how can I benchmark my computer like this?
Quite interesting - even though the M3 Max has a HUGE amount of RAM it doesnāt really help with performance (except for Iris 1x and 2x, but Artemis is even slower 
).
Seems something isnāt fully optimized hereā¦
1 Like
WhooA! Clearly, very unoptimized results.
Your current CPU, ect. is destroying that cardās raw performance.
I doubt it. Itās got 12 cores and runs about as good as the 7900X.
1X performance seems fine 2X and 4X would probably benefit from faster RAM if u care about those modes
Right, but thatās about it. The better or newer AMD CPUs arenāt really going to give more speed. Going intel might, but thatās a whole system change, and itās probably based on how fast the RAM is too.
I use a 6000 mega transfers memory
Probably the asus pro art is also not the fastest 4080 out there.
Is there anything I could do to improve the performance?
Thanks for the feedback
Would be interesting to use AIDA64 to test for RAM speed. U can perform each RAM speed test by itself (read, write, copy, latency) beause the complete RAM speed benchmark dialog will obfuscate some results in free version.
1 Like
Thereās a HUGE life beyond just 1080p renders/processes, FS. 
I have overclocked my RAM
RTX 3060 Ti Dual OC with +146 Core and +799 MEM on MSI Afterburner
Topaz Video AI v6.1.3
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 7950X 16-Core Processor 63.656 GB
GPU: NVIDIA GeForce RTX 3060 Ti 7.8359 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 13.25 fps 2X: 09.93 fps 4X: 03.37 fps
Iris 1X: 12.73 fps 2X: 08.10 fps 4X: 02.51 fps
Proteus 1X: 12.92 fps 2X: 09.63 fps 4X: 03.42 fps
Gaia 1X: 04.39 fps 2X: 03.03 fps 4X: 02.05 fps
Nyx 1X: 05.32 fps 2X: 04.54 fps
Nyx Fast 1X: 10.44 fps
Rhea 4X: 01.76 fps
RXL 4X: 01.62 fps
Hyperion HDR 1X: 26.39 fps
4X Slowmo Apollo: 19.09 fps APFast: 52.28 fps Chronos: 10.06 fps CHFast: 16.47 fps
16X Slowmo Aion: 22.02 fps
RTX 5070 +399 CORE and +2000 MEM on MSI Afterburner
Topaz Video AI v6.1.3
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 7950X 16-Core Processor 63.656 GB
GPU: NVIDIA GeForce RTX 5070 11.577 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 18.53 fps 2X: 12.98 fps 4X: 04.00 fps
Iris 1X: 19.78 fps 2X: 11.58 fps 4X: 03.60 fps
Proteus 1X: 19.61 fps 2X: 14.44 fps 4X: 05.54 fps
Gaia 1X: 08.12 fps 2X: 05.39 fps 4X: 01.46 fps
Nyx 1X: 07.30 fps 2X: 06.10 fps
Nyx Fast 1X: 13.51 fps
Rhea 4X: 00.34 fps
RXL 4X: 01.06 fps
Hyperion HDR 1X: 26.33 fps
4X Slowmo Apollo: 28.08 fps APFast: 63.21 fps Chronos: 12.21 fps CHFast: 20.30 fps
16X Slowmo Aion: 42.97 fps
This is what OCād RAM looks likeā¦
Beta with āOptimizedā 5000-series models. Something is very wrong with the benchmarking of Rhea and RXL is this beta version. Both maxed out 16GB of vram + a bunch of dram.
