the case of 854×480 to 1280×720 VAE= -1, gave me better results than changing the VAE Tile size.
- Yes, 0.85fps is probaly the ceiling of my 5060ti. It is also a 7 years old computer afer all
Btw, I did use the excellent mVolt+ app for some overclocking.
Glad it helped.
Btw my 321 frames chunk at least in my case of upsacle 854×480 to 1280×720 with 16GB vram, gave me slighly better results then 361 or 241.
Yes, I was surprised. Great work.
I will check v1.0.2 when I am back home. This hobby is starting to take too much time : )
Tried 1.0.2, but failed on every setting with target resolution 1920X1080 using the same input resolution. The autotune doesnt work either. When the last 24th setp of the autotune is being progressed, the fatal error message comes up without showing autotuned setting. more than 90 minutes of autotune process doesn’t give me any clue at all.
Only working version is still v1.0.0
Intel i5-14400, 32G RAM, RTX5060Ti(16G VRAM)
That popup indicates that your GPU is running out of VRAM. What output resolution are you using? A chunk size of 361 combined with an encode tile size of 1152 is quite aggressive for a 32 GB card.
Please also take a screenshot of the Monitor / FPS tab so we can compare how much VRAM SLP is using with how much is being consumed by background applications. I recommend running AutoTune to determine the best stable settings for your system.
For example, if you already know your system can handle chunk size 241 but not 361, you can add custom AutoTune rows to test intermediate values such as 261, 271, 311, and so on. The final tuning algorithm can only evaluate settings that were actually tested and completed successfully.
It isn’t practical for me to include every possible chunk-size interval in the default test suites. Each test takes time, hardware configurations vary enormously, and testing every possible combination could take weeks. The current suites include more fine-grained tests below 241 for systems with 16 or 24 GB of VRAM, since those systems benefit most from smaller increments. For higher-VRAM systems, extensively testing those smaller chunk sizes would generally waste time because AutoTune is unlikely to select them when larger, faster settings are stable.
v1.0.2 no longer sees the SLP26 folder since I have models in a non-default folder.
Even though I can assign a custom location for the Topaz executable, in v1.0.1 the SLP26 folder was also mapped correctly.
This has broken in v1.0.2.
I used to use temporal overlap of 12 with SEEDVR2, but gradually lowered it down to 8 because I couldn’t see any difference and I get a speed boost and less vram usage. If you have doubt, just use my video compare app to see if you notice the difference. In fact, I recommend you try overlap of 0 and 4 also and compare so you get a better understanding how this setting works. It’s very easy to compare with my app. Press space bar to pause/resume.
@skv89 looks like another issue with non-default paths when using AutoTune.
Can you at least post a full screenshot. What version of the app are you using? I need more info to diagnose the problem.
Hi @skv89
In the latest version, it gives this error and doesn’t start continuously, but ffmpeg is already installed on the system and is also in the Topaz directory.
C:\WINDOWS\system32>ffmpeg
ffmpeg version 9.0.1-essentials_build-www.gyan.dev Copyright (c) 2000-2026 the FFmpeg developers
built with gcc 16.1.0 (Rev2, Built by MSYS2 project)
configuration: --enable-gpl --enable-version3 --enable-static --disable-w32threads --disable-autodetect --enable-cairo --enable-fontconfig --enable-iconv --enable-gnutls --enable-libxml2 --enable-gmp --enable-bzlib --enable-lzma --enable-zlib --enable-libsrt --enable-libssh --enable-libzmq --enable-avisynth --enable-sdl2 --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxvid --enable-libaom --enable-libopenjpeg --enable-libvpx --enable-mediafoundation --enable-libass --enable-libfreetype --enable-libfribidi --enable-libharfbuzz --enable-libvidstab --enable-libvmaf --enable-libzimg --enable-amf --enable-cuda-llvm --enable-cuvid --enable-dxva2 --enable-d3d11va --enable-d3d12va --enable-ffnvcodec --enable-libvpl --enable-nvdec --enable-nvenc --enable-vaapi --enable-openal --enable-libgme --enable-libopenmpt --enable-libopencore-amrwb --enable-libmp3lame --enable-libtheora --enable-libvo-amrwbenc --enable-libgsm --enable-libopencore-amrnb --enable-libopus --enable-libspeex --enable-libvorbis --enable-librubberband
libavutil 61. 1.101 / 61. 1.101
libavcodec 63. 1.101 / 63. 1.101
libavformat 63. 1.101 / 63. 1.101
libavdevice 63. 1.101 / 63. 1.101
libavfilter 12. 1.101 / 12. 1.101
libswscale 10. 1.101 / 10. 1.101
libswresample 7. 1.101 / 7. 1.101
Universal media converter
usage: ffmpeg [options] [[infile options] -i infile]... {[outfile options] outfile}...
Use -h to get full help or, even better, run 'man ffmpeg'
Thanks for sharing. Latest as in v1.0.2.3? Will look into this. I really need to get the autotune feature fully compatible and working as intended across all the different possible hardware configs. It’s been proving much more difficult than I thought.
Yes, version 1.0.2.3, even after changing all the settings, still doesn’t start the SLP; it keeps giving me an OOM error. Only version 1.0 works; the others still give the OOM error.
Still doesn’t start SLP? Can you send me the log? The other 5070ti user in the v1.0.2 thread was able to run SLP with v1.0.2.3
SLP is working, but it keeps giving an OOM error during VAE encoding. I’ve tried all the settings, but it doesn’t work. However, it works with the same settings in version 1.0. You probably already know this, but let me remind you anyway: you need to tell the AI to make it work not only on your own system, but on all OS and GPU makes and models, ffmpeg versions, and all types of systems.
I only have a rtx3060 with 12gb ram and I have just tried version 1.0 today. I’m no expert and find it all a bit confusing. However in saying that when running starlight precise previously my maximum FPS was 0.2. The one I’m upscaling now using version 1.0 is running at 0.6. This is a welcome boost and I wish to thank skv89 for his hard work. I’ll try and understand a bit more about it all and then see how version 1.03 goes.
Wow that is a 3x speed increase!! I never heard of anything that much. You have a RTX 3000 series and I honestly did not expect would benefit much. I heard from some RTX 4000 series owners that they saw 2x speed increase but RTX 5000 / Blackwell series see much less speed gains. I only saw roughly 85% speed increase and I have 96gb vram. For less vram users, they will see less speed gains.
I just posted the new v1.0.3 that took into account the user reported issues and now there are many new improvements and diagnostic tools. Hopefully all the previously reported issues have been resolved. It works on my system but I can’t guarantee it will work the same on the infinite possible hardware configs and system environments out there.
Please continue the discussion in the new thread SLP Tuner Launcher v1.0.2 (Major SPEED BOOST) with Benchmark & Autotune built-in! - #50 by skv89
When the short test clip finished I ran a 9 minute video at a 3x upscale. The speed fluctuated in this. It reached a peak of 0.6fps but sometimes dropped to 0.0 for a while. It mainly stayed between 0.3 and 0.5 though and the video finished ( well it ran until 5mins 34 seconds of the 9 minutes because Windows update automatically restarted the computer when I was in bed). I am running the remainder of it now by cutting the original where it finished. Thanks again. Topaz should give you a job.
Are you using Topaz’s built in speed estimate? Please don’t use that as it is completely broken and jumps all over the place. Please use the speed monitor in the Monitor / FPS tab in my app as that is accurate.
