Video Enhance AI not using all of computer


RTX 2070

When I use topaz, it doesn’t seem to use all of my PC. I am upscaling a dvd rip to 2x res, using the “Theia Fine Tune Detail” option.

It only uses about 40% of my GPU, and 60-70% of my CPU.

This isn’t just a faulty reading from the program I’m using (MSI Afterburner), as if I start 2 at once, they are not twice as slow, about 30% slower.

Is this normal? It doesn’t really matter than much, I can just run 2 at once, I’m just curious if it’s a issue with my PC or something.

no, it’s is not your pc.

It does for some reason NOT use the full power of the gpu.
I thought it was just the “new” rtx 30x0 series, but your rtx2070 experiences show it is not.

The only solution to use your gpu to the max, is running 2 or 3 parallel instances at once.
As long as you other hardware (cpu/ram/harddisks) is capable.
Thats the recommendation of the veai developers.

We all have this problem…

my rtx3090 never hits the 50% on a single VEAI instance
Only when i run 3 instances of VEAI parallel i see a gpu load of 90%+
So i suggest you to that also, coz that really saves rendering times

for example:

one by one

instance = 1h rendering
instance = 1h rendering
instance = 1h rendering
1+2+3 one by one = 3h


instance = 2 h
instance = 2 h
instance = 2 h

1-2-3 parallel = 2h (combined) = 1h saved (33,33%)

but keep an eye on your queue lists (to be rendered videos)
coz one instance overwrites the queue (save.json) of the other instances.


I have dual RTX 3090 and VEAI always runs at 100% all the time. Power consumption is always at max.

Well that is IMHO impossible.

Since even the developers of veai say we need more instances to get 100%
There are many threads on this forum approving my experiences, that the nvidia gpus are not used to the max. How do you check that yours are used by 100% ?

I use the nvidia experience Alt-Z to check the system and gpu load.
And what i see there (one instance right now at 39-47% GPU Load )
is approved by GPU-Z and other systemtools like SIV64X right now and before.

They didn’t say you have to run multiple instances to get 100% GPU usage. They said that you need to run multiple instances to use all 24GB VRAM of RTX 3090.
The GPU usage is wrong also. When CUDA, Tensor cores are being used, the GPU usage is no longer accurate. Just like Adobe uses CUDA cores. But in real time, it is using 100% the GPU. The more instances you run, the slower VEAI becomes. So running multiple instances won’t speed up the processes. The timer is just broken when multiple instances are running. I calculated before to see if multiple instances helped. But it didn’t work that way. So my recommendation is only run 1 instance 1 footage at the time. The only way to speed up the process is using more than 1 GPU, or using TIF image sequence.


I dont know how you come to your conclusions, but i fear in this case you are wrong.

And even with 3 instances … i never got more than 12 gb vram usage.
You say the nvidia tools are wrong, the gpu usage is wrong, the timer is broken, but you are the only one who is right? sorry, i am testing and verifying this for month now.
And i believe you are wishfully thinking.

But in one point i agree.
Using 2 GPUs will speed up the process, as long as you run at least 2 instances…

GPU usage has nothing to do with VRAM usage, tensor cores or CUDA cores. Playing a game with 100% usage with 5GB of VRAM is possible. Sorry to say but your time with VEAI is a waste.

Also, I’m running both GPU at 30% and 0% usage with 350W power draw each GPU. Tell me why my GPU has 0% but draws 350W? See the problem here?

1 Like

I never said that gpu load has anything to do with vram.

Well, you have been the one complaining more than one time that the new beta works slower than the last release. And many others did not confirm that. Including me. I told you then and now you have either no idea what you are talking about or your system is configured wrong/broken.

You are not worth any further answer

Topaz team has already stated that VEAI is unable to utilize the full amount of resources of an RTX card and we won’t see dramatic performance for months while the continue to optimize it. This is one of the main reason why they recommend multiple instances if you have the headroom.

Hardware usage between the CPU and GPU are entirely based upon what model is being used and at what scaling you’re running at. But generally an RTX 30 card should never be around 100% usage unless you’re running 2-3 instances at 400% scaling. If it’s anything less than that then you’re reading the incorrect measurement.


This is just false, VEAI doesn’t always use Tensor Cores or CUDA for that matter, since it can also be run with Radeon cards. I ,for example, use RX5700XT to upscale content daily and that gpu has no CUDA cores whatsoever or tensor cores for that matter.

Secondly, it doesn’t use 100% gpu, period. This can clearly be seen if you install any third party monitoring software (like Afterburner from MSI) and monitor the clockspeed of your gpu /gpus. Clockspeed / clockspeeds will bounce up and down, since the load is not constant.

Instead of CUDA cores, AMD called it Stream processors, they are all the same, serve the same purposes. Without it, you won’t be able to encode or decode video. Don’t spread mislead information.

VEAI used to use CUDA cores to run. AMD GPUs were not supported until 2020. But then, VEAI decided to use fp16 with tensor cores for better quality and performance on newer GPUs, led to slower performance on low-end GPUs. Recently, VEAI seems to add fp32 which has even better precision than fp16, but at cost of performance which will be reduced in half. But Ampere GPUs were released with fp16=fp32 that might help increasing quality at the same performance.

VEAI uses 100% or not, depends on the models. Power goes up and down, voltage goes up and down for each frame. After finished 1 frame, power goes down, then goes up for the next frame. It WON’T be consitent. With 1080p to UHD or 8K, VEAI uses almost 400W of power. As you see in the Task Manager only shows 3D related things (like gaming). For tasks like machine learning, only tensor cores/CUDA cores or Stream processors are being used. Not all GPU components are forced or required to run. Again, this is not gaming where you can see 100% GPU usage. Unless you push VEAI heavily with 8K upscaling or so (but still, not all components are used). Also, not all models use 100% GPU due to CPU/SSD/RAM bottlenecks. GPUs are created mostly for gaming, it works 100% its power for gaming, but not rendering tasks or machine learning workloads. Most components are designed for gaming. But render tasks or machine learning tasks are NOT gaming at all. It took me 2 seconds to understand that, but people still refuse to spend 2 seconds for that. Google shows everything about machine learning. Don’t just let the machines learn itself and be smarter than you.

Still, people just don’t look up things on the internet, and believe that they know everything.


They stated that months ago and it’s no longer true based on other’s testings.

You are the only one saying so and so you are the only one living in your little ego world …

It was and is nonsense … no matter how often you repeat it

1 Like

I just have a lowly 1080 Ti still. But GPU is constantly at 99-100%, aka, fully saturated. Vram is a different matter. I had set the slider to use max memory (allegedly using max over 10G), but GPU-Z consistently puts vram usage at ca. 2.5G, and never above.

Thanks very much for your input here, where it matches my experiences with VEAI.

I was puzzled when I discovered that my GPU usage was only around 40% on average when processing a series of videos with my NVidia Quadro RTX 5000, where for months I just assumed it was 100%, until I checked a few weeks ago.

Each of those videos takes around 25 to 30 minutes to process.

So to see if I could push the utilization higher, I ran another instance of VEAI and the average utilization doubled to around 75% to 80%, and yet each video still took around 25 to 30 minutes to process, proving the reported utilization is correct.

Adding a third instance pushed the utilization to 100%, where this time each video took around 30 to 35 minutes (so I assume about 33% utilization for each in parallel).

I wish I knew about this before because it would have saved me a lot of time by processing the videos in parallel, instead of one at a time as I’d been doing before.