@reiner is correct, the benchmarks are great but no GPU benchmarks for VEAI, only CPU. To be honest not sure how many are relying on CPU processing for VEAI here, although it’s strange to see a production app where Ryzen 5900X is beaten by a 10900K.
Hopefully someone will have a GPU tests for 6800/6900s and the 3XXX RTX GPUs
AMD R9 5950X 16CORE, 32G DDR4 3600MHZ, RTX 3070 8G, STRIX X570-E GAMING, SSD 980 PRO 500G, ALL PCIE4.0
THIS IS THE TEST WITHOUT OVERCLOCK.
0.09SEC/FRAME 720X480 TO 1920X1080
Ryzen 3950X
128GB 3200
Radeon 6900 XT
1TB 970 Pro
Gaia-CG v5 200%: 0.25s / frame
Theia-Fidelity-v4 200%: 0.28s / frame
I don’t mean to be pushy, but where is this chart of users experience.
Can we change the format of a thread like this to an XML chart locked in on top and a form for people to submit their experience with minimum info and would be nice fields?
Then we could move to pass this loosy goosy info and discuss realistic expectations, time-saving tips, how to improve our systems, and even maybe what to do with our new found awesome pictures and video.
q734227051-8940
In my two systems：
10900K
720P to 2K(200%) Artemis v9
RTX2080:0.22~0.25sec/frame
RTX3070:0.22~0.24sec/frame
…
10700F
720P to 2K(200%) Artemis v9
RTX3070:0.34~0.36sec/frame
RTX3080:0.33~0.36sec/frame
So what’s the point of upgrading GPU?
my system is running AMD 3900x
RTX1060 don’t use it for VEAI
Quadro P4000 0.17~0.21sec/frame
My card is half the price of yours and meant for commercial use. The turing version of Nvidia commercial cards start at $2000, but the older Pascal versions like mine you can buy for $500 apiece. I don’t know if you are using this software for personal or commercial, but those cards right now in this case are more useful for gaming than for commercial uses like video editing.
If I can get it to work right I should be able to run two instances per card. I will figure out why it’s not working correctly.
Just sharing some testing with Handbrake.
the original file was 534MB then 1.2GB after cleanup with VEAI
Compressed it with Handbrake using H.265 Nvidia (Meaning Graphics card) and the preset to default and everything else auto. I got these results:
quality 18 1GB
quality 20 837MB
quality 22 647MB
quality 24 486MB
quality 26 384MB
quality 30 246MB(at this setting I started noticing the quality of the video being watchable, but hair strands weren’t as defined as before)
I find this very encouraging myself. I have Blueray rips for some movies I couldn’t get the file size down much at all. Plus I recently decided to double the size of my TV shows that are 576P. Was worried about how big the final file size will be, but not anymore.
Of course it’s okay to play games occasionally but it’s obviously prepared specifically for editing and rendering videos. Why not choose a professional graphics card because it is not necessary. Although professional graphics cards are supported by better materials and more professional software. they are almost useless in daily life unless they are used by professional design companies. Why choose 16-core CPU or more core CPU? Video editing and rendering can not rely on single-core processing alone. Multi-core means faster processing speed. There are many professional video production people who have done many tests and there are many related videos on YOUTUBE.
I have tested that this computer can process 5 VEAIs at the same time, 480 upscaled to 2K, the temperature of the CPU and graphics card is 65 degrees. The load of CPU and graphics card is about 80%. The computer does not have any freezes and you can watch videos online at the same time. It takes 40 hours to process 5 two-hour movies simultaneously. On average each movie takes 8 hours. Next time I will try to run more VEAI at the same time.
Sweet, it seems you have done lots of testing and got this down. When I brought up multiple VEAI and got them running, before I jacked up my computer, they didn’t use any more of the GPU. Instead, they ran slower. Instead of a constant 0.19 spf, it got up to 0.26 spf.
Is there something I need to change for more than 45% of my GPU being used by VEAI?
I am at nearing my wit’s end to repair my computer, so if I reinstall Windows I want it set up better to be doing this stuff because the rest of this year is going to be doing this. I don’t know about next year yet, but this year is all about me getting plex going that works even offline and getting 200 plus movies and shows watchable if not upscaled to be looking good on 4K TVs even one day.
PS Does anyone have an HDR TV and how does it look with this upscaling?
You got the lastest and greatest from AMD. Are thoses Frames Pre Second a good rate?
I wonder if this software prefers NVidia cards?
When I replace me GTX 1060 card something like this would be a consideration. I found rumors that even computer games where only doing the minimum amount of work to make AMD work. I hope with this latest gen of cards that attitude changes.
S/F will slow down when running multiple VEAIs at the same time. This is also determined by the number of CPU cores. Of course, if conditions permit, I will use more core CPUs with dual graphics cards to crossfire. But this still requires a lot of investment to upgrade the hardware. Recently INTEL announced that it will release a new version of DIRECTML. This will also increase the speed of VEAI. This is good news for using any graphics card that supports DIRECTML.
Too bad no Navi 1 card on the list.
The W5700 is very cheap for a Workstation Card these days.
