Build a benchmark guide // list for buying video cards

Dear Topaz, other software house build a small benchmark (free) to test cpu and gpu power with their tool. I mean about cinebench from maxon to test cpu and Octanbench from Otoy to test GPU and understand speed of gpu with their 3d render engine.
Often is not clear when we (user) need more power the real difference between cards.
If you publish a benchmark we can start to build a simple dabase or a simple open spreadsheet where everyone that can test his videocard can send you the result.
This tool can help user to understand what video card (or computer macos) buy, and help you to do a small survey about who use what and understand better where you are going to develop for what kind of hardware.
both tool that i citated actually are the benchmark tool used on every website that do benchmark, and is a great adv for their developer.
please do something of that. people are not enought disciplined to do simple and drived test, and often desire to false it to demonstrate to have more power if you give a simple test scene.

3 Likes

I agree with you !
such a spreedsheet has been done in 2020 until VEAI 2.1, on the facebook VEAI page, but since then nobody updated it. it’s available here :

for what It’s worth and if someone want at topaz to create such a list here, here are some of my test using 2.6.4 normally (i’ll need to redo them to be sure, as some where done with 2.6.4 RC, Beta, or 2.6.3 beta…

certainly done with 2.6.4
1050ti GTX: Artemis Ahq/Alq/Amq/Proteus
720x396 16/9 → 1920 x 1080 = 1.21s/f
854x480 16/9 → 1080p = 0.27s/f
960x720 crop 4/3 → 1440 x 1080 = 1.30s/f
720p 16/9 → 1080p = 1.20s/f

Chronos Fast
1440x1080 32.715 → 29.97 = 0.95s/f

1080 GTX : Artemis Alq/amq/ahq/proteus
640x360 → 1080p = 0.56s/f
640x480 → 1440 x 1080 = 0.64s/f
720p 16/9 → 1080p = 0.48s/f

2.6.3 and before :
720p → 1080p
Intel HD Graphics 630 (i7-7700) = 2.7 s/frame
Intel UHD Graphics 770 (i7-12700) = 2.24s/f
1050ti GTX = 1.80s/f
1080 GTX = 0.57 s/f

1 Like

Here have another newer spread sheets created by @shikuteshi include V2.64 data :grinning:

1 Like

thanks, was not aware at all of this topic (rarely visit any other topic except the Beta one and the official last release, will add my numbers there :wink: Thx !

My quick results:

VEAI 2.6.4:
running on a 5600G iGPU with input video of 480x720, 200% scale
Proteus: 0.45 seconds per frame
Artemis Medium Quality: 0.44 seconds per frame
Chronos Fast V3 at 2x input FPS: 0.28 seconds per frame
Artemis Dehalo: 0.43 seconds per frame

Same video, same version, same scale but on an ASUS TUF 3070 ti OC
Proteus: 0.09 seconds per frame
Artemis Medium Quality: 0.07 seconds per frame
Chronos Fast V3 at 2x input FPS: 0.03 seconds per frame
Artemis Dehalo: 0.07 seconds per frame

stop guys, it’s exactly the problems about i talk, tell us some numbers is a waste of time.
the performance are related to many factors and we can keep the same for all:

  • all should use same video to test different model
  • all should use same parameters and same version of software
  • all should declare name, vram, kind vcard etc
  • all should tell us if video is on fast or slow support, where to write and what codec.

simple clarification : if you change codec or image sequence format change the performance of model, if you change video AI model need more or less time to recognize and process frames, if you change parameters or resolution obviously change the result.
too much infos… if there is a demo video / preset that we can start from icon or simple command line that veia process same video with all different models, different cleaning and upsampling and downsampling process, writing on same location, and write the result for each one, we can do a statists about power of different GPU, or… in other way is a waste of time.
to do that we should have a scientific and repetible approach.
may be that with 3.0 prototype we can do that more easely.

If you want consistent results, you need to make a post with clear concise requirements and a sample video to test with. Otherwise you’re going to get what’s above

i ask that Topaz do it, only them know exaclty how to test correctly their products with strong and weak point.
i can post a video clip and rules to test it, but i can do the wrong starting point, i ask a benchmark tool for this exact reason :smiley:

1 Like

I mostly depend on GamerNexus reviews for any gear guide However @shikuteshi spreadsheets is worth considering.
https://community.topazlabs.com/t/what-does-veai-performance-look-like-between-amd-nvidia/18581