Users can run benchmarking (Process > Benchmark, or, Ctrl/Cmd + B) to compare results across different machines.
Topaz Video AI v5.5.1
System Information
OS: Windows v11.24
CPU: Intel(R) Xeon(R) w5-2465X 127.25 GB
GPU: NVIDIA GeForce RTX 4070 11.73 GB
GPU: NVIDIA RTX A4000 15.79 GB
Processing Settings
device: 0 vram: 0.95 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 18.07 fps 2X: 12.19 fps 4X: 03.44 fps
Iris 1X: 19.24 fps 2X: 11.45 fps 4X: 03.27 fps
Proteus 1X: 17.71 fps 2X: 12.16 fps 4X: 04.18 fps
Gaia 1X: 06.11 fps 2X: 04.19 fps 4X: 02.86 fps
Nyx 1X: 07.21 fps 2X: 06.20 fps
Nyx Fast 1X: 13.46 fps
Rhea 4X: 02.35 fps
4X Slowmo Apollo: 24.13 fps APFast: 55.48 fps Chronos: 13.15 fps CHFast: 20.42 fps
16X Slowmo Aion: 23.71 fps
Topaz Video AI v5.5.1
System Information
OS: Windows v11.23
CPU: 12th Gen Intel(R) Core(TM) i5-12500 63.745 GB
GPU: NVIDIA GeForce RTX 4060 7.7705 GB
Processing Settings
device: 0 vram: 0.9 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 10.06 fps 2X: 07.30 fps 4X: 02.35 fps
Iris 1X: 10.83 fps 2X: 06.30 fps 4X: 01.75 fps
Proteus 1X: 10.23 fps 2X: 07.22 fps 4X: 02.39 fps
Gaia 1X: 03.42 fps 2X: 02.30 fps 4X: 01.56 fps
Nyx 1X: 04.13 fps 2X: 03.55 fps
Nyx Fast 1X: 07.53 fps
Rhea 4X: 01.37 fps
4X Slowmo Apollo: 15.39 fps APFast: 40.87 fps Chronos: 08.24 fps CHFast: 12.61 fps
16X Slowmo Aion: ERR fps
Topaz Video AI v5.5.1
System Information
OS: Windows v11.24
CPU: Intel(R) Core(TM) i7-14700K 95.715 GB
GPU: NVIDIA GeForce RTX 4080 SUPER 15.671 GB
GPU: Intel(R) UHD Graphics 770 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 29.41 fps 2X: 16.18 fps 4X: 04.39 fps
Iris 1X: 28.47 fps 2X: 19.24 fps 4X: 04.69 fps
Proteus 1X: 28.28 fps 2X: 19.46 fps 4X: 05.05 fps
Gaia 1X: 10.60 fps 2X: 07.28 fps 4X: 04.68 fps
Nyx 1X: 12.15 fps 2X: 10.29 fps
Nyx Fast 1X: 22.00 fps
Rhea 4X: 03.98 fps
4X Slowmo Apollo: 37.38 fps APFast: 86.61 fps Chronos: 23.30 fps CHFast: 30.06 fps
16X Slowmo Aion: 45.03 fps
Hello.
Just a heads-up! Intel is releasing a new microcode for the 200 series (some MB OEMs are finalizing test now) that should increase performance and improve inference workloads, ect.
