I’am using Topaz Video Ai with 32Gb of Ram (2*16Gb @ 3200 Mhz).
Recently I had memory problems and I need to replace my memory for one new kit
With 32 Gb I had a output using Proteus model of 10 fps
After discovery what was the memory with problems I run again Topaz with the same preset and I had 5 fps of output usisng the same model settings
If I buy 64 Gb using a Rtx 4070 super and a Ryzen 9 3900X my output goes to 20 fps or the GPU and CPU will limited the output frame rate to some other value of fps?
Is it worth buying 64 instead of 32 or does the price difference not make up for the performance difference?
Staying on a relic chipset/MoBo is just crippling.
Upgrade to DDR5/32gb (if there’s a MoBo/RAM deal… 48gb MAX… ) 2-stick, upgrade your PSU (future proof yourself) and forget about it.
You can see in TaskManager how much ram TVAI uses, when you run out, buy 2 x 32GB. Ryzen9 3900X has a FSB of 1800, take 3600Mhz (Dual Channel) Memory
TVAI gets not faster as more Ram you have, but it gets super low when no Ram ist left
After using TVAI for a while now on both Macs and Windows, at least on Windows machine I’m starting to realize there is a limit… You can have the fastest CPU, GPU and the most RAM, but there is a limit the AI model will run at… This is where I’m at in my instance…
I have a Ryzen 7 9800X3D w/32GB DDR5, RTX 5080 w/16GB VRAM, on Win11 Pro. X870 chipset MoBo.
TVAI 6.2.0.0b and 6.1.3 have the same results in real world… Benchmarks between the 2 versions is a horse of a different color.
I’m currently running a deinterlace upscale to 1080p using IRIS MQ and here is what I’m seeing…
In TVAI is is reporting 19.0 to 20.5 fps.
CPU
AMD Ryzen 7 9800X3D 8-Core Processor
Base speed: 4.70 GHz
Sockets: 1
Cores: 8
Logical processors: 16
Virtualization: Enabled
L1 cache: 640 KB
L2 cache: 8.0 MB
L3 cache: 96.0 MB
Utilization 69%
Speed 5.18 GHz
Up time 0:19:33:35
Processes 263
Threads 5134
Handles 126000
GPU 0
NVIDIA GeForce RTX 5080
Driver version: 32.0.15.7283
Driver date: 3/14/2025
DirectX version: 12 (FL 12.1)
Physical location: PCI bus 1, device 0, function 0
Utilization 67%
Dedicated GPU memory 3.6/16.0 GB
Shared GPU memory 0.3/15.6 GB
GPU Memory 3.9/31.6 GB
CPU
AMD Ryzen 7 9800X3D 8-Core Processor
Base speed: 4.70 GHz
Sockets: 1
Cores: 8
Logical processors: 16
Virtualization: Enabled
L1 cache: 640 KB
L2 cache: 8.0 MB
L3 cache: 96.0 MB
Utilization 69%
Speed 5.18 GHz
Up time 0:19:33:35
Processes 263
Threads 5134
Handles 126000
As you can see, my CPU and GPU are only utilized 65 to 69% of the time. VRAM is 3.6GB used out of 16GB total… and I’m only getting an average of 19.7 fps.
So my bottleneck is really the AI model itself… It will only go so fast no matter how much horse power you have. Like a governor on an engine…The governor will only allow the engine to work so hard and then no more.
Below are the benchmarks my system experiences between 6.1.3 and 6.2.0.0b with 480 and 1080. My RTX 5080 SOLID has not been overclocked. It is stock!