Topaz Video AI Beta v6.1.4.0.b
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9900X 12-Core Processor 31.602 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 26.74 fps 2X: 18.36 fps 4X: 04.49 fps
Iris 1X: 28.53 fps 2X: 16.22 fps 4X: 05.48 fps
Proteus 1X: 28.31 fps 2X: 21.38 fps 4X: 06.12 fps
Gaia 1X: 12.89 fps 2X: 08.44 fps 4X: 03.98 fps
Nyx 1X: 10.73 fps 2X: 08.38 fps
Nyx Fast 1X: 20.39 fps
Rhea 4X: 00.43 fps
RXL 4X: 02.19 fps
Hyperion HDR 1X: 32.55 fps
4X Slowmo Apollo: 35.32 fps APFast: 74.17 fps Chronos: 16.67 fps CHFast: 25.45 fps
16X Slowmo Aion: 44.33 fps
Results from ALPHA release with optimizations for 50-series. Both ran one after the other just now. Rhea and RXL used 25% of total VRAM in this test which has the same default 1080p.
Topaz Video AI Alpha v6.1.1.1.a.trt
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 9900X 12-Core Processor 31.602 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 37.36 fps 2X: 20.22 fps 4X: 05.47 fps
Iris 1X: 36.40 fps 2X: 19.56 fps 4X: 05.87 fps
Proteus 1X: 31.24 fps 2X: 23.92 fps 4X: 06.53 fps
Gaia 1X: 11.28 fps 2X: 08.30 fps 4X: 05.75 fps
Nyx 1X: 10.47 fps 2X: 08.63 fps
Nyx Fast 1X: 25.34 fps
Rhea 4X: 04.77 fps
RXL 4X: 04.50 fps
Hyperion HDR 1X: 32.31 fps
4X Slowmo Apollo: 47.13 fps APFast: 77.89 fps Chronos: 28.54 fps CHFast: 41.06 fps
16X Slowmo Aion: 49.31 fps
1 Like
The amount of DRAM begins to matter when you work with eg. 4k to 8k upscaling or enhancing 8k video. However, it doesnāt use much.
At least not for me. Seem to differ between versions too. In v5.3.6 upscaling 4k to 8k use 5-6GB of VRAM and ~8GB of DRAM. Values might be dynamic and depend on how much the user has of one or the other; as in if you have low VRAM, the software wonāt use all of it, only up to a certain amount and then put the rest on DRAM and vice versa. Perhaps an employee at topaz knows how it works?
When enhancing input that is 8k (no upscaling, also v5.3.6) it consumes less VRAM than when upscaling 4k to 8k. However, enhancing 8k, no upscaling uses tons more DRAM. The ffmpeg process was using 14GB, while Video AI will use 4GB just having the input showing.
In Alpha v6.1.1.1.a.trt enhancing 8k input, the software and processing use very different amounts of vram and dram. In this version, it seems input preview consumes vram rather than dram, while processing 8k the ffmpeg process quickly increased to 20GB dram (then started reducing memory used, which might be due to reaching 90-95% of my 32GB capacity) and 4GB of VRAM, which instead is less than v5.3.6 used.
Interesting none the less.
Edit* Differs a lot between models of course.
Testing Rhea with 8k input currently. Itās consuming 24GB of memory (which is at 94% capacity currently) and 12.7GB of VRAM (however, 6GB of that might still be the app itself and the preview consuming 6GB, while the process itself consume ~6.5GB VRAM.
Pagefile got big. Currently 25GB. Typically, the pagefile goes unused on my system, which points towards video ai wanting to use more than the 24GB DRAM it currently uses processing 8k with rhea in the alpha version.
The 5080 is drawing 340W on average (out of 360W max, however, Iāve flattened curve after 1015mV and the GPU tend to use a couple of steps down from where voltage was flattened, so it currently sits at 995mV) and 0.2 fps processing speed, which probably comes down to it having to read data from the pagefile.
Interpolating 8K with Apollo (no enhancement model, just interpolation) will draw the full 360W of power at 960mV. Same amount VRAM and DRAM, but about 4GB smaller page file. Processing speed 2.2fps interpolating from 24fps to 60fps.
Thanks. I properly started tuning my memory a few weeks back. The amount of time it took just to find relevant information about all the different timings, what they do, which depend on which and general timings rules was ridicules. Combine that with some 100+ reboots to test different configurations, and every config needing to be tested for stability/timings compatibility and if voltages are too high/low. Luckily, thereās stability tests that catch errors quite fast. There have been double digits bsodās though, not going to lie.
