As an effort to improve my Roon server performance, I did the following experiment.
Server A: ROCK on NUC 7i5BNH (fanless in Akasa Newton S7 case)
Server B: Roon server on Ubuntu 20.04.6 installed on MacPro Late 2013 (aka trashcan)
Server C: Roon server on MacOS (Monterey) on MacPro Late 2013
To do an A/B comparison, I got a 15 day trial susbscription, which is installed either on Ubuntu or MacOS, and did the comparison. My house is wired under a compound subnet, and all devices are connected to the LAN and there is a central router for the compound. All servers feed signals to the DAC (dCS Rossini Apex) using the network input. My library has about 5000 albums on 5TB USB HDD and 5000 albums on Tidal/Qobuz (130k tracks).
ROCK vs. MacOS
MacOS runs faster, not much delay in search. ROCK has random delays in loading artist, album information. However, ROCK sounded noticeably better - resolution, dynamics, and tonal purity. MacOS sound was in comparison more in my face and veiled/congested. I know this is not a surprising outcome, but I was still surprised how much SQ difference I could hear.
ROCK vs. Ubuntu
Here again, Ubuntu performed nearly as fast as the MacOS, with a few seconds of delay at most on search. SQ difference between the two was much closer, but upon tedious A/B comparison, I would have to say ROCK has a slight edge in terms of clarity and bit smoother midrange.
Conclusions:
In terms of performance, ROCK has the idiosyncratic slowness that neither MacOS or Linux has. One could perhaps attribute it to the hardware specs, but it’s hard to believe that the NUC 7i5 with 8GB RAM built in 2018 would substantially underperform compared to the old trashcan (3.7GHz Quad-core Intel Xeon E5).
So I suspect that there is still some software inefficiencies in the way ROCK is implemented on NUC, in “some” configurations. This has long been my speculation to ROON but it was nearly impossible to confirm as everyone’s setting is different.
I terms of pure sound quality, I will have to stick to ROCK, although I may be happy with a well built dedicated Linux Roon server. MacOS is not recommended. This is rather expected as the device has fewer secondary processes moving down from MacOS to Linux to ROCK. If I put the SQ scale of 1 (MacOS) to 10 (ROCK), Linux would be rated at around 8-9, which is also consistent with the simplicity of processes in the device.
So I am content with the ROCK for now, and slowly contemplating whether I should build another ROCK with 2024 spec or get a Titan…
Same as before in your other threads, I don’t have this so it’s not a fundamental ROCK thing. But interesting that other servers don’t seem to exhibit the problem for you.
Roon has never claimed SQ differences for ROCK or Nucleus and one would think they would if there were any. This would certainly be a selling point for Nucleus.
Search is handled in the cloud, so I wouldn’t take much importance from the results. Search performance is likely affected by many external factors, most likely all network-related.
In my experience, Roon on an Intel Xeon E3 doesn’t perform as well as an i3.
@Martin_Webster Please read before you comment. I was running two machines nearly simultaneously using two separate Roon licenses through the same subnet. Within a few seconds, both machines access the same cloud (same artist/album) through the same network server. I think the result is pretty objective.
I did read your comment, and I stand by what I said. Your test was entirely subjective. Resource demand, i.e., cloud services, network congestion, caching etc. impact performance, and can change significantly every second.
Do a search on the same machine, and the response times vary!
I have seen the search slowness using Rock and then moved to DietPi (Debian based) and could monitor the performance of search and turning on maximum performance in the BIOS and DietPi made a difference. Though when Roon Cloud is running slowly it still has an impact on all platforms, though more cores make a big difference for search.
I also tried Roon server on my M1 based Mac as well and eventually decided to stick with DietPi. Though all worked fine and I never noticed a difference in SQ but I wasn’t looking for any either.
Maybe it is in the nature of trying to be more appliance like than a full performance Linux distribution.
I was able to make DietPi run at the speed of Rock by turning off turbo performance. I also tested limiting cores for Roon process and this had a big impact on search in particular. 7-8 cores had a big positive impact on search at least for me. Most of Roon seems to run in single thread, but search seems to be more multi threaded in functionally.
Not even for TailScale support
Got to say I had 4 good years running on Rock, but the 8 months I have been using DietPi have been great. Artwork on my TV, TailScale, Roon Extension Manager as well as Plex Server all auto running great.
As well as a DNS server and a few other non intrusive extras that do not get in the way.
I think for many users adding TailScale will be good and having something like Extension manager would also be nice.
This was my performance test across the 3 machines and operating systems I was testing Roon on. It led to a bit of tweaking and the understanding that Roon had done a fair bit of work on getting search cloud and multi thread compatible, but it needed at least 6-7 cores for a search for The The to return (at that point in time) in under 15-20 seconds.
@Michael_Harris This sounds great. Could you write the instructions on “how to set up a DietPI for Roon server and TailScale”? Starting from a proper hardware, and installation guide. That would be great. Or is there one already?
Andrew I am not a great one for documenting the processes I follow as sorry.
But if you follow any DietPi (on X86 install guide) then you will be good to go.
Roon is just a simple menu software install from this.
I did document the Roon artwork screen output on HDMI as I never found this anywhere so if you get to the point that everything works and need that I will happily help you through it.
If you want to start a thread there are quite a number of us here who have been through this process that would be happy to pitch in and help you through it.
Thank you. I understand that DietPi was intended for SBC, but can I possibly try it out on my trashcan MacPro (late 2013)? I have a couple of them at my disposal. Or does it defeat the purpose of using DietPi?
Yes, DietPi was originally intended for SBC, but since a few years there is a build for x86-based platforms as well (this includes your trash can).
The procedure is the same and the menu driven software implementation as well.
Create a boot stick with the current build of DietPi x86, install onto your Mac.
Not much more to it, for details Google is your friend.
You do have to ensure that Roon user is part of the dietpi Linux group or it will have permissions reading any drives you have. You can find out how to do this under the dietpi bits on here.
