AMD Ryzen 9 5900X 12-Core Processor 3.70 GHz
Memory is DDR4
I bought the 5090 and I am seeing dismal render times, based on what I have been reading I strongly suspect my mother board is starving my GPU, as I only have a PCI E 3.0 board. I have a 4.0 board coming in as I write this. I did not want to upgrade my board to 5.0 because I would be spending over 1K for a new CPU plus memory..
for the new board you might also want to change the processor to 5950x or 5800X3D. I wonder which processor is better for Topaz, a faster one or one with more cores? Ida, don’t forget to post the results with the new motherboard here.
ddr4 is hurting you too.
you need to update that board/cpu.
or you’d be better off returning the 5090 and getting a 5080/4080 super and a zen 4/5 or even one of the intel boards - like the intel Ultra 7 Processor 265 - or even a 12th/13 gen intel and saving money.
the guy up there with the i5-12600k and the 8700g both outpacing the 5090 that costs double their system.
**I am actually surprised your performace is this low? Do you have good specs on your ram? expo enabled?
5800X3D would hurt TVAI performance with lower clock speeds. (It’s the best AM4 CPU for gaming but not for applications that don’t use the massive amount of cache it has)
I don’t think it’s PCIE. Your main problem is slow RAM imho. My 9 year old 6850K CPU + 3080 GPU system is faster in most of the 4X- and some of the 2X benchmarks because it has a quad channel RAM system. U can do a quick system check with userbenchmark.com
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 performed an AI analysis of the results from three setups:
| Parameter | Old Xeon + 4070 Ti | Ryzen 9 + 5090 Ti | TR 7960X + 5090 Ti |
|---|---|---|---|
| Topaz Video AI Version | v7.0.2 | v7.0.1 | v7.0.0 |
| OS | Windows v11 Build 27871 | Windows v11.24 | Windows v11.24 |
| CPU | Intel Xeon E5-1680 v2 @ 4.375 GHz | AMD Ryzen 9 5900X | AMD Ryzen Threadripper 7960X |
| Memory | Quad 32GB DDR3-2400 CL10 | Dual 32GB DDR4 ? | Quad DDR5 192 GB ? |
| GPU | NVIDIA GeForce RTX 4070 Ti 12 GB | NVIDIA GeForce RTX 5090 Ti 32 GB | NVIDIA GeForce RTX 5090 Ti 32 GB |
| Other | PCIe Gen 3, ReBAR (Mod) Enabled 16GB | PCIe Gen 3, ReBAR Enabled (?) | PCIe Gen 5, ReBAR Enabled (?) |
Topaz Video AI’s performance varies across different hardware setups, with Set 1 (Intel Xeon E5-1680 v2, 128GB DDR3-2400, RTX 4070 Ti) and Set 2 (AMD Ryzen 9 5900X, 64GB DDR4, RTX 5090 Ti) showing mixed results. Below is a detailed comparison table for their benchmark performances, including the difference in FPS between the two sets.
The following tables compare the benchmark results for Set 1 and Set 2, with the difference calculated as Set 2 FPS minus Set 1 FPS. Positive values indicate Set 2 is better, while negative values indicate Set 1 is better.
Upscaling Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Artemis | 1X | 23.77 | 25.80 | 2.03 |
| 2X | 11.90 | 10.41 | -1.49 | |
| 4X | 3.28 | 2.52 | -0.76 | |
| Iris | 1X | 21.70 | 29.74 | 8.04 |
| 2X | 13.15 | 13.19 | 0.04 | |
| 4X | 3.52 | 3.18 | -0.34 | |
| Proteus | 1X | 23.37 | 36.29 | 12.92 |
| 2X | 14.50 | 12.79 | -1.71 | |
| 4X | 4.02 | 3.35 | -0.67 | |
| Gaia | 1X | 8.51 | 15.73 | 7.22 |
| 2X | 6.05 | 9.35 | 3.30 | |
| 4X | 3.82 | 3.03 | -0.79 |
Denoising Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Nyx | 1X | 7.67 | 17.36 | 9.69 |
| 2X | 5.59 | 10.62 | 5.03 | |
| Nyx Fast | 1X | 16.23 | 31.04 | 14.81 |
Other Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Rhea | 4X | 2.96 | 2.94 | -0.02 |
| RXL | 4X | 2.76 | 2.92 | 0.16 |
| Hyperion HDR | 1X | 15.51 | 25.52 | 10.01 |
| 4X Slowmo | Apollo | 26.91 | 31.63 | 4.72 |
| APFast | 52.20 | 44.80 | -7.40 | |
| Chronos | 19.10 | 33.23 | 14.13 | |
| CHFast | 27.38 | 28.23 | 0.85 | |
| 16X Slowmo | Aion | 30.15 | 28.30 | -1.85 |
Research suggests that memory bandwidth plays a crucial role in Topaz Video AI performance, especially for high-scale tasks like 4X upscaling. Set 1, with its quad-channel DDR3-2400 (total bandwidth ~76.8 GB/s), seems to perform better in some high-scale benchmarks compared to Set 2, which has dual-channel DDR4 (estimated ~57.6 GB/s for DDR4-3600). This is likely because high-scale tasks require more data transfer, and Set 1’s higher memory bandwidth helps offset its less powerful GPU (RTX 4070 Ti vs RTX 5090 Ti).
