I currently run an i5 9400 Ubuntu box headless server/DVR. I was exploring using an Nvidia 1060 as a dedicated transcoder or using an i9-9900k. However, I was advised that the 9900k would give me more benefit as a server processor. Now, this got me to wondering if there was a tipping point is between a rapid HW solution like Quicksync or an Nvidia card vs. brute force through software on multiple cores. Are there jobs that make more sense in software vs hardware? Are there times to disable hardware acceleration?
I am not going to make a change, but when consumer CPUâs are at 32/64 cores, will hardware transcoding still be optimal?
Neither of which really addressed my question. Are there ones I am missing that address the use of massive numbers of CPU cores?
The forum posts I checked were:
And then another thread I have loked had a good answer by you regarding hw vs sw, and how strong the Ryzen 7 3700x is compared to an i7 8700.
It also touts the strength of the Ryzen 7 mated to an Nvidia video card. But âscreamingâ is hard to quantify. Nor does that look at when things get to 64/128 threads. Is there a point where software (CPU with very high numbers of cores) in brute force will be more effective than hardware (Intel QSV (seems like this is approachable) or Nvidia (seems nearly untouchable))?
each 4k transcode requires about 17000 passmark, so you do the math on any current or future cpu.
further, other processes are few or single threaded, so pure core count is not the solution alone, you need to have sufficient single core power and qty to handle audio/subs/tone mapping (for each individual stream, in addition to the passmarks for the cpu video transcodes).
As with a lot of things this is subjective - technically or practically?
Technically - SW/CPU encoding is the better quality encode result. Does that matter when youâre trying to watch some show on a 6 inch phone screen on a plane? Or even in just another room of the house? Or while at the in-laws? Not really. Go GPU for this sort of stuff - the âusageâ stuff.
Now, if youâre encoding for archiving purposes and not just âat that very moment passive consumptionâ, thatâs when I suggest go CPU over GPU.
good point, for plex transcoding on the fly, as long as the quality is âgood enoughâ, it is all that is needed. Itâs not like the transcode is being saved.
Besides the point, when converting 4k/hdr to sdr, quality-wise most will be better off using an original SDR source file, vs trying to convert 4k hdr to sdr with tone mapping.
Are you advocating that I trust random users over the support forums where Plex employees like yourself offer support? Moreover, I am not asking about QSV vs NVENC so much as CPU cores doing work in software becoming more effective than hardware transcoding. So, for our purposes, letâs think in terms of 2160 HEVC HDR and the lowest things are transcoded to are 1080/30. So, granted that a dedicated GPU is better than CPU for transcoded. But what about QSV vs a 32 core Threadripper 3970x?