Glad I put in the time though, seeing as tuned memory has a bgiger impact on CPU performance than CPU-tuning does when it comes to AM5.
As a comparison:
Tuned CPU with expo enabled: A 7-8% performance increase in CPU benchmarks.
Tuned CPU with manually tuned memory: A ~15% performance increase in CPU Benchmarks.
1 Like
Topaz Video AI v6.1.3
System Information
OS: Mac v15.04
CPU: Apple M2 Ultra 64 GB
GPU: Apple M2 Ultra 48 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 3840x2160
Benchmark Results
Artemis 1X: 03.89 fps 2X: 02.38 fps 4X: 00.55 fps
Iris 1X: 02.26 fps 2X: 01.17 fps 4X: 00.57 fps
Proteus 1X: 02.97 fps 2X: 02.55 fps 4X: 00.60 fps
Gaia 1X: 01.41 fps 2X: 00.97 fps 4X: 00.71 fps
Nyx 1X: 00.97 fps 2X: 00.95 fps
Nyx Fast 1X: 02.57 fps
Rhea 4X: 00.29 fps
RXL 4X: 00.30 fps
Hyperion HDR 1X: 19.74 fps
4X Slowmo Apollo: 04.71 fps APFast: 16.75 fps Chronos: 01.54 fps CHFast: 02.28 fps
16X Slowmo Aion: 06.02 fps
Topaz Video AI v6.1.3
System Information
OS: Windows v11.24
CPU: Intel(R) Core(TM) Ultra 9 285K 191.27 GB
GPU: NVIDIA GeForce RTX 5080 15.517 GB
GPU: Intel(R) Graphics 0.125 GB
Processing Settings
device: 0 vram: 0.84 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 24.70 fps 2X: 11.60 fps 4X: 03.06 fps
Iris 1X: 26.45 fps 2X: 13.53 fps 4X: 03.62 fps
Proteus 1X: 25.93 fps 2X: 14.22 fps 4X: 03.62 fps
Gaia 1X: 11.68 fps 2X: 07.73 fps 4X: 03.95 fps
Nyx 1X: 10.41 fps 2X: 08.18 fps
Nyx Fast 1X: 18.48 fps
Rhea 4X: 03.27 fps
RXL 4X: 03.33 fps
Hyperion HDR 1X: 19.57 fps
4X Slowmo Apollo: 22.89 fps APFast: 51.42 fps Chronos: 14.88 fps CHFast: 26.75 fps
16X Slowmo Aion: ERR fps
I donāt know if good but in general I am super underwhelmed when compared to my i9 13900K 128GB 4070 system (former system I just transitioned from) 
TVAI 6.1.X version doesnāt have the RTX 5000 fixes/optimizations and will under perform by up to 50%.
There is a TVAI beta 6.1.4.0.b which is supposed to have 5000 optimizations but they DONāT work in that version.
The ONLY version with proper RTX 5000 performance is alpha-6.1.1.1a
topaz-video-ai-alpha-6-1-1-1-a
Any visibility on an official release that would support the 285 with the 5080? And in the interim would you recommend that I try the alpha? Also do you guys have any plans to support NPU on my chipset? And finally does my benchmark indicate a failure \ crash on 16x?
I have been using the alpha the last weeks for different Proteus h265 tasks and it has been working ok for that. It doesnāt really seem to be more buggy than release version and especially beta 6.1.4. which is really a complete failure since neither RTX5000 nor older NVIDA hardware seems to work properly with it and there is absolutely no feedback about that for days.
That said, I expect a proper beta very soon due to 6.1.4 being so bad, so maybe waiting for that could be the best option.
Generally, with the 285, u want to have the fastest RAM u can get I think, especially if u have a lot of tasks which involve scaling.
1 Like
Thanks for the detailed advice, new build is maxed on RAM based on chipset and motherboard limitation: [4x Kingston DDR5-5600 48GB]