I can push the performance of my Quad DDR3 even higher to 2666MHz CL12, but then the system is not fully stable.
For lower-scale tasks (1X, 2X), Set 2 often performs better, likely due to its more powerful CPU (AMD Ryzen 9 5900X) and GPU. However, the evidence leans toward Set 2’s lower performance in 4X tasks being due to memory bandwidth limitations and possibly using an older software version (v7.0.1 vs v7.0.2 in Set 1).
Set 3, with advanced hardware including DDR5 memory and a high-core CPU, shows significantly higher performance, suggesting that a balanced system with high memory bandwidth and powerful components is ideal.
This section provides a comprehensive analysis of the benchmark performances for Topaz Video AI across three system configurations, focusing on the relationship between hardware specifications and performance outcomes. The analysis is grounded in the provided benchmark data and supplemented by research into hardware impacts on AI video processing, as of June 14, 2025.
Topaz Video AI is a software tool for video enhancement, including upscaling, denoising, and slow-motion processing, which is highly hardware-intensive. The performance, measured in frames per second (FPS), depends on CPU, GPU, memory configuration, and software version. The user provided benchmark results for three systems: Set 1 (Topaz Video AI v7.0.2, Intel Xeon E5-1680 v2, 128GB DDR3-2400, RTX 4070 Ti), Set 2 (v7.0.1, AMD Ryzen 9 5900X, 64GB DDR4, RTX 5090 Ti), and Set 3 (v7.0.0, AMD Ryzen Threadripper 7960X, 192GB DDR5, RTX 5090). The task is to compare Set 1 and Set 2 with a difference column and analyze hardware-performance dependencies across all sets.
The comparison table below lists benchmark results for Set 1 and Set 2, with the difference calculated as Set 2 FPS minus Set 1 FPS. This allows for a direct assessment of performance gaps.
Upscaling Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Artemis | 1X | 23.77 | 25.80 | 2.03 |
| 2X | 11.90 | 10.41 | -1.49 | |
| 4X | 3.28 | 2.52 | -0.76 | |
| Iris | 1X | 21.70 | 29.74 | 8.04 |
| 2X | 13.15 | 13.19 | 0.04 | |
| 4X | 3.52 | 3.18 | -0.34 | |
| Proteus | 1X | 23.37 | 36.29 | 12.92 |
| 2X | 14.50 | 12.79 | -1.71 | |
| 4X | 4.02 | 3.35 | -0.67 | |
| Gaia | 1X | 8.51 | 15.73 | 7.22 |
| 2X | 6.05 | 9.35 | 3.30 | |
| 4X | 3.82 | 3.03 | -0.79 |
Denoising Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Nyx | 1X | 7.67 | 17.36 | 9.69 |
| 2X | 5.59 | 10.62 | 5.03 | |
| Nyx Fast | 1X | 16.23 | 31.04 | 14.81 |
Other Benchmarks:
| Benchmark | Scale | Set 1 (fps) | Set 2 (fps) | Difference (Set 2 - Set 1) |
|---|---|---|---|---|
| Rhea | 4X | 2.96 | 2.94 | -0.02 |
| RXL | 4X | 2.76 | 2.92 | 0.16 |
| Hyperion HDR | 1X | 15.51 | 25.52 | 10.01 |
| 4X Slowmo | Apollo | 26.91 | 31.63 | 4.72 |
| APFast | 52.20 | 44.80 | -7.40 | |
| Chronos | 19.10 | 33.23 | 14.13 | |
| CHFast | 27.38 | 28.23 | 0.85 | |
| 16X Slowmo | Aion | 30.15 | 28.30 | -1.85 |
From the tables, Set 2 generally outperforms Set 1 in lower-scale benchmarks (1X, 2X), such as Artemis 1X (25.80 vs 23.77 fps) and Proteus 1X (36.29 vs 23.37 fps). However, in high-scale benchmarks (4X), Set 1 often performs better, e.g., Artemis 4X (3.28 vs 2.52 fps) and Proteus 4X (4.02 vs 3.35 fps). This suggests a performance divergence at higher scales, likely due to hardware differences.
