Complete HT/HTPC/Encoding overhaul - x265 on GPU Advice appreciated

[Allan74]

I apologize in advance for what may become a Novella of sorts.....

I have been running the same setup at home for more years than I would like to admit.
All x264/1080p 'friendly' to suit my 46" LED TV, wife's iPad, Dual Core Win7 based HTPC and misc devices, including my back end Workstation Hardware.

I use Handbrake/VidCoder when I need to, as well as FFMPEG for all audio, regardless of transcoding acquired files, as I have learned quickly after multiple Terabytes and thousands of titles that sacrificing a DTS soundtrack for AC3 is well worth the 1GB/Title space savings for what I am doing (and not willing to do to continually upgrade server storage).
I currently store everything on my Plex Server @ 1080p/8192kbps/AC3.

I want to do a COMPLETE transformation, on the ripping/encoding end, to suit new hardware (4K TV, new HTPC etc) and move to using exclusively x265/HEVC.

What I am not willing to do is build a new HPC Grade double-digit multi-core Workstation to accomplish my back end needs.

I am currently running an i7-7700k on a z270 platform that gives me essentially everything that I could want for what I do currently.

I don't 'Game' and find no real need to chase further upgrades...except for 1. A new Video Card Platform with x265 Hardware Video encoding in mind.

Now, I have been doing this long enough to know that I will in fact be giving up a bit of quality at the end of the day using a GPU hardware encoder over using a CPU and software, but I just can't justify the money for a new system or the time for using the current CPU that I have.

For those familiar with GPU Video Encoding/Transcoding of BluRay and BluRay Rips, I am looking for suggestions on which way to go, AMD VCE or Nvidia NVENC ?

As far as software, I will stick with Handbrake/VidCoder/FFMPEG.

Specific hardware that I am considering, based on price and availability are somewhere in the Vega56/64 range for AMD or GTX 1060/1660Ti or RTX2060 for Nvidia hardware.

I would appreciate if anyone familiar with this, could offer some insight in regards to performance, quality etc, pitting the AMD VCE vs. the Nvidia NVENC.

Thanks in advance.
Allan

[sneaker]

Best cards for HEVC currently available should be Nvidia Turing with GTX 1660 Ti being the cheapest. I wouldn't ever buy AMD for video encoding. Nvidia is simply ahead in both compression efficiency as well as software support.

Have you tested your i7-7700K? Kaby Lake does have HEVC 10 bit hardware encoding.



P.S.: "x265" is an H.265/HEVC software encoder developed by MulticoreWare. "H.265" and "HEVC" are the names of the general compression standard.

[Allan74]

P.S.: "x265" is an H.265/HEVC software encoder developed by MulticoreWare. "H.265" and "HEVC" are the names of the general compression standard.

LOL..... I remember now.

Although, in my own 'Malcolm-X' defense, and as a Gen-X'r.....'X' will always be alt/open/generic/etc to me

As far as using the CPU that I have.... I just want something faster without completely rebuilding.
The possibility of a Threadripper (once the next gen is released) may be on the table, but for today, a video card is an easy bandaid.

I have been playing around with x264 Handbrake/VidCoder NVENCx264 on an old GT 730 w/384 CUDA cores (my htpc card) that I have and noticed that the GPU was hardly even being taxed and even with a passive cooler, I only saw temps increase about 10c over idle. Opening GPU-Z in mid transcode showed the 'Video Engine Load' at 99-100% with the card itself at about 20%. It took 30mins to transcode a higher bitrate 1080p/264 90m file to 7168kbps....and I tried other bitrates as a comparison using the same source and it was literally the same time frame each time.

This only left me with more questions.

Is the on-chip Nvidia encoder affected by higher clock speeds ? better performance with more CUDA cores ?
...or is the NVENC chip just the chip and everything else is disconnected ?

I have a handful of older cards to play around with using x264 until I decide on a new card for H.265 encoding.

I apologize for what may seem simple questions, but I am just trying to wrap my head around this so that I can purchase a card once and be done (for now anyway).

thanks.

[sneaker]

As far as using the CPU that I have.... I just want something faster without completely rebuilding.
The possibility of a Threadripper (once the next gen is released) may be on the table, but for today, a video card is an easy bandaid.

The i7-7700K has an integrated GPU and video encoder unit (like Nvidia and AMD). So you can run e.g. x265 on the CPU or use the Intel hardware encoder. In HandBrake you can enable the Intel encoder in preferences (restart HandBrake after), then select it in the video codec drop-down list.

[Selur]

Is the on-chip Nvidia encoder affected by higher clock speeds ? better performance with more CUDA cores ?
...or is the NVENC chip just the chip and everything else is disconnected ?

