Poor DSD512 upsampling performance on NUC10I7

Add to that the basic profiles which are, if i remember correctly:
Performance, Balanced and Power save.
The last one shuts off speed stepping and may help lower max power consumption of the machine. In my scenario i successfully powered the NUC10i7 from a Nordost QSource, but only in the Power saving mode.
(In the other modes the NUC simply rebooted directly after OS bootup, when Roon started investigating media changes)

1 Like

excellent point - power supply plays a huge role. with my three NUC’s I figured they are terribly picky in this regard. One 8i7 would not power with another 8i7 LPS these are two slightly different boards and CPU’s but with own characters :slight_smile: and 10i7 has its own device.

1 Like

Is everyone familiar with Danny’s explanation of Processing Speed in relation to upsampling?

1 Like

How many cores are used at dsd Upsampling in the multi core mode? All or only 2 as supposed above?

My guess was correct, it uses 2. This sort of parallelisation isn’t easy to do!

It is the single thread performance of the NUC you need to look at

I prepared a comparison table here Intel NUC - Roon Rock - only for TIDAL - #8 by simon_pepper

This covers from the NUC5i3, NUC7i3 (as per a Nucleus), NUC7i7 (both the 2 core & 4 core version, the later used in Rev B Nucleus+), NUC10 i3, i5 & i7, and with the best single thread performance the NUC8i7 but at a cost of 28W TDP over the others 15W TDP

In a past Benchmark I was able to use a NUC5i3 to perform DSD256 upsampling, while sustaining playback in other zones, indicating the DSP is core bound.

For DSD512 upsampling, either the use of 2 core Sigmar Delta processing or you need to use a Desktop based CPU with a greater single thread performance.

NUC10i7 is 25 W TDP.

Simon, you reference TDP “cost” too often. The cost factor/difference is negligible vis a vis Roon, since most Roon activities are loading the CPU well below base frequency across some/all cores.

Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

Per above, TDP more or less represents power dissipation at the high end of continuously sustainable operation. Idle or low end operation dissipates much less power, regardless of TDP spec.


Just to confirm when I measure my 8i7 power consumption it’s usually 8W.

That’s actually faster than I thought.

See Nucleus+ Dropping When Upsampling to DSD512 Using Qobuz & Fios

You’re expecting too much from this mobile (meaning weakened to reduce power consumption) CPU. There are certainly faster CPU than mobile i7.

See Roon Nucleus verses Apple Mac Mini M1 - #63 by wizardofoz

Up to 2 cores per zone if Parallelize Sigma Delta Modulator is enabled.

You need a high single thread CPU performance if you care about Roon DSP performance.

If you have the time, you can try installing Windows 10 (not 11) on it. I think there were posts that discussed they got a higher processing speed indication from Windows than ROCK (regardless of whether this difference in processing speed number is meaningless or not. One may argue that if you get a higher speed the OS is not doing a good enough job to lower the CPU clock rate as the unneeded performance is wasted as power consumption, heat, or if you believe it, noise).


And if course there are m-itx if you want a desktop in a small form factor. But, so many have written off, ye old desktop tower. A larger form factor can equal easier cooling, and quieter operation. Fans can spin slower, if at all,. Etc.

1 Like

I tested a bit more. And at some songs got dropouts also in parallel mode.
Then I switched off hyperthreadingn of the processor in bios. Wanted to try rock to work only with physical cores and not virtual. Now rock recognizes just 6 cores and not 12 as before. As expected.
The song I had dropuots before plays now without.
The processing speed starts now with 2.3 and goes down to 2 and after three minutes to 1.9.

I do not know whether that were coincidece and not the result of disabling of hyperthreading. Will do tests further and report.

1 Like

A question to roon: would you please implement a function to allow user to choose a number of cores for dsd upsampling. So not only to switch parallelized processing but also to choose the cores number woulb be very helpful. Thanks!

If you want Roon to see that suggestion, I would suggest making a Feature Suggestion thread about it.


Thanks, done.


@Serguei , while the option to play with cores for DSD upsampling may be an interesting feature, I doubt it would be of any serious help. There are several things to reflect upon in this discussion.

ROON explanation of “Processing speed” that @Neil_Small reminded us all about (thank you Nail by the way). Bottom line, this is not exactly the CPU performance indicator. I don’t know the way how one can see the CPU Cores/Threads utilization in ROCK, i think we are not supposed to mingle with ROCK OS. But I would like to show you how it looks for ROON Server running on Linux Ubuntu 22.04 Server. This is DSD@512 playback:

You may see that actually ROON balances its load on 4 cores / 8 threads fairly nicely (frankly much better than I thought it would) And there is WAY LOTS’a 'head room in CPU. Basically “RoonAppliance” is taking 1/8th of the CPU capacity for DSD@521 with 7th order (CLANS), that is. Guess what! “Pocessing speed” in ROON is 1.9 for this. And this is on i7-8665U CPU, in fanless body.

Another interesting metric may be the NUC CPU frequencies. Actually the CPU power settings I have mentioned above influence exactly this - time windows in seconds for higher/turbo frequencies.

It’s just a screen shot, the stats are very dynamic, but I was surprised that spread of frequencies on different cores is from 1000 to 3600 Mhz (short bursts). Importantly, you may see the CPU is not breaking a sweat with temperatures, all around 50C, it’s all a light breeze.

Bottom line - DSD@512 playback on NUC8i7 may be very stable.

If you continue to experience drop outs the bottleneck is somewhere else SSD i/o, Ethernet, Memory bank, etc.

Although as discussed above NUC10i7 may not be necessarily a huge step up from 8i7 nevertheless it has slightly more head room, another words the bare bone performance must be better in this regards.


Today i did some further tests.
I changed nothing. But get today many dropouts in parallel mode. The processing speed in roon is the same as before. 1.9-2.2x.
Then I played many dsd512 files. Without the necessity to upsample. Plays perfect. So the reason for dropouts my not be at my streamer or at lan.
Defects at the ram, I would rule out, because another functionality is without concerns. Ssd also.
I still did not play with cpu parameters.
But this unpredictable behaviour of the occurrence of dropuots at different times is strange.

As the evidence that may nuc functions correctly is the dsd256 upsampling which functions fine (processing speed at ca. 3-4.2x).

After Igor’s last message with cpu metrics I have doubts that the cpu tuning will help.

Do someone else upsample to dsd512 on a nuc without concerns?

It’s quite possible still that the cpu is throttling just enough to stutter with higher rate conversions

Older 7i3 in an Akasa Plato fanless case.
Upsampling to dsd512 with zero stutters or dropout at just 1.3x processing speed.
This was being fed Qobuz from a Lumin T2


You still have Hyperthreading disabled right?

Is TurboBoost On or Off?

Also try disabling C-States, see if it makes a difference.

Hyperthreading is still disabled.
Turbo is on.
Disabling c-states - will try it out. Thanks!
Do you have some special bios settings?
Is it roon os or rock what you installed on your 7i3?