Before and after benchmark results from alpha release with optimizations for 50-series:
Note*
Some processing speeds are slower (Aion for example). These are likely due to change of CPU and non-optimized ram speed or tightened ram timings. The before results with the Ryzen 7 7800X3D had manually set ram timings and overclocked speed to 6200MT/s as well as high fclk.
The results from alpa with 50-series optimization are run with a Ryzen 9 9900X with only PBO activated and CO All -27 and regarding ram I’ve only enabled expo which run at 6000MT/s and much loser timing as well as default fclk.
Windows version also differ between the two as you can see. Benchmarks before optimization were run on Windows 11 23H2, benchmarks with optimizations on Windows 11 24H2.
Both benchmarks were done at fabric clock. Since 5080 is an overclocking beast, I’ll add a third benchmark with the alpha build where the rtx 5080 is overclocked, however, not overclocked to the brink of crashing - which is around +400-450. The overclocked benchmark was done at +350 core and +1500 memory with a tiny undervolt to reduce likelihood of a crash due to the transient load nature of VAi (Aion being the worst offender where power draw shifts between 20% and 80% several times per second), curve flattened at 1015mV
The 5080 doesn’t use more than at most, 80% power, even with no power limit set
Before optimization stock 5080 + tuned RAM and Ryzen 7 7800X3D:
Topaz Video AI Beta v6.0.4.0.b
System Information
OS: Windows v11.23
CPU: AMD Ryzen 7 7800X3D 8-Core Processor 31.604 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 25.02 fps 2X: 15.73 fps 4X: 04.83 fps
Iris 1X: 26.44 fps 2X: 15.08 fps 4X: 04.85 fps
Proteus 1X: 26.19 fps 2X: 19.37 fps 4X: 05.22 fps
Gaia 1X: 11.22 fps 2X: 07.73 fps 4X: 05.03 fps
Nyx 1X: 09.80 fps 2X: 08.03 fps
Nyx Fast 1X: 17.37 fps
Rhea 4X: 03.63 fps
RXL 4X: 03.48 fps
Hyperion HDR 1X: 33.59 fps
4X Slowmo Apollo: 32.49 fps AFast: 73.45 fps CH: 14.78 fps CHFast: 28.98 fps
16X Slowmo Aion: 52.62 fps
After optimization (alpha build) stock 5080 + Expo ram and Ryzen 9 9900X:
Topaz Video AI Alpha v6.1.1.1.a.trt
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9900X 12-Core Processor 31.602 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 34.09 fps 2X: 16.45 fps 4X: 04.66 fps
Iris 1X: 33.13 fps 2X: 19.14 fps 4X: 05.30 fps
Proteus 1X: 34.37 fps 2X: 20.11 fps 4X: 05.47 fps
Gaia 1X: 10.69 fps 2X: 07.84 fps 4X: 04.95 fps
Nyx 1X: 10.00 fps 2X: 08.06 fps
Nyx Fast 1X: 23.06 fps
Rhea 4X: 04.20 fps
RXL 4X: 04.03 fps
Hyperion HDR 1X: 30.21 fps
4X Slowmo Apollo: 41.97 fps APFast: 69.74 fps Chronos: 27.07 fps CHFast: 38.19 fps
16X Slowmo Aion: 39.52 fps
After optimization (alpha build) overclocked and voltage limit 5080 + Expo ram and Ryzen 9 9900X:
Topaz Video AI Alpha v6.1.1.1.a.trt
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9900X 12-Core Processor 31.602 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 36.38 fps 2X: 16.77 fps 4X: 04.66 fps
Iris 1X: 37.00 fps 2X: 20.10 fps 4X: 05.30 fps
Proteus 1X: 37.20 fps 2X: 19.80 fps 4X: 05.46 fps
Gaia 1X: 11.41 fps 2X: 08.35 fps 4X: 05.02 fps
Nyx 1X: 10.63 fps 2X: 08.43 fps
Nyx Fast 1X: 24.71 fps
Rhea 4X: 04.50 fps
RXL 4X: 04.28 fps
Hyperion HDR 1X: 32.31 fps
4X Slowmo Apollo: 43.47 fps APFast: 70.41 fps Chronos: 28.69 fps CHFast: 40.29 fps
16X Slowmo Aion: 46.94 fps
Looking at the generative models which seem to be the ones most affected by tensor cores, Artemis, Iris and Proteus specifically, the optimizations increase processing speed by about ~26-27%.
EDIT*
I went ahead and tuned my ram a bit and would like to show that ram speed and/or ram latency does have a significant impact on upscaling and some of the interpolation models. My 2x16GB 6000mhz@cl30 (with expo enabled) kit is now running at 6200mhz@cl30 with fclk at 2067, uclk=mclk and tighter sub-timings.
Apollo fast and 4x upscaling for the 4 basic models see a 10% uplift in processing speed. Some other around ~5%. So don’t neglect your ram.
VAi doesn’t use much ram, as it doesn’t store frames in ram. Ram is just used as a middle hand because frames are being processed by both CPU and GPU. Looking at upscaling as an example, and the ffmpeg export command. The GPU handles the generative part of enhancing a frame, while the CPU handles upscaling of frames, eg: scale=w=3840:h=2160:flags=lanczos:threads=0 with -sws_flags "spline+accurate_rnd+full_chroma_int" added before the filter graph which slightly refines the “scale” filter.
Reason some other models are seeing a ~1-2% difference is more likely just margin of error. Take Rhea and the basic models without upscaling as examples. There will always be a slight difference in processing speed even when running benchmarks back-to-back without changing any setting.
After optimization (alpha build) overclocked and voltage limit 5080 + manually tuned ram and Ryzen 9 9900X:
Topaz Video AI Alpha v6.1.1.1.a.trt
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9900X 12-Core Processor 31.602 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 37.73 fps 2X: 19.17 fps 4X: 05.41 fps
Iris 1X: 37.22 fps 2X: 20.42 fps 4X: 06.06 fps
Proteus 1X: 37.30 fps 2X: 23.41 fps 4X: 06.33 fps
Gaia 1X: 11.53 fps 2X: 08.46 fps 4X: 05.79 fps
Nyx 1X: 10.63 fps 2X: 08.64 fps
Nyx Fast 1X: 25.06 fps
Rhea 4X: 04.23 fps
RXL 4X: 04.43 fps
Hyperion HDR 1X: 33.26 fps
4X Slowmo Apollo: 46.36 fps APFast: 77.09 fps Chronos: 28.91 fps CHFast: 41.14 fps
16X Slowmo Aion: 49.31 fps
