Performance Matrix Roon Server


(Mikael Ollars) #1

I know there are recommendations available for chosing hardware, but there are quite few performance reports from real world examples. Are there more people than me who wants to know what they can expect from, say, a Gen 6 i5 NUC running ROCK compared to an older PC with an Intel Core i5 3470?

If so i suggest that we create a matrix within a limited set of boundaries to keep it simple and reasonably comparable and understandable. The only way i know of getting performance figures out of roon is by using the DSP and looking at the “processing speed”.
And then we need to keep it simple and basic so i suggest two values based only on upsampling RedBook material into one stream for this test. Only one mandatory value here so most people can add. (If your DAC doesn’t do DSD, you are welcome to contribute the 24/192 upsampling performance figure.)

If you have an energy meter those values are highly appreciated too, one value when idling and the other two when disks are spinning and your sending an upsampled stream to an endppoint.

I started a public spreadsheet on Google Docs, available here

And please do add if you have the opportunity!


MOCK - Tell us what strange gear you've installed ROCK on!
What kind of chip best runs Roon Core? More cache, cores, speed...?
(Daniel Beyer) #2

A good way would be to put the cpu of both into a comparison site to see. Using your example,

http://www.cpu-world.com/Compare/142/Intel_Core_i5_Mobile_i5-6260U_vs_Intel_Core_i5_i5-3470.html

The desktop cpu will win in terms of cores and core speed at the expense of using more energy. Since most NUCs use a mobile cpu that can be a general rule of thumb. Lower energy cost, lower cores and less speed than a similar desktop cpu.

Personally, if you are building a machine to stick in a closet, I would not go with a NUC, I’d just build a full on desktop. It will in the end be cheaper and much faster, and upgrade-able.


(Mikael Ollars) #3

Absolutely Daniel, but that wont tell me squat about its Roon Server performance! :slight_smile:


(Mikael Ollars) #4

I would appreciate it highly if someone with a NUC i3/i5/i7 would add their performance and power figures to the spreadsheet! :slight_smile:
I just rebuilt my MOCK exorcising the AMD E-350 barebone motherboard and replacing it with a modern Asus Prime J3355I-C.
I reused everything, SSD, Internal drive and chassi. Even the memory was usable! I have, however, changed the memory for a pair of 1866Mhz DDR3L for a slight increase in performance and further lowered power consumption.
It actually started up just fine when i disabled the UEFI boot and from the ROCK GUI i reinstalled the OS. Everything still in place, full library of around 110K tracks and personal setting etc.

And, most important, the powerconsumption was basically cut in half! Very nice, but what was even better was the more than doubled performance!
Now uppsamping RedBook to DSD128 is done at almost 2x and i even tried DSD256 which works but is barely above 1x so i wouldnt rely on it. (Couldnt hear any glitches though)

So, how does a cheap low power barebone compare to a NUC?


(Mikael Ollars) #5

The spreadsheet has been updated with some performance figures from my freshly built i7 8700 Coffee Lake computer. A couple of observations;

  1. The performance according to Signal path info was surprisingly low? There must be something strange goin’ on! I mean, upsampling RedBook to 192Khz on a 6core/12thread CPU must surely tax it way less than the 20x indication. I actually had almost the same figures while simultaneously ripping a Bluray to hard disk at the same time as i was compressing another Bluray movie with Handbrake!!! I can assure you that the 12 threads were burning some W!
  2. The power consumption was very much all over the place? Seems very different from older Intel chipsets consuming power at a much more even rate.

I would still like to see some figures from a Nucleus(+) or NUC!

Oh, the figures i noted in the spreadsheet were basically the same with a ROCK install on the same hardware!


(Daniel Beyer) #6

not all cores are used. So a lower core but higher Mhz CPU wil perform better. My i7 4790k does the same at 42.3x.


(Mikael Ollars) #7

That makes sense, but i was under the impressions that ”parallellize Delta Sigma modulator” would allow the Core to use more cores for heavy tasks?
Btw, do you know if there has been some changes to the code concerningthe processiong speed indicator?
I seem to recall reports of performance figures up in the thousand x’s?


(Andrew J Shepherd) #8

The delta sigma modulator is used for only one task. It does not apply to all DSP.

AJ


(Daniel Beyer) #9

Yes, the performance indicator was changed as the devs did not feel it was giving as useful information as it could. See:

Also, see this:


(Mikael Ollars) #10

Thank you Daniel! :slight_smile: Those references were useful!

And i think i’ve figured out why the i7 8700 displays such modest figures when upsampling RedBook to either 192Khz or DSD128… It’s Windows/Intel’s way of managing processor performance! I setup the Roon Core on my HP All-In-One touchscreen PC, an i5 6500 Quad Core running @ 3.2Ghz as a default. While happily upsampling RedBook material to 192Khz the Signal Path reported some figures of about 17x which is way more than required, but when looking simultaneously at Windows Task Manager Performance Tab, i realised the CPU was running at 1.25Ghz and utilising some 3-4%. I changed to upsampling to DSD128 and the processor was running at 2.5-2.8Ghz and showing some 15-20% usage. Roon reported a Processing speed of about 6.5x which adds up!

