When I set the encoder to H264 (NVIDIA) of H265 (NVIDIA) the video fails to be processed and “Driver error” is displayed. I suspect this is down to my having a relatively old Nvidia driver installed on my Windows 10 system, i.e. v456.71 and hoping it’s not down to my GTX 1070 video card no longer being supported.
Can someone please enlighten me as to what the MINIMIUM Nvidia driver version I need to have installed for Video AI 3.0.0.
My current driver and setup works fine with Video Enhance AI.
Thanks @ForSerious. From what I read 472.12 and 512.95 are the most stable recent releases so I installed 472.12 and found that this was enough to get H264/5 encodes working on Video AI 3.0.
I’ve found that the encoding speed is about 20% slower with Video AI 3.0 compared to Video Enhance 2.6.4. That said, the encoding speed is extremely slow at around 1 fps or 1 second per frame. My GPU usage during encoding is minimal (max 5%) even though I have my GPU selected in the settings. I have to wonder if the GTX 1070 is supported for GPU encoding.
@ForSerious - what encoding speed and GPU usage do you get?
What model are you using, what is the source resolution and the target resolution?
It can actually take longer to upscale a 1080p source to 4K than a 480p or 720p original because there are more pixels to process.
FWIW, I am upscaling DVD rips of Star Trek Deep Space Nine. These have always been tricky to upscale because the live action is 24fps, but the CGI is 30fps. In TVEAI 2.6 it took about 2 hours to upscale a 45 minute SD original, but produced an output that stuttered frequently. TVAI 3.0 takes closer to 3 hours, but produces a perfect VFR output. BTW, running two upscales at once took 5 hours, not 6 and running 3 at once took 7.5, not 9 - so if you have multiple files to process there is a definite advantage to running 3 at once.
My rig is an Intel i9-12900 with a Nvidia 3080Ti (AIO cooled). Using Proteus I push GPU to about 50% to 60% on average, when running 3 upscales - I also pull about 600watts, according to my UPS.
Just noticed that you have a GTX-1070. That doesn’t have any Tensor cores for AI processing, so all it can be used for is floating point calculations and basic DirectX functions like compression and muxing the output. Everything else is on your CPU.
Thanks for highlighting the need for Tensor cores. I’m pretty new at this so still learning. So, I guess that means I have to suck up the slow processing or upgrade to a different GPU. Good to know.
A quick test on Video Enhance AI 2.6.4 gives me 0.8 sec/frame using my GPU, but 7.8 sec/frame using only my CPU. So even with my current GTX-1070 GPU I get an enormous speed boost so that’s good.
BTW - currently I’m processing some old 8mm cine footage that has been captured at 1440x1080p 24fps. I’m putting it through Artemis Low Quality to denoise it and add black sidebars to output 1920x1080p.
I just upgraded my driver earlier today. I believe that the minimum is 520, in the popup. This is not available from Nvidia. If you download the driver from Nvidia today, it will fix the problem. If you cannot use the driver, just download the latest version.
FWIW, I updated my driver today to the latest version and 3.0.1 still says my driver is too old. FFMpeg also keeps screwing the pooch, so basically I’m stuck on 2.6 on my windows machine, and can only use 3.x on my M1 Mac, which, you know, ain’t great.