Phil261575

My GentooPlayer config seems now stable and fine : v8 installed (standard distro), kernel 6.0.7-xanmod-x86_64-GP-XAR-ST-rt14-xanmod1, some optimizations applied, HQPE affinity to the 8 P core, Roon + IRQ + USB (no use) + ETH + WebInterface affinity to the 8 E core, AVX components recompiled as proposed by GP... Overall performances are much better with this kernel: CPU average is 20% when it was 30%. Great! But this "raw" HTOP reading is related to many HQPE parameters that I may have changed, so the real benefit is probably lower (some %). 😁 I still have the exaSound NAA Endpoint not reachable at first start. HQPE log is mentionning "clNetEngine::DeviceLookup(): clNetEngine::Disco(): clSocket::SendTo(): sendto(): Network is unreachable". Once the HQPE Apply button pressed, the endpoint is found. Is it enough to find the problem? Do you need more information or a TeamViewer session? I am available today.

It seems to be OK again now. The faulty configuration was Roon playing Radio and doing Convolution (2 to 6) + HQPlayer doing Upsampling (DSD256). The correct configuration is Roon playing Radio + HQPlayer doing Convolution (2 to 6) and Upsampling (DSD256).

I do have a problem with Live Radio in Roon. My local files and Qobuz streams are played without any issue (Roon server -> HQPE server -> exaSound NAA) but Live radios do not play. Roon starts playback but ends with the error message « Roon lost control of the audio device ». Any idea?

To the new users of EXTRM / CTLO kernels, have you measure a diminution of the CPU load? What is the average raw improvement then?

Does the SSE version of HQPE change something with the use E core?

Thanks. I tried the various combinations and it is OK. I stressed Roon with DSD upscaling without HQPE and I can see that CPU that were set for it were indeed use (HTOP CPU usage). I only keep CPU affinity for Roon, other OS tasks are very low so they will be well spread by the OS.

Thanks, I will check tomorrow if it is also correct on my side. As you said, Roon is not CPU consuming except database build/update and DSP processing. But HQPE is very demanding, so I try to save as much CPU as I can for it. Currently, HQPE is processing Convolution (2 to 6 voices) and Upsampling (PCM to DSD 256). Since HQPE can not use the 8 E cores of my Intel i9-12900T processor except for OS tasks, I am trying to move all tasks but HQPE to those cores so as to preserve P cores for HQPE. One option is also to perform Convolution with Roon and Upsampling with HQPE. I saved some % when doing this 🙂, but thus not enough to move to a more performant Modulator. The ideal setup could be: -2 E cores for GentooPlayer OS tasks -6 E cores for Roon server -8 P cores/16 threads for HQPE. But I don’t know how to setup GentooPlayer so far 🙃…

Yes, Hyperthreding is activated. In fact, I try to simplify my setup: 2 servers, ROCK on NUC8 I7 and HQPE OS on I9 12900T). Since HQPE does not make use of E cores, I wanted to move Roon to those cores to preserve P cores for HQPE (no parameters set for it). If it is not possible or not recommended, I will keep the 2 servers setup since I may have less CPU for HQPE if they are shared with Roon. What do you think?

I am trying the new version of GentooPlayer. Installation and setup is simple and fine. Roon Server and HQP Embedded are up and running without problem (except the automated detection of my NAA end point that I need to analyse). I want to setup CPU affinity for Roon so it uses E cores of my Intel I9 12900T and leave P cores for HQPE. I have enable the function in Set Roon Server and define CPU range 16-23. But after reboot, HTOP displays CPUs assigned to Roon from 0 to 23. Have I done something wrong ? Should the range be defined for physical CPU (0 to 15) instead of logical CPU (0 to 23)?

Hi Miska, my actual and stable setup is ROCK on NUC and HQPE OS on a dedicated i9 12900T server. Everything is stable and simple. However, the 8 Eproc of the i9 are of very limited use. If I remember well, it is because AWX-512 instructions are not available or hidden on efficiency cores. Are you working on an alternative that would make use of those limited but available Eproc?

Another question Miska, have you published an estimation of CPU load of over sampling filters and modulators? It could be a % scale of one filter compared to the one having the minimum load or to the one having a maximum load. The use would be to anticipate incompatible combinations because of hardware constraints (but to be tested to confirm). For instance, the manual gives information on special focus and genre. And I know that my actual combination is OK in regards of focus and genre, and is fine in terms of CPU load. But moving from ASDM5 to ASDM5ECv2 is KO because of the additional CPU load. This comparison between filters could help me finding alternative filters to compensate the modulator additional load. That would optimize the test and learn time. Maybe this information had already been compiled somewhere…

I had in mind that file audio content is file format divided by 2 (I should had mentioned it in my previous post). And knowing that their may not be much recorded content above 96 kHz, I thought the 176,4 kHz filter with 88,2 kHz content could be enough. But if the 352,8 kHz has no or very limited impact on CPU, I will go on with it. Thanks for the very clear answer as usual.

Thanks Miska. I am only performing PCM upscaling to DSD256. I tried DSD convolution with my NUC server once and it indeed failed because of the heavy load 🤪. I once thought of HQPlayer Pro as an alternative to process files once for all with optimum parameters my NUC server was not capable of. But impossible because of Roon convolution including PCM conversion… Most of my PCM files are 44.1 kHz, some 96 kHz, and few 192 kHz. Thus, should the convolution filters be equal to the max but adapted file format (e.g. 176,4 kHz) or to the max but adapted recording content (DXD maximum recording as you mentioned, 352,8 kHz)? If I understand you, it should be recording content so 352,8 kHz.

Additional question for Miska, is the convolution performed with Ecore on an Intel 12th gen proc or Pcore? Don’t know the impact of the convolution against upsampling. With Roon, convolution is light compared to DSD upscale (35x vs. 1.7x).

I am using the Matrix functionnality to perform the convolution (2 to 6 voices + SPL / Group Delay / Room corrections) based on Acourate generated filters. Filters are 131 072 taps and can be generated at 44.1 kHz, 48 kHz, 88.2 kHz, 96 kHz, 176.4 kHz, 192 kHz, 352.8 kHz and 384 kHz. Currently, I am using filters at 384 kHz with HQPE. What should be the most appropriated frequency while processing PCM to DSD 256 upsampling? If I remember well, HQPE automatically upsamples filters to the appropriate rate, but there might be a most suitable rate that would also save CPU cycles.