So, this has been quite useful as an experiment and learning session. I have excellent response from anything between the Asus Prime J3355C-I ($75) and the i7 8700+Asus Prime Z370-P ($450) and can upsample one zone at a time to whatever when i feel inclined to do so. (Much more with the latter of course)


(Anders Vinberg) #11

I looked into this with my new Nucleus.
It has a modern i3 processor, don’t know how much RAM, the basic SSD (I think 128 GB?) plus my added 2 TB SSD.

Power consumption with the audio stack off, just the Nucleus idle plus a switch and some minor stuff: 14 W, not 30 W
Power consumption with a MicroRendu/LPS1, Meridian 818 + DSP8000SE on: 180 W (whether music is playing or not). Note that those speakers contain DSP, several DACs, and five 150 W amps, but the amps are class AB.
I can’t measure the power consumption of the Nucleus while processing, because my power measurement comes from my PSAudio P10, and the only way to measure the Nucleus alone is to turn everything else off, and Roon won’t play with no output available.

Roon performance indicator:
44k upsampled to 176k: 45X
44k upsampled to DSD64: 5X
44k upsampled to DSD128: 2.6X
44k upsampled to 176k + convolution with 131k taps: 18X
44k upsampled to DSD128 + convolution 131k: 2.4X

The Nucleus handles my demands well.

EDIT Correction, the Nucleus idle draws 14 W


Intel NUC Kaby Lake i3/i5/i7 CPU benchmark
(Mikael Ollars) #12

Great input, thanks Anders!
It seems the i3 is a viable option for anyone not needing to upsample or do more than one zone at a time. And all at a very reasonable power consumption.


(Fredrik) #13

I measured my NUC7i5 today it has M.2 SSD for ROCK OS and second SSD for music.

This is with regular 44.1/16 no DSP

19Vdc average 0.57A (10.83W) and peak 1.5A
15Vdc average 0.69A (10.35W) and peak 2.1A
12Vdc average 0.81A (9.72W) and peak 2.8A

Average is during play to Roon endpoint. Peak was during start up.

I have not measured amp usage during a scan for example but if all cores are being used during a library scan I suppose we can see constant use high as peak is in my test.


(Anders Vinberg) #14

Yes. In fact it does the highest upsampling and convolution. And I don’t play multiple zones.

Based on what Roon Labs told me, I had concluded it would be sufficient for my needs, and I always like using the least possible amount of computer power.


(Mikael Ollars) #15

Tackar Fredrik! I was a bit confused at first, but later realised you can run the NUC on any voltage between the ones you mentioned?

If my math is correct you are seeing up to 30-35W while putting load on the i5 NUC. That makes sense. Are you feeding the NUC with your Paul Hynes PSU?
and are you measuring before the PSU or between PSU and NUC?


(Anders Vinberg) #16

As I recall, these performance multiplier numbers are much better than I saw on my Windows i5 NUC. (I could hook it up and look…)

Is it the Linux-based Roon OS that is much more efficient than Windows?
Or a newer chip?
I bet on the OS.
@Danny ?


(Mikael Ollars) #17

When i tried ROCK on my i7 8700/Asus Z370-P i saw basically the same performance figures as in Windows 10+Roon. So my bet is on the chipset/bios performing power management in my case.


(Fredrik) #18

Yes anything between 19-12vdc will work according to Intel but I think 19vdc is the prefered voltage.

I tried some up-sampling to DSD 64 and the amps went up with 0.10-0.15A so not much difference.

No I use my Paul Hynes SR 4 to power my SOtM SMS 200 Ultra. I have a HD Plex PSU with 4 outputs that power ISP fiberconverter with 5vdc, router with 19vdc, switch with 12vdc and finally my NUC with 19vdc.

I have a 0-32vdc 0-5A PSU that I use to test equipment with so I can se exactly how much amps it uses on the DC side.

CPS-3205


(Mikael Ollars) #19

I have tried another Roon server, a Dell Optiplex 7010 SFF. Equipped with an SSD it performs server duties as either a Windows 10 server or as a MOCK machine.
One difference which may affect results is that in ROCK OS i mapped my NAS as storage. However, ROCK actually increases performance while lowering power consumption! :slight_smile:
And a weird effect was that it consumed less energy while upsampling to DSD128 than when upsampling to 192Khz PCM? Almost the same power consumption as when idle!

Info added to the spreadsheet!


(Mikael Ollars) #20

A few NUC’s have been added to the spreadsheet for those who are curious!