Roon Hardware Platform Suggestions

You could also consider, if sticking with windows for a dedicated use, something like Fidelizer to reduce windows processes. I will be looking at this for my dedicated w10 system too.

I bought a used NUC i3. I have it next to my router. I run Windows 10 and it works just fine and cost peanuts. I am considering moving to Linux though.

Of be very interested to hear hope you find that works for you. I didn’t really understand how to use Fidelizer and in any case it wouldn’t be appropriate to use it for my general purpose pc. Of I go the dedicated windows box route it would certainly be an option though.

Is there a reason you wouldn’t just use ROCK?

I want to use it as a server too. It’s an i3 series 3 so I don’t think it quite meets the specs for rock.

I have run ROCK on a DC3217IYE from the Ivy Bridge series (2nd Gen NUCs), with a
Intel® Core™ i3-3217U Processor (3M Cache, 1.80 GHz), 8GM and 240GB SATA SSD

It ran fine (just make sure you are the latest firmware) and was only upgraded when I obtained a NUC5i3 (NUC5i3MYHE) and became my spare. It is presently on loan to a friend and working fine.

Thanks for sharing. I’m interested in running some sort of room correction and possibly play with filters/upsampling so I believe from all the helpful advice I’ve been provided it looks like an i5 or i7 is on the cards. I’ll tinker with my Macbook Pro first to see how this does as an easy and free gauge of my requirements.

Hmmm, I’ve just made sense of what you were showing me @simon_pepper remember this is all very new to me… I think I understand that what you’re demonstrating that relatively modest setups can do upsampling for PCM. DSD seems to need more grunt which isn’t much of an issue for me as I have little DSD in my collection. Do you know if room correction using REW or HAF requires a lot more overhead?

These generate the filter files you then load into the DSP section.
So I understand that Scott’s file had a series of node points in it, which were included in the signal path with very little impact in my benchmark tests (I don’t believe anyone else has actually undertaken this type of testing, but just jump straight to the higher more expensive NUCs)

It is my understanding that the internal ‘Computing Cores’ of the x86 microarchitecture within the CPUs of i3, i5, i7 chips has not fundamentally altered in some time. This is basically set on chip die size and silicon architecture, which has been 22nm with Ivy Bridge in 2012 and then 14nm with Broadwell in 2014, all supporting 14 pipeline stages. Yes, they have increased max processor clock speed, but this increases the heat output, the TDP. A NUC uses the lower heat output versions and limits the clocking due the restricted heat processing of the chassis footprint, so this top-end capacity is not available when used in a NUC, particularly in an audio application - who wants a screaming PC tower with huge fans to run Roon Core?

What Intel have done in the CPU generations is put more functionality onto the processor die, with more on board GPU.
The NUC generations NUC6, NUC7 & NUC8 have then offered better support for 4K displays, better WiFi, built in Bluetooth etc. But the fundamental processing capability has not changed. Yes, the benchmark performance reviews will show differences because many of the benchmarks test Graphics performance under load, I/O throughout etc, which have shown improvement with each NUC generation. However for headless operation as a media server using a Ethernet connection only, supporting one web-page with just numbers on it, you are not using any of this additional functionality or capability.

This is due to change in 2020 with 10nm die sizes (already seen in Cannon Lake), but the main plants have yet to come on-line, plus the number of pipeline stages in the CPU is around the same. So there will be step function up due to the reduce chip architecture, but it depends of whether the CPU pipes, and internal microcode can handle it, and retain clock integrity.

What is interesting is the revised delivery of NUC compute units separate from the chassis (with the I/O and storage). So with the Austin Beach range, it is a Fanless chassis, and a processing board.

This is great news for a fanless ROCK server, as you can start with a i3 module and if/when required just upgrade the compute module. No need to reinstall ROCK or transfer database etc from one unit to another to facilitate an upgrade.

Meanwhile use your existing Ivy Bridge NUC to see how ROCK works, or get a 2nd hand NUC5i3/NUC6i3 and see what the underlying processing requirement is and watch how the next CPU and NUC generation plays out.

1 Like

It was a convolution filter, just like you would get from HAF/REW, in this case via Acourate. Typically Roon will use one core per DSP zone, which can be split with Parallelize Sigma-Delta Modulator which can split this across cores if you are really hurtin’.
That thread showed that in every case tested, the i3 could do what the i7 could. We did not get to a point where the i7 could do something the i3 couldn’t. That will happen at some point, but Roon got pushed fairly hard and help up well on the i3. I suspect this will happen when Roon is asked to manage a large library, upsample to DSD AND do DSP (probably in more than one zone). That is a LOT. If you are pretty sure you’ll only use one zone for DSP and little upsampling, the i3 will get it done.
Then, you also need to ask the price/performance question. At some point this decision is akin to buying an insurance policy. If you only have i3 money, you should just get the i3 and be pretty confident it will do all you want. If you are a little uncomfortable about future needs, you ‘buy up’.

From the various posts here I have heaps to play with before I need to make any purchase. I may just be lucky that the Gen9 NUC mentioned are in good supply by the time I’m ready. Thanks for the history lesson as well mate, very interesting!

By the sounds of it, I’ll likely wait for the next gen NUC range to come and just go for an i5 which should suit me fine for the foreseeable future.

It isn’t playing DSD files that require more grunt.

It’s up-sampling a PCM stream to DSD that takes its toll.



It’s also downsampling dsd to pcm if your dac can’t handle dsd and you have dsd files in your library, that too will pound a cpu

1 Like

I do. Given the recommendation to put the Core away from the listening space, the computer noise really doesn’t matter.

But, if it did, then your characterization of screaming is way overstated. It is easily possible for a large very powerful tower to be quieter than a NUC. The larger airspace means you can use larger fans moving at a slower rate, ie lower noise. Additionally, there are specialty fans made to quiet that can be employed in a desktop case, not in a NUC. And finally, you can have direct control over fan decibel levels with the right fan controllers.

Newer digital power supplies, control whether power supply fans even come on.

And finally, there are desktop cases made with internal sound dampening materials if you need it.

I even use a fanless power supply (in my tower case) and fanless CPU cooler. So the case fans (easy to control) are the only moving parts!

1 Like

Yes, indeed. It’s amazing what is available and possible with high end computer parts.

I’m using passive cooling for EtherREGEN and OpticalRendu :slight_smile: These puppies run pretty hot and with ambient temps in the low 30c’s these help a lot…also weight down the devices nicely.


Interesting. Much higher than I would have guessed.

Since @wizardofoz lives in Singapore, I think when he says ‘ambient’ he is talking about the climate.