GPU - Erratic Performance

I have been struggling with GPU encoding for a few weeks.

I am running Windows 10, using a GTX 1050.

I found this thread that says to use driver version 376.51. When I use that driver, the GPU encoding does not work at all.

I have tried different versions of the driver, currently I am using the CUDA driver from Cuda toolkit version 11.4.0_471. Not sure how to find what the driver version is.

Here is where things get weird. The GPU encoding WORKS, but the performance is drastically different between different encodings.

I am converting videos recorded from an HDHomerun.

A 1 hour recording takes about 16 minutes. During that time the GPU usage will be about 60%. I think this is great performance.

Other times a 1 hour recording from the same source, with the same MCEBuddy profile/settings can take 1.5 - 2 hours. During that time the GPU usage will be about 7% - 9%.

The GPU is doing something, just vastly different in terms of performance.

Attached is a log, converting a 1 hour show in 16 minutes.

fast.txt (14.2 KB)

Here is the conversion log of another 1 hour show, it takes 90 minutes. The logs show it is using GPU encoding, so what causes 1 video work convert in 16 minutes and another video almost 6 times a long?

slow.txt (14.0 KB)

I have been unable to figure out what causes the performance difference. I have tried rebooting between encodings, restarting MCEBuddy - nothing is consistent.

It looks like your logs are set to INFO so it doesn’t contain the details required to see what’s going on (default is debug).

Different profiles use different features and filters. Not all hardware support all features and filters, so if a profile is running slower despite using hardware acceleration it’s likely due to the hardware not supporting filters or features in that profile and the CPU having to take over which slows it down.

Also if you’re running multiple conversions or games or GPU intensive activities at the Sam time it may impact the GPU conversion performance.