It does get no speed through overclocking, overclocking can however cause problems due to ram problems on the card.
Number of cuda cores doesn't influence the vpu and not everything else is disconnected
-> The important thing on the card is the vux chip (https://en.wikipedia.org/wiki/Nvidia_PureVideo) and don't overclock your system.

To use NVEnc you need at lease a Kepler card so too old cards won't help. (https://en.wikipedia.org/wiki/Nvidia_NVENC)

Cu Selur

[Allan74]

The i7-7700K has an integrated GPU and video encoder unit (like Nvidia and AMD). So you can run e.g. x265 on the CPU or use the Intel hardware encoder. In HandBrake you can enable the Intel encoder in preferences (restart HandBrake after), then select it in the video codec drop-down list.

WOW.... I never bothered to enable it as I am running an addon card and this is my first system back on consumer CPUs in 10 years after a few Xeon platforms.
My last 'Desktop' CPU was a Q6600 before I went to Dual Xeons.

For H.264, QSV was roughly 2x faster on a single pass compared to the GPU that I was testing (GT 730)....32mins vs 17mins on a 106min 1080p/264 Rip.
H.265 is next.

Obviously I love the ability for a 2 pass, so this is going to take some playing around with.

I am heading over to a buddy's place to play with his 2080Ti and 9900k using the same source file and software, so this will all help in my decision.

Thanks again for all the info...... I can't believe how 'in the dark' that I was.

[Allan74]

As an update to my last post....


The only thing that I have left to do is test to compare if enabling QSV Decoding has a negative affect on overall time, as at this point, I have tons of CPU and max temp delta to spare.

The CPU is currently running @ 5.0Ghz on air cooling and has been de-lid'd, so even using the software encoding exclusively, temps never got over 70-72c.
At roughly 50% CPU usage on this 265 run it's hovering at about 40c and hits 50c briefly (QSV Decode turned off to compare vs. On).

Will overclocking the Intel GPU help QSV run faster ?

thanks again for the info. I think that simply using the CPU's features will save me cash and sanity.

[Allan74]

Still using H.264 and testing with QSV.....

I have now found myself able to transcode a 2Hr 12min 1080p source file to 1080p 8192kbps (for housewife archival purposes on the media server) using Handbrake 2pass with everything cranked up for quality (speed on slow), in about 37mins.

This has BY FAR exceeded all my previous expectations for speed/quality compromise, but as with everything, NOW I WANT MORE.

I have also found that I have 'enough CPU' to be able to leave 'QSV Decoding' OFF during a 2pass run, which improves overall performance/time, at the expense of CPU load, but again, not a big deal, as it leaves the IGP to Encoding duties exclusively and the time can be measured in more than a few minutes difference when testing, so I found it to be a net positive.

I am wondering if there are any tweaks or perhaps some IGP overclocking that I can do to improve performance.... ?

Here is a rough idea of my current setup:
- MSI Z270 Gaming M5 Motherboard.
- i7 7700k @ 5Ghz (De-Lid, Liquid Metal'd), -1 AVX Offset (max 70c on Noctua U12s Dual Fan during stress testing).
- 4 x 8GB DDR4 2400 @ 2400 (XMP auto).
- Intel P600 1TB PCIe M.2 NVME
- Geforce GTX 950 (as primary display).
- Motherboard in 'Multi-Monitor' mode, IGP as Co-Processor only.

I am wondering if increasing Memory Speed, Timing tightening or IGP Overclock will have any REAL effect on the performance of my future transcodes.

I am able to handle the heat, as even peg'd @ 100% during a QSV Accelerated Transcode, IGP and CPU never get over 58-60c.

As I stated above, I am still using H.264 for testing, as I have found that (for me anyway) H.265 just seems to take almost exactly 2x longer after testing the same input files in a number of scenarios.

Thanks in advance and I sincerely apologize for the rambling.

[Cumberiron]

Is the on-chip Nvidia encoder affected by higher clock speeds ? better performance with more CUDA cores ?
...or is the NVENC chip just the chip and everything else is disconnected ?

It does get no speed through overclocking, overclocking can however cause problems due to ram problems on the card.
Number of cuda cores doesn't influence the vpu and not everything else is disconnected
-> The important thing on the card is the vux chip (https://en.wikipedia.org/wiki/Nvidia_PureVideo) and don't overclock your system.

To use NVEnc you need at lease a Kepler card so too old cards won't help. (https://en.wikipedia.org/wiki/Nvidia_NVENC)

Cu Selur

HI Selur! Number of cores does not matter in NVENC encoder, but I think the card has only 1 frequency generator, when you overclock the card, it really effects the transcoding speeda, but just a bit