Video AI v3.2.X - User Benchmarking Results

Fresh off the press from my base model mac studio ultra m2


Topaz Video AI v3.2.9

System Information

OS: Mac v13.04

CPU: Apple M2 Ultra 64 GB

GPU: Apple M2 Ultra 48 GB

Processing Settings

device: 0 vram: 1 instances: 1

Input Resolution: 1920x1080

Benchmark Results

Artemis 1X: 15.31 fps 2X: 08.17 fps 4X: 02.83 fps

Proteus 1X: 14.33 fps 2X: 07.80 fps 4X: 02.28 fps

Gaia 1X: 05.90 fps 2X: 03.95 fps 4X: 02.69 fps

4X Slowmo Apollo: 10.95 fps APFast: 37.62 fps Chronos: 05.87 fps CHFast: 07.55 fps

1 Like

Actually, an 8 core 5800x would give u almost the same performance as a 5950x. I usually only use the first 8 cores of my 7950x for tvai with a 4090. Your performance should be between 10 and 12fps on 2X, depending also on your RAM speed.

1 Like

Fastest CPUs that board will take are a bit faster than my 5800x, and thatā€™s a bit faster than yours. So your results worry me, as I am considering a GPU (+ PSU!) upgrade from my 3060 ti to a 4080 or 4090. But some of my benchmarks are similar to yours with your 4090:
,
System Model B550 AORUS PRO AC
Processor AMD Ryzen 7 5800X 8-Core Processor, 3801 Mhz, 8 Core(s), 16 Logical Processor(s)
Memory 32 GB DDR4 @ 3600 MHz

Topaz Video AI  v3.2.9
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 7 5800X 8-Core Processor               31.906 GB
GPU: NVIDIA GeForce RTX 3060 Ti  7.8496 GB
Processing Settings
device: -2 vram: 0.99 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	12.45 fps 	2X: 	08.22 fps 	4X: 	02.24 fps 	
Proteus		1X: 	12.85 fps 	2X: 	08.24 fps 	4X: 	02.16 fps 	
Gaia		1X: 	04.18 fps 	2X: 	02.89 fps 	4X: 	01.99 fps 	
4X Slowmo	Apollo: 17.69 fps 	APFast: 41.95 fps 	Chronos: 09.61 fps 	CHFast:	16.06 fps 	

As you can see, your Artemis and Proteus 2x and 4x results are similar or worse than mine, though other results are much faster. But overall, with what I use TVAI for, the upgrade from 3060 ti would not be worth any extra cost for me, let alone a 4080/90.

This all leaves me wondering if other factors are at play here - CPU throttling or whatever. Either way, I canā€™t see that a CPU upgrade for you or me on our present boards is going to make a big difference. Maybe the mobo/CPU and memory sides need addressing too? Sorry that I canā€™t help more.

EDIT - One last thoughtā€¦ have you checked for thermal throttling? That 4090 pushes out a lot of heat and if your case cooling canā€™t handle it, it (and/or the CPU) may throttle and cause performance loss. Just a thought.

1 Like

I was waiting for these results!! Thank you :slight_smile:

As can be seen here again is, that at 1X, the GPU is the biggest performance factor. At 2X and especially 4X, the CPU/RAM speed becomes more important (due to image scaling + color correction being performed on the CPU).

The most powerful GPUs cant be fully utilized by TVAI atm, even with the fastest CPU/RAM when scaling @1080p or higher is involved. Only possibility is to run 2 TVAI processes in parallel on a CPU with 16 cores since TVAI doesnt scale much beyond 8 CPU cores for a single process leaving spare CPU capacity for a parallel 2nd job.

1 Like

Hereā€™s my results. Iā€™m thinking of upgrading to Mac Studio M2 Max, Any advice from Mac users?
it looks like its nearly 3x faster than my current M1

Topaz Video AI  v3.3.0
System Information
OS: Mac v13.04
CPU: Apple M1  16 GB
GPU: Apple M1  10.667 GB
Processing Settings
device: 0 vram: 0.9 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	03.49 fps 	2X: 	02.29 fps 	4X: 	00.89 fps 	
Proteus		1X: 	03.49 fps 	2X: 	02.28 fps 	4X: 	00.74 fps 	
Gaia		1X: 	01.17 fps 	2X: 	00.86 fps 	4X: 	00.67 fps 	
4X Slowmo		Apollo: 	03.56 fps 	APFast: 	13.83 fps 	Chronos: 	01.14 fps 	CHFast: 	01.73 fps 	


Topaz Video AI v3.3.2
System Information
OS: Windows v10.22
CPU: 13th Gen Intel(R) Core(TM) i9-13900K 95.715 GB
GPU: Intel(R) UHD Graphics 770 0.125 GB
GPU: NVIDIA GeForce RTX 3090 23.77 GB
GPU: NVIDIA GeForce RTX 4090 22.096 GB
Processing Settings
device: 2 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis 	1X: 	34.86 fps 	2X: 17.51 fps 	4X: 04.98 fps
Proteus 	1X: 	29.24 fps 	2X: 15.80 fps 	4X: 04.79 fps
Gaia 		1X: 	13.54 fps 	2X: 09.42 fps 4X: 04.86 fps
4X Slowmo 	Apollo: 	43.73 fps 	APFast: 	96.17 fps 	Chronos: 	29.31 fps 	CHFast: 	36.24 fps

1 Like
Topaz Video AI  v3.3.0
System Information
OS: Windows v11.22
CPU: AMD Ryzen 9 7950X 16-Core Processor              31.737 GB
GPU: NVIDIA GeForce RTX 4080  15.688 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	27.16 fps 	2X: 	12.86 fps 	4X: 	03.30 fps 	
Proteus		1X: 	24.69 fps 	2X: 	11.43 fps 	4X: 	03.20 fps 	
Gaia		1X: 	10.61 fps 	2X: 	07.16 fps 	4X: 	03.39 fps 	
4X Slowmo		Apollo: 	37.00 fps 	APFast: 	57.18 fps 	Chronos: 	23.11 fps 	CHFast: 	30.86 fps 	

1 Like

Wow. When comparing to a I7-10700k 4080 configā€¦there is little to no noticable performance boost. Crazy.

This post is for V3.2 benchmark.
There is already a new topic for V3.3 benchmark.

Thanks for posting. I was waiting to see what the M2 Ultra could do as well.

Good review on Topaz here with M2 chips

1 Like

Thatā€™s because TVAI canā€™t really make good use of more than 8 cores, so the 7950X essentially performs like a 7800X (or 10700K) when running a single process.

The bigger cache of the 7950X is also not of much use because the TVAI tasks are more about streaming data (every frame is different from the previous one) rather than a longer operation on the same large data set, which is more common in games for example.

The advantage of a CPU with more than 8 cores, is that it will run 2 simultaneous TVAI processes faster than a CPU with fewer cores. Thatā€™s ofc when the GPU is not utilized to 100% by the first task, which is usually the case when upscaling is part of the process.

1 Like

Thx for your take on my thoughts and the analysis overall :raised_hands:. Now I do understand better and take it into consideration, when building my upcomming gaming/work(photo-video-edit)-machine.

A post was merged into an existing topic: Video AI v3.5.X - User Benchmarking Results