Media Encoder not using maximum GPU performance

[Stephen5C1C]

I hav i5 4570 16gb ram SSD and 3gb Gtx1060 Nvidia. When i activate mercury acceleration with cuda the render speeds dont even improve and the GPU will be operating at 5% o less the processor and ram will be less tha 80%. A 30sec project with basic 3D layers and camera depth of field takes 50minutes to render.

I have tried changing program setting in nvidia control panel for the app to use maximum performance and overclocked the card bt nothing changed.

What could be the problem?

[meylorod]

Any luck with this issue? I am having similar problems. My CPU is at 100%, while the GPU is at 6% usage while rendering. 

[Christopher Shipe]

I am not 100% sure, but I think the software of Media Encoder cannot maximize the GPU since it has to render frame by frame in certain scenarios. The reason it may be limited to frame by frame is because FRAME 2 may require the information of FRAME 1 to be finished before it can start processing/rendering itself (Unlike some projects where multiple frames can be rendered at once which gives you a faster export).

An example would be like eased keyframes, to render a solid that is keyframed (eased) from one position to the next, I think the software can only make calculations about the solid's position in FRAME 2 based on where the solid was in FRAME 1.

I am not sure if you are even using eased keyframes within your project, or if this limit just applies to keyframes in general but I remember reading somewhere that easing keyframes can affect export speed and I believe this was the reason why.  Hopefully there is some correlation haha, sorry for the long explanation if not.

Long story short, it's a limit on the software rather than hardware... that's what I've got from what I've read.

[RjL190365]

Here's the issue:

The rendering performance is only as good as the weakest link, in this instance. The CPU is very underpowered by current standards. And when a CPU is too weak, the GPU will just sit idle, waiting for the CPU to play catch-up. In fact, I would dare say that the GTX 1060, weak (by current standards) as it is, is still too much GPU for that 4-core/4-thread CPU.

[meylorod]

I have an 8 core CPU, not a 4 core. Also, I installed Cuda Toolkit and that changed the balance of the CPU/GPU render percentages to 55% GPU and 20% CPU. There's still a lot of room for processor usage, especially in the CPU.

[RjL190365]

My reply was to the original poster. That i5-4570 is only a 4-core CPU without multithreading capability.

Even with heavy GPU effects, the GPU still sits nearly idle while the CPU gets slammed in that system. Exactly the opposite of what is happening in your PC.

And under normal circumstances, if the GPU usage is too high for even a simple frame rate conversion while the CPU usage remains low, that is a sign that the GPU is too weak for the CPU.

[Invix]

I've got a 3070 ti and an i9 12900k, and both max at 40% usage without any other software open in the foreground. It uses CUDA for rendering. But why isn't it maxing out and blazing through the render en though it's got the space to do so?

[Robert329276361yzp]

This is the same problem I am having at the moment. Just to *ensure* data rates werent and issue I changed the render destination from the source. My source is a 8x 2TB SSD RAID5 on a dedicated RAID controller (about 1.5GBps write, 2GBps read) and the destination is 2x 1TB M.2 RAID0 (about 3GBps W/R). These are separate from my OS which is on 2x 1TB M.2 RAID1. **Yes, I have 4x M.2 slots on my MoBo, with 5 remaining SATA6 ports available (4x 10TB HDD RAID5, 1x DVD R/W).

I'm rocking a 12 Gen i9-12900K, 2x 16GB DDR5 @6000MHz, and a RTX3060.

I'm transcoding 4K Sony MXF to 1080 ProRes Proxy. And like... there's no clear throttle point besides the software just not being able to utelize the system? My GPU is sitting derpy at 15%, CPU hanging out at 25% or less, my drives aren't even breaking 2% usage. 

Make it make sense? The only other thing I can think of to do at the moment is run a performance test to verify it isn't system wide / something wrong with the computer itself that I need to track down. 

[Warren Heaton]

Transcoding camera originals to ProRes is CPU based.  I would expect to see very little GPU usage for that.