Is NVENC better than 6 cores?

[nick_m]

Hello,

I'm running HandBrake 1.2.1 on a Win7 machine with an i7-3930k processor and a Quadro 2000D. The CPU does not have QuickSync, so I was thinking of upgrading to Win10 and swapping the GPU for an NVENC-compatible card for better performance. Would NVENC provide a significant performance boost over a 6-core CPU? It's currently stock, but I was thinking of overclocking it. If swapping cards is a better choice, would just any GTX 1050/ti/1060 do?

Thanks,
Nick

[s55]

First point, while NVEnc is typically faster (usually a lot faster than that kind of CPU), it can be heavily bottlenecked by a slow CPU since decoding, filters, audio etc are still all on the CPU. I would be surprised if your CPU doesn't slow an NVenc encoder down a fair bit.

Second point I'd make is you'd also be trading quality and/or filesize in order to gain that extra speed. If that trade off is something your already happy to make, you might be better of using a faster encoding preset on the video tab for x264/5 as that can make substantial differences in speed and likely still produce better results than NVEnc.

Final point, for best results, a lower end RTX card is where your going to get the best NVEnc experience. (i.e a 2060 or 1660)

[nick_m]

Thanks, s55.

I have an i9-10990X with a Quadro RTX4000 coming, so I'm just trying to determine what to do with the old box.

I would be surprised if your CPU doesn't slow an NVenc encoder down a fair bit.

It makes sense that my system would perform only to the level of its least-capable component if everything happened in parallel. However, the docs state (emphasis mine) "Every stage prior to and after video encoding including decoding, filters, audio/video sync, audio encoding, muxing, etc., is performed by the CPU." That lead me to believe the encoding was a discrete step in a serial-process using the GPU. They then go on to state "As a result, it is normal to have high (even 100%) CPU utilisation during encodes." There's an apparent contradiction that I can't reconcile: that the CPU-based processes going on before and after encoding cause high CPU usage during NVENC-based encoding. Can you shed any light?

you'd also be trading quality and/or filesize in order to gain that extra speed.

Is there something inherent in NVENC encoding that causes a loss of quality? All other things (settings) being equal, wouldn't NVENC simply reduce overall processing time? Since it's not enabled on my system, I can't see whether it's an "either/or" setting relative to other encoding options. I'm still trying to determine what settings work best and it would be nice for the trial-and-error to be somewhat faster.

a lower end RTX card is where your going to get the best NVEnc experience. (i.e a 2060 or 1660)

I see reasonably priced RTX 2060 and GTX 1660 cards, but nothing like an RTX 1660; was that a typo? It's worth replacing my Quadro 2000D anyway, so it might be worth running the experiment for the additional cost of a Win10 license.

One last question: many Blu-ray players advertise up-scaling of DVDs to 1080p; is that something HandBrake can do effectively? For some reason I haven't been able to sort out, playing a DVD on my up-scaling DVD and Blu-ray players looks like crap after I upgraded from a 720 to a 1080 projector.

Thanks for all the help,
Nick

[s55]

It makes sense that my system would perform only to the level of its least-capable component if everything happened in parallel. However, the docs state (emphasis mine) "Every stage prior to and after video encoding including decoding, filters, audio/video sync, audio encoding, muxing, etc., is performed by the CPU." That lead me to believe the encoding was a discrete step in a serial-process using the GPU. They then go on to state "As a result, it is normal to have high (even 100%) CPU utilisation during encodes." There's an apparent contradiction that I can't reconcile: that the CPU-based processes going on before and after encoding cause high CPU usage during NVENC-based encoding. Can you shed any light?

It's a pipeline, so Decode -> Filters -> Encode. All happening at the same time. While a frame is encoding, others are being decoded and filtered.
As such, if the app can't decode or process though filters frames fast enough, then the encoder will just wait for the frames to become available.

Is there something inherent in NVENC encoding that causes a loss of quality? All other things (settings) being equal, wouldn't NVENC simply reduce overall processing time? Since it's not enabled on my system, I can't see whether it's an "either/or" setting relative to other encoding options. I'm still trying to determine what settings work best and it would be nice for the trial-and-error to be somewhat faster.

It's fixed function hardware. Not an implementation of x264/5 which are far more advanced encoders which far more clever options that improve compression efficiency and quality.

I see reasonably priced RTX 2060 and GTX 1660 cards, but nothing like an RTX 1660; was that a typo? It's worth replacing my Quadro 2000D anyway, so it might be worth running the experiment for the additional cost of a Win10 license.

1660 is a GTX named card, with the RTX encoder. the 1650 has the old pascal encoder that the older 10-series cards had.

One last question: many Blu-ray players advertise up-scaling of DVDs to 1080p; is that something HandBrake can do effectively? For some reason I haven't been able to sort out, playing a DVD on my up-scaling DVD and Blu-ray players looks like crap after I upgraded from a 720 to a 1080 projector.

You should avoid upscaling files with HandBrake at all costs (It can only be done in the CLI. The GUI blocks it at present)
Your basically bloating the filesize for no good reason. Typically, letting your TV/Monitor handle the upscaling will yield much better results as they should have an optimised upscaler built-in for that specific screen.

[nick_m]

Thanks for everything, s55!

[s55]

Either way, try it out on the RTX4000 and you can make a judgement wither you find the results acceptable or not. There is a level of subjective-ness in these things anyway.