Topaz Video AI v5.5.1
System Information
OS: Windows v10.22
CPU: AMD Ryzen Threadripper 3970X 32-Core Processor 127.87 GB
GPU: NVIDIA GeForce RTX 3090 23.756 GB
Processing Settings
device: -2 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 24.22 fps 2X: 13.16 fps 4X: 03.62 fps
Iris 1X: 22.39 fps 2X: 14.43 fps 4X: 03.67 fps
Proteus 1X: 23.39 fps 2X: 14.11 fps 4X: 03.78 fps
Gaia 1X: 08.30 fps 2X: 05.77 fps 4X: 03.21 fps
Nyx 1X: 10.27 fps 2X: 08.38 fps
Nyx Fast 1X: 20.41 fps
Rhea 4X: 03.02 fps
4X Slowmo Apollo: 32.48 fps APFast: 55.96 fps Chronos: 19.04 fps CHFast: 26.95 fps
16X Slowmo Aion: ... fps
Topaz Video AI v5.5.1
System Information
OS: Windows v11.24
CPU: Intel(R) Core(TM) i7-9700K CPU @ 3.60GHz 31.843 GB
GPU: NVIDIA GeForce RTX 3080 9.8174 GB
GPU: Intel(R) UHD Graphics 630 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 19.28 fps 2X: 09.41 fps 4X: 02.75 fps
Iris 1X: 19.16 fps 2X: 11.19 fps 4X: 03.26 fps
Proteus 1X: 18.67 fps 2X: 11.73 fps 4X: 03.21 fps
Gaia 1X: 06.55 fps 2X: 04.59 fps 4X: 02.93 fps
Nyx 1X: 08.38 fps 2X: 06.58 fps
Nyx Fast 1X: 15.65 fps
Rhea 4X: 02.64 fps
4X Slowmo Apollo: 22.30 fps APFast: 45.85 fps Chronos: 14.64 fps CHFast: 21.89 fps
16X Slowmo Aion: ... fps
Topaz Video AI v5.5.1
System Information
OS: Windows v11.23
CPU: Intel(R) Core(TM) i7-14700K 31.781 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
GPU: Intel(R) UHD Graphics 770 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 42.86 fps 2X: 18.89 fps 4X: 05.86 fps
Iris 1X: 35.83 fps 2X: 23.68 fps 4X: 05.94 fps
Proteus 1X: 41.09 fps 2X: 21.28 fps 4X: 06.03 fps
Gaia 1X: 15.51 fps 2X: 10.77 fps 4X: 05.53 fps
Nyx 1X: 17.41 fps 2X: 14.47 fps
Nyx Fast 1X: 30.79 fps
Rhea 4X: 05.64 fps
4X Slowmo Apollo: 39.95 fps APFast: 80.08 fps Chronos: 30.38 fps CHFast: 33.73 fps
16X Slowmo Aion: 32.93 fps
Wow!
It’s cool to see the 9000 series is still holding up after so many years.
Yeah, that old build from late 2020 is still capable of producing around 60% fps of my main build, so it can still help me speed up my upscale vaulting process daily. I also have M1 Pro base model MBP but the performance on MacOS is not as good as Intel CPU + Nvidia build; only M2 Ultra or maybe M4 Ultra next year can compete with.
Topaz Video AI v5.5.1
System Information
OS: Mac v15.02
CPU: Apple M2 Max 96 GB
GPU: Apple M2 Max 72 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 11.37 fps 2X: 07.02 fps 4X: 02.39 fps
Iris 1X: 10.44 fps 2X: 04.02 fps 4X: 01.51 fps
Proteus 1X: 11.56 fps 2X: 07.50 fps 4X: 02.11 fps
Gaia 1X: 03.28 fps 2X: 02.22 fps 4X: 01.63 fps
Nyx 1X: 03.41 fps 2X: 03.03 fps
Nyx Fast 1X: 09.73 fps
Rhea 4X: 00.85 fps
4X Slowmo Apollo: 12.61 fps APFast: 43.25 fps Chronos: 03.69 fps CHFast: 06.18 fps
16X Slowmo Aion: 12.60 fps
Topaz Video AI v5.5.1
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 3950X 16-Core Processor 31.928 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 29.78 fps 2X: 13.01 fps 4X: 03.17 fps
Iris 1X: 39.50 fps 2X: 14.39 fps 4X: 03.84 fps
Proteus 1X: 41.10 fps 2X: 15.15 fps 4X: 04.03 fps
Gaia 1X: 16.55 fps 2X: 11.05 fps 4X: 03.56 fps
Nyx 1X: 18.25 fps 2X: 13.52 fps
Nyx Fast 1X: 32.23 fps
Rhea 4X: 03.55 fps
4X Slowmo Apollo: 35.09 fps APFast: 54.96 fps Chronos: 32.95 fps CHFast: 29.80 fps
16X Slowmo Aion: 31.18 fps
AsRock X570 Phantom Gaming-ITX/TB3 Mainboard (BIOS vP5.60, 01/18/2024)
AMD Ryzen 9 3950X
16GB x2 (32GB) DDR4-3600 16/19/19/39/85/1T
MSI 4090 Suprim Liquid X AIO
nVidia Studio Driver v566.36 (12/10/2024)
Topaz Video AI v5.5.1
System Information
OS: Windows v11.23
CPU: AMD Ryzen 7 7800X3D 8-Core Processor 27.604 GB
GPU: NVIDIA GeForce RTX 3080 9.8174 GB
GPU: AMD Radeon(TM) Graphics 3.949 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 20.69 fps 2X: 14.50 fps 4X: 04.55 fps
Iris 1X: 20.21 fps 2X: 12.87 fps 4X: 03.91 fps
Proteus 1X: 20.05 fps 2X: 14.06 fps 4X: 05.28 fps
Gaia 1X: 06.76 fps 2X: 04.71 fps 4X: 03.21 fps
Nyx 1X: 08.66 fps 2X: 07.18 fps
Nyx Fast 1X: 16.87 fps
Rhea 4X: 02.76 fps
4X Slowmo Apollo: 24.47 fps APFast: 65.65 fps Chronos: 14.91 fps CHFast: 24.35 fps
16X Slowmo Aion: 41.31 fps
Replaced thermal pads and thermal paste on my HP RTX 3080 a few weeks ago, yesterday i also changed the fabric fans from 2x80mm and 1x92mm to 2x Noctua NF-A12-x25. 3D-printed shroud, far from optimal, as the fans diameter is larger than the width of the card by about 25mm. Not able to direct that air into the copper flanges as that would block majority of the hot exhaust air, forcing it towards the motherboard instead which would cause a lot of other parts to get higher temps.
