I’m thinking of upgrading my PMS and I quite like the idea of the new 8th Gen Coffee Lake CPU based Intel NUCs. However there’s i3, i5 and i7 versions to consider.
I will be unlikely to require more than two simultaneous streams and although no 4K stuff presently, I want to know I can if/when I want to. I’ll be using an HDHomeRun quad tuner and I want Plex DVR to be able to save recordings as .mp4 files (which currently is not possible on my Mac Mini) and so there should be no need to transcode on the fly as all clients can play MP4s.
I will also be running server stuff like email, DHCP, DNS etc.
I’ve dismissed the idea of using an i3 and I like the idea of a fast i7, but that would no doubt be louder than an i5 based NUC due to the fan. But ignoring any noise issue, would an i5 be acceptable? Barely or easily? Of would an i7 really be a better choice?
So, hope those with experience of running PMS on different platforms and CPUs could offer some advice here.
Any modern intel CPU will breeze through several transcodes using hardware acceleration.
An i3 would do fine but I would go i5… I have an i7 and it would be overkill for only two streams.
While the hardware transcodes are available on low end CPUs it’s worth noting that the software transcodes produce a higher quality-per-bit than hardware, so you may prefer to do the transcodes in software on that basis. The typical recommendation is a passmark of 2000 per 1080p transcode so that should give you a general idea. You can lookup the chips in the NUCs you are considering and the passmark is easily found from the chip model.
Not been able to find any PassMark scores for either the i7:8559U or the i5:8259U. I guess they’re still a bit new. Other comparisons between them put the i7 about 10% faster than the i5.
Yeah despite the lack of benchmarks either should be easily fine for two transcodes.
My personal thoughts regarding HW or SW transcoding is this that a lot hinges on the input file.
Though it’s true that software decoding can give better results… it’s all relative to the source file.
I personally only have untouched remuxes.
When forcing them to HW transcode the loss in quality in negligible on a 55"… However I would hazard a guess that if the source file is already heavily compressed then I guess there could be a more noticeable quality hit, so in that case SW decoding would be a better option.
When you do purchase I guess the best option is how you perceive the comparison.
For me Quicksync is the biggest thing that ever came to Plex.
But like I said just my personal view.
First, I am completely on the same page as you. But I’m starting to see the “tipping point” in the real world.
Somewhere around the Kaby Lake generation of Quick Sync graphics, my wife can no longer tell the difference between a software transcode of a full-quality blu-ray rip (to a Roku stick) and a hardware transcoded version, on a 60-inch TV.
I really should give credit to Intel there. They’ve made some strides.
I’ve not played with it on the recent hardware. If you really want to do A/B testing, usually you can see the difference more dramatically when you transcode down to a lower bitrate. The HW transcodes tended to be particularly bad at the low Mbps and sub-Mbps range.
You asked if it can do 2 and so I was answering that specific question. I did not mean to imply that it was capable of only 2 (should be capable of ~5). Keep in mind that the 2000 number is a gross estimate so it could fluctuate higher or lower depending on the source media.
That would explain why they are not on the wikipedia page nor really present in the benchmarks.
No disagreement, Intel can never compete with well-tweaked software encoding (aka the Plex encoder, or Handbrake) in a bit-for-bit comparison. I always use software encoding in Handbrake. But…
My house wifi is solid. So the connectivity to the Roku is good enough that it’s not noticeable.
I have solid connectivity via LTE but obviously it’s a lower bitrate. However, on something as small as a phone or tablet, it’s also not as obvious as earlier generations.
So while I totally agree about software encoding being better, Intel is getting surprisingly close to “good enough” territory unless you have dreadful connectivity (as you noted!). With QSV able to do hardware encoding of HEVC, I wonder if it could beat/match software AVC bit-for-bit. Hmm, perhaps time for a weekend test in Handbrake.
So i7 or i5 is still the question. I wonder if the i7 would be significantly noisier due to cooling fan than the i5. Anyone have any experience of this comparison with earlier generations?
In fact, do the NUC fans run continuously or only when the CPU is working and getting hot? In fact, are they fixed speed or variable according to cooling requirement?
I realise this new generation of NUC could operate differently, but I would be interested to know more about the earlier gens. to help my overall understanding of NUCs.
As a datapoint, my NUC (4th gen, Haswell): The fan runs continuously at a minimum speed that’s settable in the BIOS. It’ll spin up with higher temperatures and spin down when the temps go down. If I’m within a few feet of it I can hear it when it is at a high speed but not from across the room.
