It would be very helpful if we also had the possibility to perform user benchmarks in TPAI as in TVAI.
There are benchmarks on the Internet, but only with the latest hardware and only for one OS platform (Mac OR Windows) and not cross-platform.
With the benchmarks of all participating TPAI users you have a good overview of the performance of your own computer, and it also helps in the selection and purchase of new hardware!
I would be happy if I am not the only one who requests this feature and if it will be realized in one of the next releases!
We have a system requirements page but if there’s enough demand from users to see where their machine stands then this is potentially something we could build.
@tim.he Thanks for your answer!
But it’s not only about seeing where our device stands, but also about finding out which individual components bring added value in terms of performance when purchasing new hardware.
In the TVAI User Benchmarks forum, for example, a user was able to decide for himself that a cheaper graphics card could definitely keep up with more expensive models and chose the inexpensive model.
And it would be a good help in deciding whether to change the OS because of the performance.
Not that I have any great plans for upgrading at the moment but since I’m running a somewhat unusual (I suppose) setup with Windows installed as a VM with GPU passthrough with Linux Mint as host it could always be nice to see how that compares with other computers.
(I currently pass through a AMD RX 6700 GPU and usually dedicate 12 threads of the 24 I have on my Ryzen 9 5900x and 16GB of RAM to Windows)
It is not only performance but issues that arrive when using certain platforms and OS. Both in this forum and Skylum (luminar NEO) most of the problems happen with Macs and Nvidia GPUs (drivers mostly). I use Windows 11 and AMD RX 5600 XT GPU and rarely have any problems with updates.
I think some performance tests would be helpful, but this would require the use of a standard picture that users would download because picture size and perhaps content could affect times. Unless Topaz includes a benchmark test in the program. We can only compare one version of PAI to be consistent.
While using TVAI data, the idea is to apply this concept to TPAI and TG.
Here’s a developer-focused report using anonymized aliases (User E, User F, etc.) to protect user identities, while retaining all detected patterns, version comparisons, and hardware insights from the benchmarking data
Despite typical expectations of the 4090 outperforming the 5080, results are module-dependent and not universally so. This hints at hardware-model pairing effects.
Beta vs. Shipping Code Regression/Improvement Highlight
User E: Results for v6.2.0 and v7.0.1 are closely matched on Artemis 1X, indicating stable performance across major versions for flagship setups. Small dips (e.g., Iris 1X: 2.7 fps lower) warrant closer look for regression testing consistency.
High-End Machine Envelope
User H: Threadripper 7960X + RTX 5090 delivers unprecedented highs (Artemis 1X at 56.59 fps, Nyx Fast at 42.18 fps). Use these as upper bounds for optimization and regression reference.
Intra-Cluster Hardware/Software Oddities
Apple M3/M4 Max setups:
User I: Mac performance outlier with Hyperion HDR at 74.79 fps (vs. 25–30 fps on similar Windows/RTX machines). This may indicate Apple platform optimizations or testing bugs, and should be cross-examined.
Outlier Examples
Users such as User J report failed/“ERR” benchmarks (Quadro P1000), suggesting model-driver incompatibility; relevant for product support and code stability improvements.
For Rhea 4X, some mid/high-tier systems show disproportionately low fps versus comparable cluster peers (potential model bottleneck or hardware issue).
Recommendations (Alias Version)
Track regressions/improvements with same-hardware cross-version runs.
Investigate per-module outliers and scaling effects across clusters for optimization targets.
Compare OS platform results for modules that deviate (e.g., Mac’s Hyperion HDR vs Windows).
Prioritize fixes for configurations showing failed tests (“ERR”), especially on less common hardware.
Use flagship machines (e.g., User H) as reference points for maximum achievable performance.
All references now use pseudonyms for user privacy, but every pattern and testing insight remains intact for developer analysis.
Here is a targeted upgrade report for two end users, using aliases (“User A”, “User B”, etc.), based on the provided benchmarking data.
One user represents a bottom cluster profile who decides they wan t to move to the top cluster of performance. The system recommends that they must upgrade their entire system. What the have will not benefit from the addition of new components.
A key concept here is that the larger tha sample size, that is the greater the number of users participating in the benchmark, the more accurate the representation of the entire Topaz compute community.
The second user has expressed who is from the middle cluster of performance to the top cluster of performs has been given a recommendation that requiring only a new GPU (RTX 50-series) due to an already adequate CPU.