However, end result was better than i thought. Ran TVAi16h of non-stop rendering while i slept and kept it going when i went to work. No power limit on the GPU, with a slight overclock (mostly at the lower voltages). Over 16h non-stop rendering the GPU had an avg. 318W(rtx 3080 max power is 320W) power consumption, avg. clock of 1830MHz(base clock of a 3080 is 1440 and boost clock 1710Mhz). Also running a memory overlock of 800Mhz.
With the new gpu fans limited to 58% up to 82C (fan speeds are set to only increase if core temp goes above 82C, and past that a max speed of 70%) and case fans between 50-55%(they follow either GPU fan curve or CPU fan curve - whichever is the highest, but with different negative offsets depending on case fan), it kept an average temperature of below 80C core, 91C hot spot and 86C memory junction.
CPU is air-cooled by a Noctua D15, case fans are also Noctua. At the above speeds, the computer is essentially completely silent, while maintaining temps that cause little to no degradation.
The odd memory showing ~28GB is due to manually allocated 4GB to the iGPU which i’ve offloaded most things to, as well as one display, that doesn’t depend on a high end gpu. Since TVAi will make use of all rtx gpu load it gets its hand on, other software can suffer.
Thanks for the ethusis posting modding process with such a detail. I had done the similar modding like your RTX 3080 on my RTX 3080FE around 3 years ago (full set of Noctua fans in PC case and RTX 3080FE replaced with AIO waterblock + new thermal pads).
Damn, waterblock too? Must have been a solid setup.
It’s funny, the 3D-printed shroud i got made was due to pure luck and coincidence. I don’t have a 3d-printer, have never printed anything in my life. I don’t know anyone with one either. I started considering it just 2 weeks ago, but quickly dismissed the idea due to the above reasons and having no idea what the cost of finding a professional to do the design and print it would cost.
Just a week ago though, i went for a drop-in at a hair saloon i’d never been at. Got along well with the owner so he gave me a key ring with the flag of his home country “Here you go, i’ve made this myself.” and points towards his 3d-printer. I jumped at the opportunity and presented my idea. He does some IT stuff on the side so he found my it super interesting. So he went home and designed it that same evening for fun. Didn’t ask for more than $20, material and design inkluded.
Tbh, I rather go with your mod instead lol, I found out my modding might not worth at all after all the experience I had. That waterblock was expensive , it did immediately drop the VRAM temperatures while gaming, but I found out the temperature would slowly back to 90 degrees, I had to open the waterblock to replace the pad frequently. 🫨 maybe just something was wrong at that time, either issue from installation or pads, which cause overheating, so that ended up switching the closure back to its originals. For using TVAI, I found Artemis and Nyx model.stress GPU and cause lesoverheating, so I almost just let it do Proteus work.
Ah, i see. The radiator couldn’t keep up with cooling the liquid which slowly caused it to rise.
Below is a picture from last nights render. No average interval sample amount set in HWiNFO. I just reset it right after initiating TVAi.
Only renderd 1080p. A mix of Chronos or Chronos Fast + Prot or Art. No upscaling, just enhancement and fps increase from a variety of of 24-30 fps on the input files to 59.94 fps.