But the Pro Navi 2 cards are also coming, but didnt hear any rumors about it.
2x RTX 5000, TR 3000, 24 Core, 128 GB ECC.
Win 10
All standard settings.
output Tiff 16 bit.
1.9.0
Gaia CG V5. 0.23 sec. image.
Gaia HQ V5. 0.23 sec image.
Artemis HQ - (Does not work with Dual GPUs atm) —> Beta V11 → 0.19 sec. image
Artemis MQ V10 - 0.19 - 0.20 sec. image
Artemis LQ V10 - 0.18 - 0.19 sec. image
Artemis AA V9. 0.19 - 0.21 sec. image
Dione TD v1. 0.19 - 0.20 sec. image
Theia Detail V3. 0.24 sec. image
Theia Fidelity V3. 0.24 sec. image
#Note: The curve is already flattening. Buying two small inexpensive gpus with 8 gb each, pcie 4 and fp16 around 20 TF is a good decision i think.
Like 6700 XT or 6800, when prices come down.
Updated Dual GPU numbers look to post above.
Are you still looking for numbers here or is this old post?
I gave a quick test of two of them on latest VEIA v2.0.0 on ASUS Hero VIII X570 system with Aorus RTX3070, R9 5950X CPU, 32GB 3400/CL14 RAM and got below:
Artemis HQ v11
0.19-0.20 sec/frame
Artemis LQ v11
0.19-0.20 sec/frame
Gaia CG v5
0.29-0.35 sec/frame
To be honest I wish they improved the program to give “final” numbers for timings after processing, instead of users just guessing numbers that are constantly changing. Can’t even see a total time taken amount after processing which most programs have for long functions.
I’m surprised so many CPU cores are active and used during these GPU runs actually - the 5950X is about 30% used and many cores being used while processing.
I’m not sure that’s best test video is it? Would it not be better to use something that changes scene instead of doing same thing entire time?
Hi Geogan,
If you place the cursor (or arrow) over each video after batch processing it will display exactly how many seconds each video took to render. Hope that helps.
Cheers,
Paul
Oh I didn’t realise that. Cheers!
sharing my benchmarks:
CPU: Ryzen 1800x
RA: 32GB DDR4 3200MHz B-Die
GTX 1080Ti
RAM: 32GB 3200MHz
VEAI Version: v2.0.0
Gaia High Quality 0.5: 0.76 - 0.91 sec./frame
Gaia CG v5: 0.76 - 0.93 sec./frame
Theia Fidelity v4: 0.43 - 0.51 sec./frame
Artemis HQ v11: 0.32 - 0.38 sec./frame
Artemis AA v9: 0.32 - 0.37 sec./frame
Artemis LQ v11: 0.32 - 0.37 sec./frame
RTX 3080:
Gaia HQ v5: 0,29 - 0,33 sec/f
Theia Fidelity v4: 0,28 - 0,32 sec/f
Artemis LQ v11: 0,23 - 0,25 sec/f
Can you please try to upscale to 4K? I would like to see RTX 3080 performance with GaiaHQ and ArtemisHQ.
I haven’t checked back here in awhile since I forgot about it. Only just now remembered after getting a new rig and seeing that what appears to be the latest Nvidia drivers causing Ampere not to work in VEAI.
Specs:
Ryzen 5950x
RTX 3090
32GB 2133Mhz RAM (I’m waiting for the proper kit to come in so this was pulled from my old rig. I’ll update again if I remember to do so.)
VEAI 2.1.1
Nvidia Studio Drivers 461.92 (Same as Game Drivers. Just not the latest Game Driver version as it doesn’t work)
Umm…
0.21 - 0.26 seconds per frame across the board.
Every single model got around this range. And I noticed that the GPU clock just wouldn’t stick to max which is probably why there’s such a range.
I wasn’t until I tried to upscale a 1080p video to 4K did I see very slight performance differences.
Artemis ~0.35 seconds a frame
Gaia/Theia ~0.40 - 0.45 seconds a frame
Now while it seems like the low resolution video is the cause and these results are may now be deprecated. Allow me to first point out that within Task Manager, looking at GPU usage under Graphics_1 showed no more than 35% usage at best. No other metric that Task Manager can monitor changes under VEAI workloads.
And opening a second instance of VEAI doesn’t increase the usage. Only halves the speed of each instance of VEAI so you’re neither going any faster or any slower.
This seems to all come back around to Topaz not yet being able to make use of all the Tensor Cores within TensorFlow as of this time.
Looking back on the 720p 2x upscale Task Managers only reports 22% usage. So there’s definitely an unoptimized maximum cap on the performance you can get out of VEAI as of this time.
As a result I’m personally going to come to the conclusion that as of this moment no further testing is needed unless one is curious about what the current performance you can expect out of a given hardware is.
Though there’s also a good chance that memory speed is creating an artificial cap and I’ll have to retest these scenarios after the new kit comes in. Assuming my local post office hasn’t actually lost it. (it’s 5 days late since ‘arriving at my local post office’ >.> most likely missing since I’m getting every other package from them)