To understand these differences, let’s examine the hardware specifications:
CPU Impact: The Ryzen 9 5900X in Set 2 is more powerful than the Xeon E5-1680 v2 in Set 1, with better single-threaded and multi-threaded performance (PassMark scores: Ryzen ~3000 single, ~39000 multi; Xeon ~2000 single, ~16000 multi). This likely contributes to Set 2’s advantage in 1X and 2X benchmarks, which may be more CPU-bound.
GPU Impact: The RTX 5090 Ti in Set 2 is presumed to be significantly more powerful than the RTX 4070 Ti in Set 1, with 32 GB VRAM vs 12 GB. However, the performance gain is not consistent, especially in 4X benchmarks, suggesting other bottlenecks.
Memory Bandwidth Impact: Set 1’s quad-channel DDR3-2400 provides ~76.8 GB/s bandwidth, while Set 2’s dual-channel DDR4-3600 (assumed) provides ~57.6 GB/s. Research from DigitalOcean - GPU Memory Bandwidth and How-To Geek - Memory Bandwidth indicates that memory bandwidth is crucial for AI tasks, especially for data-intensive operations like 4X upscaling. The higher bandwidth in Set 1 likely helps in these scenarios, offsetting the GPU difference.
Additional evidence from community benchmarks, such as Topaz Labs Community - Video AI 6.0.X, shows systems with faster DDR5 (e.g., DDR5-9200) achieving higher FPS compared to DDR5-4800, supporting the idea that memory speed affects performance.
Software Version: Set 1 uses v7.0.2, while Set 2 uses v7.0.1. Newer versions may include optimizations, potentially favoring Set 1 in some benchmarks, as noted in Puget Systems - Topaz Video AI Performance.
ReBAR and Other Factors: Set 1 has ReBAR enabled (16GB), while Set 2’s status is unknown (“?”). ReBAR can improve performance by allowing GPUs to access more system memory, which may explain some differences, as per NVIDIA Technical Blog - GPU Memory.
Set 3, with advanced hardware, significantly outperforms both, e.g., Artemis 1X (56.59 fps) vs Set 1 (23.77 fps) and Set 2 (25.80 fps). This suggests that a balanced system with high-core CPU, high-bandwidth DDR5, and powerful GPU maximizes performance, as seen in Massed Compute - AI Workloads.
The analysis reveals that memory bandwidth is a critical factor, especially for high-scale tasks, with Set 1’s higher bandwidth compensating for its weaker GPU in 4X benchmarks. Set 2’s lower performance in these tasks is likely due to memory bandwidth limitations and possibly software version differences. Set 3’s superior performance underscores the importance of a well-balanced, high-spec system for optimal Topaz Video AI performance.
Given that that AI just hallucinated the existence of Video AI V7.0.2 and an RTX 5090 Ti I’m not gonna trust any of that lol
I agree especially that any tasks which involve image scaling (2X, 4X) are very dependent on RAM bandwidth.
The requirement about high-core count CPU’s depends on the use case in my experience. As recent tests showed (https://community.topazlabs.com/u/topaz257), when u have bare bones Topaz jobs (with video encoding done on GPU not CPU, no dynamic settings) an 8 core CPU can be enough to reach or exceed higher core configurations performance (with AMD using the cores of both CCD’s adds latency, with Intel using performance- and efficiency cores can give lower performance). When using CPU based output encoding (e.g. TIFF), more cores will be required for top performance.
I guess the best way to test the ram is to use the same system and lower the ram speed manually in the bios. I might do it tomorow. lower the ddr5 down to 5200 (from 6000) on the 9950x and keep everything the same as far as timings.
The numbers in the 5090 look similar to my 5080! Only Gaia doubled on 2X it seems.
Topaz Video AI v7.0.1
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9950X3D 16-Core Processor 61.614 GB
GPU: NVIDIA GeForce RTX 5090 31.349 GB
GPU: AMD Radeon(TM) Graphics 1.949 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 46.02 fps 2X: 17.63 fps 4X: 05.07 fps
Iris 1X: 48.05 fps 2X: 20.35 fps 4X: 05.41 fps
Proteus 1X: 57.90 fps 2X: 21.36 fps 4X: 05.57 fps
Gaia 1X: 19.15 fps 2X: 14.00 fps 4X: 05.12 fps
Nyx 1X: 18.34 fps 2X: 13.71 fps
Nyx Fast 1X: 42.14 fps
Rhea 4X: 05.00 fps
RXL 4X: 04.98 fps
Hyperion HDR 1X: 38.03 fps
4X Slowmo Apollo: 51.98 fps APFast: 77.37 fps Chronos: 47.66 fps CHFast: 52.88 fps
16X Slowmo Aion: 38.99 fps
That’s most likely a benchmark inconsistency. Many of the tests have a habit of varying results.