Those average temperatures/loads/powers are over 9h as you can see in the bottom right corner. Intake fans at 50%(2x NF-A14 PWM, 1x120mm corsair fan that came with the case), outtake at 55%(1x NF-A14 PWM, 1x NF-S12A PWM) and the new 2x NF-A12x25 PWM fans on the gpu at 58%. Cpu fans at no more than 40% majority of the time. Have the Corsair 4000D Airflow case.
How’s it going with your 4090?
Since 3080 was a bit of an outlier performance wise in the 30 series, I’ll probably hold on to it, unless they release a card similar to 4090 in the 50 series, but at $1200 +/- $200 at most. I don’t need extremely high VRAM for what i do but would appreciate more tensor cores and overall faster processing of AI based gpu loads.
4k input videos use about 8-9Gb out of 10gb with the 3080. Not sure if it leaves a bit reserved, or if 8-9gb is what enhancing 4k with TVAi requires. How much vram does your 4090 use when you render a 4k video?
same, mostly 1080p → 2160p, I mainly do upscaling on Artemis and Proteus, sometimes Nyx, or frame interpolate model for video platforms.
See the status of 1080p → 2160p Proteus from 9700k + 3080FE (original) below:
The same process of Proteus speed from 14700k + 4090FE is about 1.8x the 9700k + 3080FE build, the status is below.
I don’t have the waterblock + radiator on 3080FE anymore, so I don’t know how much fps can be increased with. These days, I would just use the default setting and original part if the performance boost and time savings did not exceed about 25%. This way, I would not waste too much time and money on a small boost.
Topaz Video AI v5.5.1
System Information
OS: Windows v11.23
CPU: 13th Gen Intel(R) Core(TM) i9-13900K 32 GB
GPU: NVIDIA GeForce RTX 4090 24 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 41.34 fps 2X: 16.24 fps 4X: 04.76 fps
Iris 1X: 35.95 fps 2X: 21.16 fps 4X: 05.87 fps
Proteus 1X: 40.12 fps 2X: 23.16 fps 4X: 05.76 fps
Gaia 1X: 14.42 fps 2X: 10.33 fps 4X: 05.26 fps
Nyx 1X: 17.65 fps 2X: 13.40 fps
Nyx Fast 1X: 32.12 fps
Rhea 4X: 04.52 fps
4X Slowmo
Apollo: 37.54 fps
APFast: 75.55 fps
Chronos: 32.44 fps
CHFast: 32.83 fps
16X Slowmo
Aion: 32.44 fps
OK…first one I think. I just built this so not much tuning at all yet. For some reason TVAI will not give me a Aion score at the end. It just runs forever every time.
Topaz Video AI v5.5.1
System Information
OS: Windows v11.24
CPU: Intel(R) Core(TM) Ultra 9 285K 47.428 GB
GPU: NVIDIA GeForce RTX 4090 23.576 GB
GPU: Intel(R) Graphics 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 43.59 fps 2X: 19.89 fps 4X: 05.42 fps
Iris 1X: 39.44 fps 2X: 20.02 fps 4X: 05.08 fps
Proteus 1X: 42.99 fps 2X: 20.65 fps 4X: 05.05 fps
Gaia 1X: 16.53 fps 2X: 11.40 fps 4X: 04.91 fps
Nyx 1X: 18.39 fps 2X: 15.52 fps
Nyx Fast 1X: 32.91 fps
Rhea 4X: 04.79 fps
4X Slowmo Apollo: 40.09 fps APFast: 84.58 fps Chronos: 33.83 fps CHFast: 32.75 fps
16X Slowmo Aion: ... fps
AsRock Z890I Nova WiFi Mainboard (BIOS v2.22, 12/10/2024, Intel uCode 0x113)
Intel Core Ultra 9 (Series 2) 285K
24GB x2 (48GB) DDR5-9200 42/56/56/134/968/2T
MSI 4090 Suprim Liquid X AIO
nVidia Studio Driver v566.36 (12/10/2024)
Yeah, I just tried the just (in the last few hours) released 0x114 uCode BIOS on my system. No bump in performance here. Still experimenting though. I’ve been waiting on it. BIOS v2.23.AS01
Great to see some 285K benches!
Interestingly, the 1X results are top of the line (also compared to other 4090 builds) but the 2X and 4X fall a little bit short. I guess there is some more performance to be had by optimizing RAM timings. If u have something like AIDA64 for RAM benchmarks by chance, that would be really interesting to see.