I do a lot of proteus upscaling from 6K video to 8K, I use a 4070 Ti video card on a 12th gen i7 CPU, and find i get around 0.5 FPS. I’d like to improve that on a budget (ie, i don’t want to dole out $4000 for a x090 series card). Beyond RAM speed you mentioned here, can you theorize anything else that would really help?
A post was merged into an existing topic: Video AI 7.0.X - User Benchmarking Results
OK, I’ll replace it as soon as DDR6 comes out 
Hello.
Just remember, software, especially creative/inference workloads, does not technically support UV/OC hardware; if data accuracy, stability, ect. (not to mention the extra heat/PROPER cooling system needed, lifespan of the hardware, ect… unless you are one of those that change/swap out your hardware… VERY often) is your main desire over so-called best-in-class “bragging” metric numbers.
If I understand you correctly, you are saying that undervolting can lead to hardware instability. The usual reason it is claimed to be used, is to lengthen the life of hardware, or reduce power draw, at the cost of longer processing times or less frames per second. Those two ideas are conflicting—unless you undervolt and find that it crashes right away. I presume that anyone reporting that they have undervolted is only doing so because they have been able to get everything running without issue.
Modern CPUs and GPUs apply ever more complex mechanisms to make them more efficient, to maximize speeds and to minimze safety margins (because AMD/Intel both want to win the benchmarks). Many are for regulating the current voltage/clock speed combination which is constantly fluctuating depending on workload, temp, silicon quality, … .
OC/UC ist just a way for the user to influence these mechanisms manually. There is nothing inherently instable in OC/UV. U can even make your system more stable (eg. Intel had some CPUs with too aggressive voltage curve causing silicon damage over time which could be partially fixed by user applied UV).
I OC/UV (now more UV, before more OC) every PC (and laptop) system since more than 30 years, also those I use professionally and they all run perfectly stable (ofc sometimes only after some finetuning), often for years.
Nonsense. No GPUs are designed for this extended behavior due to slew of reasons regardless of what the naysayers states.
Allow the complexity concerns to the architects of these GPU’s designs. No GPU is designed with these manipulated/modified frequencies in mind for every day use.
That doesn’t answer anything. Nonsense to what of the four things I said?
When I looked into overclocking, I was instructed by everyone that it can burn out your hardware faster than normal. Your hardware will burn out eventually, but the increased heat from overclocking will increase the chances of that happening sooner.
From that school of logic, underclocking should have the opposite effect on hardware since it is reducing the heat.
From that school of logic, underclocking should have the opposite effect
…again. The architecture was NOT designed in mind with OC/ UV values from its architects thoroughly research findings. Those scientist/authors goes through yEaRs of research… but if you allow naysayers/clueless minds persuade you… well, so be it. I’m not trying to save you from yourself! 
That’s an uninformed reply. I am admitting right now that I erroneously thought undervolting was the name for underclocking—though they are almost the same concept, except for the goal of trying to keep the same performance.
Engineers at companies like Nvidia have the job of figuring out safe parameters to run their GPUs at—with the catch that some will not run as good as others. They have to find parameters that will run best on most of the GPUs. All overclocking is, is using similar, if not the same, tools to adjust the same parameters as those engineers—in hopes that you got one of the units that can handle higher parameters better than the majority.
Some history:
In the past few years, algorithms have been implemented to maximize those parameters basically automatically. There’s little point in overclocking since its already been done on your GPU by default out of the box. That’s why undervolting is appealing. Underclocking might be appealing for some of the same reasons. The worst that’s going to happen while undervolting or underclocking is your GPU might not get as much power as it needs for a particular task and crash—but that’s explained well in every course on the subject.
1 Like
Nothing in this TL;DR reply is factual for any GPU architects’ designs. Maybe some ill manufacturer’s notes, but GPU architecture’s scientists/researchers (NVIDIA, ect.) are NOT the same as manufacturers’ (ASUS, ect) footprint.
