Limits has stayed at 0 at all times with Direct SDM disabled. Still get pops even at -30dB fixed volume, listening to Keith Jarrett streams on Tidal (not compressed).
I noticed as I was screen sharing with dad that DSD512 started off ok (listening to quiet Keith Jarrett Tidal stream to pickup pops easily) until I started browsing his Roon library on his Mac…
i.e. when Roon Core starts doing more work (on the same machine as HQP Embedded of course), and CPU usage (and network usage?) goes up, the pops with DSD512 become much more apparent.
When I next visit him I’ll take the sonicT i7 from the home office and into the listening room and connect it via USB directly to the Pro-Ject S2 , i.e. by-pass ultraRendu, for testing purposes.
@jussi_laako - until Jesus can test a sonicT i7 himself (or someone else that comes to read this thread), have I run out of things to try in HQP config?
@andybob - are you running Roon Core on the same machine as HQP? Or separate machines? If both are on the same machine, what are you machine’s specs and can you run the 2s filter variants at DSD512 without issues, with the single Roon Core + HQP machine?
After asking andybob about if you have a single machine or two separate machines, I thought since I’m screen sharing on the old man’s iMac (2010, old i5, 8gb of old RAM) then why not separate HQP Embedded from Roon Core to test what I was thinking…
So I quickly setup Roon Server on his iMac with just his Tidal account for now. No need to import his USB drive for a quick test.
And I stopped RoonServer on his sonicT i7 so that HQP Embedded has the machine all to herself.
And voila - DSD512 playing perfectly fine with xtr-lp-2s filter… Even with RoonServer extremely busy importing Tidal library and images and me browsing his library, not a sing pop…
So my hunch that the sonicT i7 can’t handle both HQP Embedded DSD512 up-sampling on the same machine as Roon Core looks correct.
But why! Do others out there have a single machine i7-7700 doing both RoonCore and HQP DSD512 up-sampling duties?
Sigh, he scored a free sonicT i7 last year. Now I’ll need to give him my ROCK NUC i7 for smoother Roon Core duties.
I use a single machine with an i7 7700 and this gear including a CUDA GPU, all running Windows 10 in a neighboring room. I run Roon Server and HQP and a monitoring program called Cam on that machine. I can run all filters except xtr at DSD 512; xtr-2s is fine.
I think you’ve isolated the issue (well done), but I don’t know why my i7 7700 is ok and the sonicT isn’t. Looking back over the linked thread I see I was reporting getting some pops with CUDA off. I would think that separating HQP from Roon would be enough without CUDA.
CUDA is an NVIDIA protocol that shares processing with the GPU. HQ Player supports CUDA if you have an NVIDIA CUDA capable GPU. Roon doesn’t support CUDA atm but has generally lower processing requirements.
Noted, I’ve passed this onto Sonore to look at. I’m sure Jesus will communicate with Jussi in the background to look at what can be done (if anything). The next model sonicT i9 (guessing) will probably come with a nice GPU now (guessing again).
Thanks for your help ! My question to you about single vs 2-box solution is what got me to try the latter (yes I’m patting myself on the back there ).
Do you mean why Tidal streaming has such a large impact?
Dad’s initial observation was only Tidal streaming was being affected with pops. But yesterday he confirned (and I noted somewhere above) that the pops is not just limited to Tidal - USB drive playback is just as affected.
And popping gets more intense/frequent when browsing and searching Roon while HQP is simultaneously upsampling to DSD512- probably caused by Roon’s spikes/bursts in CPU usage, I’m guessing.
As noted above - separating RoonServer and HQP Embedded to separate machines solved the issue completely. But Jesus will look at how to get both working on the same machine, when he gets a sonicT i7 to play with.
You can select the same filter for both if you like, then it is just like before.
This is a feature that has been requested every now and then. Thinking behind this is that RedBook and other (48k) 1x content requires fairly steep filter in general because there’s not really much “slack” between 20 kHz and Nyquist, while hires content (depending on one’s view) can trade some of bandwidth for slower roll-off filter.
For example slow roll-off poly-sinc-mqa (or the more MQA’ish poly-sinc-mqa-mp (*)) works fairly well for ordinary hires content, while for ordinary RedBook one could use a somewhat steeper poly-sinc-short. This also allows mixing linear-phase and minimum-phase filter between hires and non-hires content.
*) Even though the filters have “MQA” in their name, they are still much steeper than what MQA delivers. They are just tuned with MQA content in mind.