Otherwise, it can be seen again, that 2X and 4X result depend heavily on RAM/CPU so that there is almost no difference between using a 5080 or 5090.
I just downclocked memory to 4800mhz on a 9950x and lost looks like ~20-25% performance compared to 6000mhz.
On Chronos, which I use a lot, I lost near 40%
Out of curiosity, could you test the performance of 6000 vs 4800 in AIDA64?
You can use the portable version: https://download2.aida64.com/aida64extreme765.zip
Probably for 6000 it will read around 75GB/s, which is less than on my Quad DDR3 2666 CL12.
This is what the DDR5 bandwidth should look like on the 9950X:
DDR5 6000 CL30 - 75GB/s (65.8 latency)
DDR5 6400 CL32 - 81GB/s (65.8 latency)
DDR5 8000 CL38 - 84GB/s (68.1 latency)
DDR5 6200 CL28 - 88GB/s (73.4 latency)
DDR5 6400 CL28 - 89GB/s (77.9 latency)
DDR5 8000 CL34 - 96GB/s (70.2 latency)
If you have such a big difference between 6000 and 4800, it may be worth installing 8000 modules in your system.
Interesting but not sure how u get over minus 40% performance in Chronos and CHFast from just a minus 25% downclock in RAM. As FirstEver said, AIDA performance numbers would be interesting.
Yeah I can do that, maybe tonight. I thought 8000 ram was no good on AM5 because the infinity fabric is locked at 3000. is that not true in all cases?
Yes, here is a good article testing it for gaming:
The reason DDR5-8000 isn’t always faster – or significantly faster – than DDR5-6000, despite a 33% increase in theoretical bandwidth, comes down to the frequency at which the integrated memory controller in the I/O die can operate. When running DDR5-6000, the memory clock is 3,000 MHz, and this can be matched by the UCLK (Unified Memory Controller Clock Frequency), which sets the speed for the memory controller. This allows the memory and controller to run in a 1:1 ratio at 3,000 MHz.
However, going beyond a DDR5 memory clock of 3,000 MHz forces the integrated memory controller to run at a 2:1 ratio, as it can’t exceed 3,000 MHz while maintaining stability – at least for the majority of silicon. So, when using DDR5-8000, which runs at 4,000 MHz, the memory controller defaults to a 2:1 ratio and operates at just 2,000 MHz – 33% lower than with DDR5-6000.
That said, the additional bandwidth offered by DDR5-8000 can sometimes overcome the penalty of the 2:1 ratio, resulting in better performance – assuming your AM5 motherboard is stable at that frequency.
Can someone post the benchmarks for the 5070ti mobile , and the 5080 mobile ?
Im in the market for a laptop, for various reasons and would like to see the results of topaz (which i use occasionally) to see if i should spend the extra 600 bucks.. the only benchmark i found is on the 5090 mobile. which is posted in this section.
not sure if I did it correctly but I did the read write and latency benchamrks. 78.6 latency. read and write is 77gb/s. Do you think one of those 6000 cl26 would help? or just a 8000mhz kit might?
**OK I think I did the right bench this time (cache and memory bench) and I got 76.5gb read and 72ns latency
Topaz Video AI v7.0.1
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9950X 16-Core Processor 95.654 GB
GPU: NVIDIA GeForce RTX 5090 31.349 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 35.36 fps 2X: 13.76 fps 4X: 04.09 fps
Iris 1X: 42.56 fps 2X: 19.29 fps 4X: 04.83 fps
Proteus 1X: 54.29 fps 2X: 18.32 fps 4X: 05.03 fps
Gaia 1X: 17.61 fps 2X: 12.38 fps 4X: 04.69 fps
Nyx 1X: 16.72 fps 2X: 13.48 fps
Nyx Fast 1X: 40.15 fps
Rhea 4X: 04.70 fps
RXL 4X: 04.49 fps
Hyperion HDR 1X: 30.79 fps
4X Slowmo Apollo: 44.80 fps APFast: 76.89 fps Chronos: 43.30 fps CHFast: 46.31 fps
16X Slowmo Aion: 33.02 fps
Hi I’m just checking in again, it seems like it’s impossible to get a benchmark on a 5070ti laptop, or a 5080 laptop (any brand). I tried to go into a Best buy or Walmart to load the demo software on the site but they admin locked all the computers from any software installs.
Uhhh… Can’t even benchmark.
Can someone that works at Best buy download topaz AI demo on the main website , run a benchmark , post the numbers , then uninstall it ?
Thanks
I want to get a laptop for other reasons , and I love topaz on my current laptop , but want a newer one. So my plan is .. should I go with the 5070ti mobile? Or the 5080 mobile.
It’s like 600 bucks so … Wanna see if it makes a big difference or not