Topaz Video AI v6.2.1
System Information
OS: Mac v15.0401
CPU: Apple M2 Pro 16 GB
GPU: Apple M2 Pro 10.667 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 07.51 fps 2X: 04.59 fps 4X: 01.03 fps
Iris 1X: 05.50 fps 2X: 03.15 fps 4X: 00.76 fps
Proteus 1X: 07.54 fps 2X: 04.85 fps 4X: 00.74 fps
Gaia 1X: 02.44 fps 2X: 01.72 fps 4X: 01.34 fps
Nyx 1X: 02.06 fps 2X: 02.03 fps
Nyx Fast 1X: 05.04 fps
Rhea 4X: 00.36 fps
RXL 4X: 00.39 fps
Hyperion HDR 1X: 27.56 fps
4X Slowmo Apollo: 08.63 fps APFast: 32.76 fps Chronos: 02.39 fps CHFast: 03.96 fps
16X Slowmo Aion: 05.67 fps
Topaz Video AI v6.2.2
System Information
OS: Windows v10.22
CPU: Intel(R) Core(TM) i7-6850K CPU @ 3.60GHz 31.91 GB
GPU: NVIDIA GeForce RTX 3080 9.8174 GB
Processing Settings
device: 0 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 19.98 fps 2X: 11.18 fps 4X: 03.14 fps
Iris 1X: 18.64 fps 2X: 10.97 fps 4X: 03.69 fps
Proteus 1X: 19.43 fps 2X: 13.07 fps 4X: 03.60 fps
Gaia 1X: 07.12 fps 2X: 05.15 fps 4X: 03.44 fps
Nyx 1X: 06.51 fps 2X: 04.60 fps
Nyx Fast 1X: 17.41 fps
Rhea 4X: 02.52 fps
RXL 4X: 02.49 fps
Hyperion HDR 1X: 28.03 fps
4X Slowmo Apollo: 23.26 fps APFast: 40.60 fps Chronos: 16.01 fps CHFast: 25.09 fps
16X Slowmo Aion: 26.69 fps
I haven’t seen anyone else post this configuration before, The ram DDR5 6400 CL 32.
Topaz Video AI v6.2.0
System Information
OS: Windows v10.22
CPU: 12th Gen Intel(R) Core(TM) i7-12700K 31.723 GB
GPU: NVIDIA GeForce RTX 5070 Ti 15.534 GB
Processing Settings
device: 0 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 31.11 fps 2X: 16.18 fps 4X: 04.64 fps
Iris 1X: 29.14 fps 2X: 16.90 fps 4X: 05.07 fps
Proteus 1X: 30.74 fps 2X: 19.55 fps 4X: 05.09 fps
Gaia 1X: 09.14 fps 2X: 06.78 fps 4X: 04.69 fps
Nyx 1X: 08.53 fps 2X: 06.61 fps
Nyx Fast 1X: 21.02 fps
Rhea 4X: 03.58 fps
RXL 4X: 03.48 fps
Hyperion HDR 1X: 27.37 fps
4X Slowmo Apollo: 34.15 fps APFast: 59.55 fps Chronos: 23.41 fps CHFast: 33.67 fps
16X Slowmo Aion: 43.38 fps
1 Like
I was wondering how the 5070 Ti would do! I was thinking about that video card or the 9070 XT.
I’ve been happy with it. I think Topaz probably works best on nvidia. Some times you can find topaz benchmark on pudget systems’ blog. They might have done a post about the 9070 series. Techpower up sometines includes topaz benchmarks in its GPU reviews.
Topaz Video AI v6.2.1
System Information
OS: Windows v29.09
CPU: 13th Gen Intel(R) Core(TM) i7-13700K 127.76 GB
GPU: NVIDIA CMP90HX (like GeForce RTX 3070ti, but mining card with slow Floating Point) 9.8373 GB
Processing Settings
device: -2 vram: 0.75 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 00.88 fps 2X: 00.63 fps 4X: 00.24 fps
Iris 1X: 01.19 fps 2X: 00.68 fps 4X: 00.22 fps
Proteus 1X: 00.86 fps 2X: 00.63 fps 4X: 00.24 fps
Gaia 1X: 00.26 fps 2X: 00.18 fps 4X: 00.16 fps
Nyx 1X: ERR fps 2X: ERR fps
Nyx Fast 1X: 01.12 fps
Rhea 4X: 00.15 fps
RXL 4X: 00.16 fps
Hyperion HDR 1X: ERR fps
4X Slowmo Apollo: 00.96 fps APFast: ERR fps Chronos: 00.66 fps CHFast: 01.03 fps
16X Slowmo Aion: ERR fps
But cmp90hx very fast at hashcat, like 3070ti, and cost only $100 on secondary markets.
Does it make a difference if you change the device to 0 and put the memory usage to 100%?