User A: Bottom Cluster, Full System Upgrade
Current Specs
CPU: Intel Core i7-8700K
GPU: NVIDIA GeForce RTX 3060
RAM: 64 GB
Artemis 1X: 9.47 fps
Rhea 4X: 1.15 fps
Analysis & Upgrade Path
Performance falls well short of the top cluster benchmarks (compare to RTX 5090/5080 and high-end Ryzen/i9 CPUs, which achieve Artemis 1X >35 fps and Rhea 4X >4.5 fps).
Both CPU and GPU would heavily bottleneck modern upscaling and slow-motion tasks.
Required Action: Purchase a new system with a current high-end CPU (e.g., AMD Ryzen 9 7950X3D or Intel i9-14900KF) AND a latest-generation RTX 5090 or 5080 series GPU.
No partial upgrade is adequate; jumping clusters mandates a full platform refresh.
User B: Middle Cluster, GPU-Only Upgrade
Current Specs
CPU: Intel Core i9-12900KF
GPU: NVIDIA GeForce RTX 4060 Ti
RAM: 128 GB
Artemis 1X: 12.13 fps
Rhea 4X: 1.42 fps
Analysis & Upgrade Path
The CPU is already high-performance and matches those in the top cluster, but the GPU is a limiting factor for upscaling models.
To reach top-tier (e.g., Artemis 1X ≥ 40-55 fps, Rhea 4X ≥ 4.7 fps), only the GPU needs to be replaced.
Required Action: Install a new RTX 5090 or 5080 series GPU; this single upgrade, combined with the strong CPU, will shift this machine into the top performance cluster.
Power supply and case compatibility should be checked for the next-gen GPU, but other system components are sufficient.
Alias
Cluster
CPU
GPU
RAM
Artemis 1X
Rhea 4X
Needed for Top Cluster?
User A
Bottom
i7-8700K
RTX 3060
64 GB
9.47 fps
1.15 fps
New CPU + RTX 50-series GPU
User B
Middle
i9-12900KF
RTX 4060 Ti
128 GB
12.13 fps
1.42 fps
RTX 50-series GPU only
These personalized recommendations identify the most direct and effective path to the top performance tier for each user scenario using anonymized identifiers and respecting modern hardware requirements.
Performance Upgrade Recommendation – Low to Mid-Tier
Example: Low to Mid Cluster Upgrade
Selected Low-Performance User
User Name: [removed for privacy reasons]
Current Equipment:
CPU: Intel Core i7-8700K @ 3.70GHz
RAM: 63.936 GB DDR4
GPU: NVIDIA GeForce RTX 3060 (11.826 GB VRAM)
Performance (Artemis 1X): 9.47 fps
Percentile Ranking (Artemis 1X): Bottom 20–25% of all reported results (based on multiple records in attached files)
Recommended Upgrade Path
To move into the median of the “mid” cluster, an upgrade to a 40-series or high-end 30-series GPU combined with a recent-generation CPU is typical.
Upgrade Recommendation:
CPU: 12th Gen Intel Core i9-12900KF (from another mid-performing user)
RAM: 127.78 GB DDR5 (or at least upgrade to higher-speed DDR4/DDR5)
GPU: NVIDIA GeForce RTX 4060 Ti (15.731 GB VRAM)
Projected Performance
Performance (Artemis 1X): 12.13 fps (the “mid” cluster typical, as per actual benchmarking)
Projected Percentile: 40th–45th percentile, solidly in the middle of the pack
Direct, Actionable Advice
Upgrading from a 3080/3060 GPU + 8700K CPU system to a 4060 Ti GPU + 12900KF CPU (and maximizing RAM bandwidth) will move a user from low to median performance on Topaz Video AI.
This leap will deliver a ~28% FPS boost for core upscaling workloads, and also improve throughput on all other video enhancement features.
Additional, cost-saving options: just upgrading GPU will give a major bump, but pairing with current-gen CPU and fast RAM is preferred for long-term platform reliability and bandwidth.
Decision Model Logic
User’s benchmarking percentile and FPS are calculated directly from data, so targets are precise.
Upgrades and expected outcomes are matched to real configurations in the mid cluster, not theoretical specs.
Confidence is high due to empirical data ties between technical specs and observed performance.
Conclusion
This example demonstrates a real, empirically justified upgrade path from “low” to “mid” cluster, utilizing only available benchmarking and hardware data.
I can see Topaz Video’s benchmarks, but can’t see topaz Photos’ benchmark.
The previous iterations of Topaz Photo products had performance worse on 5xxx. So my question is - is 5080 faster than 4080Super and is 5090 faster than 4090? And how much faster? Anyone tested? Where are the benchmarks? Yes, I’ve use “search” function but did not find them…
