HEVC encoding slower on new GPU

[Ron30]

I use a software called VideoProc Converter (their website looks scammy as hell, but it's actually proven to be useful over the last few years) to convert a large number of personal old video files into x265/HEVC files. This saves me a ton of storage space and as the videos are fairly old and not the highest quality to begin with, I don't notice any drop in picture quality. I recently upgraded from a GTX 1080 to an RTX 4070 Ti and decided to compare the conversion speed between them. I converted a video to HEVC mp4 using my 1080 and timed how long it took. I then swapped in the new 4070Ti and attempted to convert the same file, with the exact same settings and found that it took about 25-30% longer. I repeated the test several times and made sure I had the latest NVIDIA drivers installed. The software has detected both GPU's correctly and does have hardware acceleration enabled.

 

I'm trying to understand why a more powerful card, is performing the workload slower. Per https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new, there's fewer NVENC chips on the 4070 Ti, As I understand it, this just means I can't encode more than 1 stream simultaneously, which is fine as I'm only converting 1 file at a time. Can anyone shed some light on what's happening?

 

My specs are:

AMD 5900X

Gigabyte X570 Aorus Pro Wifi

64GB 3200MHz RAM

800W PSU

Samsung 970 Nvme running at PCIe 4x

[dragonhart6505]

Well firstly have you contacted their support? The site says they support up to RTX 3080…if I recall, didn’t 40 series do away with some popular older encoding variant that perhaps the program requires for hardware acceleration?

[Ron30]

Yes, I did reach out to their support. They took some screenshots of settings and some log files and said they've "passed it onto the engineering team and will let me know if/when they find anything". To be honest, it sounded like they were just placating me, which is why I came here.

 

I hadn't noticed that the site states GT630 up to RTX 3080. Nor did I know that the 40 series dropped some encoding features. I assumed that "newer must be faster", but this could certainly account for it. Thanks for the info.

[Ron30]

UPDATE: Just to provide a follow up on this topic, I tested encoding with several different apps to see if I could find a better solution. I tried my best to match the settings between each. For each, I tried to use the mid-lower quality setting, or when given a quality number to be set, I selected 36. The default was 24 and the lowest was 51. I've no idea what this number means, but using 36 in each application that offered this setting, I would assume meant that I was comparing apples to apples. 36 also looked to be about the same place on the quality slider, as the setting I would use for the apps that had a slider with no number.

 

Original file: 6.75GB

 

VideoProc

File size: 2.35GB

Time to encode: 22m 34s

GPU utilization: 1-4% with very rare spikes up to 12%

 

WinXDVD (same developer as VideoProc)

File size: 2.41GB

Time to encode: 18m 07s

GPU utilization: 1-4% with very rare spikes up to 12%

 

AVIDemux

File size: N/A. Aborted after a couple of mins as ETA was nearly an hour

Time to encode: N/A. Aborted after a couple of mins as ETA was nearly an hour

GPU utilization: 1-3%

 

Freemake Video Converter:

N/A - did not have a HEVC / x265 option

 

Handbrake (H.265 (x265) encoder)

File size: N/A. Aborted after a couple of mins as ETA was over 35 mins

Time to encode: N/A. Aborted after a couple of mins as ETA was over 35 mins

GPU utilization: 1-4%

 

Handbrake (H.265 (NVEnc) encoder)

File size: 1.20GB

Time to encode: 9m 44s

GPU utilization: 11-14%

Quality: I think it's visually lower than the usuall VideoProc encodes I do, but I was pausing and zooming and trying REALLY hard to see the difference. If I wasn't looking for it, I wouldn't notice it, and I'm not even certain there is a visual difference.

 

Update2: The lower the number, the higher the quality. I found that 29 is the sweet spot for me, for quality vs file size.