Topaz Video AI v6.1.3
System Information
OS: Windows v11.24
CPU: AMD Ryzen 7 9800X3D 8-Core Processor 31.151 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
GPU: AMD Radeon(TM) Graphics 0.47438 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 27.54 fps 2X: 15.12 fps 4X: 03.88 fps
Iris 1X: 28.47 fps 2X: 16.59 fps 4X: 04.61 fps
Proteus 1X: 24.42 fps 2X: 16.77 fps 4X: 04.74 fps
Gaia 1X: 11.96 fps 2X: 07.98 fps 4X: 04.46 fps
Nyx 1X: 10.52 fps 2X: 08.62 fps
Nyx Fast 1X: 19.90 fps
Rhea 4X: 03.52 fps
RXL 4X: 03.30 fps
Hyperion HDR 1X: 32.46 fps
4X Slowmo Apollo: 35.31 fps APFast: 71.70 fps Chronos: 15.95 fps CHFast: 30.87 fps
16X Slowmo Aion: 44.55 fps
Topaz Video AI Beta v6.2.0.0.b
System Information
OS: Windows v11.24
CPU: AMD Ryzen 7 9800X3D 8-Core Processor 31.151 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
GPU: AMD Radeon(TM) Graphics 0.47438 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 38.22 fps 2X: 16.09 fps 4X: 04.20 fps
Iris 1X: 35.46 fps 2X: 20.72 fps 4X: 05.02 fps
Proteus 1X: 38.28 fps 2X: 19.79 fps 4X: 05.29 fps
Gaia 1X: 11.38 fps 2X: 08.19 fps 4X: 04.98 fps
Nyx 1X: 10.49 fps 2X: 08.59 fps
Nyx Fast 1X: 24.55 fps
Rhea 4X: 04.39 fps
RXL 4X: 04.25 fps
Hyperion HDR 1X: 32.16 fps
4X Slowmo Apollo: 46.93 fps APFast: 71.12 fps Chronos: 29.78 fps CHFast: 41.45 fps
16X Slowmo Aion: 52.44 fps
Topaz Video AI v6.1.3
System Information
OS: Windows v11.24
CPU: AMD Ryzen 7 9800X3D 8-Core Processor 31.151 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
GPU: AMD Radeon(TM) Graphics 0.47438 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 720x480
Benchmark Results
Artemis 1X: 143.24 fps 2X: 72.22 fps 4X: 19.31 fps
Iris 1X: 130.91 fps 2X: 67.24 fps 4X: 27.90 fps
Proteus 1X: 158.68 fps 2X: 74.76 fps 4X: 31.41 fps
Gaia 1X: 65.37 fps 2X: 35.86 fps 4X: 22.89 fps
Nyx 1X: 30.30 fps 2X: 29.04 fps
Nyx Fast 1X: 78.08 fps
Rhea 4X: 21.10 fps
RXL 4X: 21.99 fps
Hyperion HDR 1X: 100.85 fps
4X Slowmo Apollo: 187.70 fps APFast: 371.39 fps Chronos: 71.57 fps CHFast: 136.62 fps
16X Slowmo Aion: 128.72 fps
Topaz Video AI Beta v6.2.0.0.b
System Information
OS: Windows v11.24
CPU: AMD Ryzen 7 9800X3D 8-Core Processor 31.151 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
GPU: AMD Radeon(TM) Graphics 0.47438 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 720x480
Benchmark Results
Artemis 1X: 216.77 fps 2X: 89.12 fps 4X: 24.62 fps
Iris 1X: 168.10 fps 2X: 109.49 fps 4X: 32.83 fps
Proteus 1X: 212.22 fps 2X: 145.37 fps 4X: 34.89 fps
Gaia 1X: 77.71 fps 2X: 54.97 fps 4X: 32.76 fps
Nyx 1X: 29.80 fps 2X: 29.03 fps
Nyx Fast 1X: 92.07 fps
Rhea 4X: 26.39 fps
RXL 4X: 25.82 fps
Hyperion HDR 1X: 98.33 fps
4X Slowmo Apollo: 213.23 fps APFast: 339.17 fps Chronos: 165.05 fps CHFast: 190.66 fps
16X Slowmo Aion: 128.24 fps
I can’t tell for sure but I think that fitting two modules of will say 16 GB is faster than one module of 32 GB du to the dual channel most mainboards offer. If more memory will be an advantage for an individual application has to be tested though. 
Dual-Channel RAM vs. Single Module:
- Two 16GB modules (32GB total) running in dual-channel mode usually outperform a single 32GB stick running in single-channel mode.
- Dual-channel gives the CPU double the memory bandwidth, which helps in memory-intensive tasks—like video editing, upscaling with TVAI, gaming, or encoding.
When More Memory Matters:
- Whether more memory helps depends entirely on the application and workload.
- If you’re maxing out 32GB, then going to 64GB (even with a potential tradeoff in speed) could be a net win.
- If you’re using under 32GB, then dual-channel is the bigger benefit.
TL;DR:
- 2x16GB > 1x32GB (in terms of performance due to dual-channel).
- But total capacity trumps channel config if you’re running out of RAM.
