I am currently trying to download a new HQPe image. Trying as in trying to get the image home before I shut down the computer tonight … I am a bit concerned, I must admit. It’s been like this for some time, I had the same issues last update too. Anyone else thinking the download speed is in the region of 56K V90 modem call line speed?
Hi Jussi, I just got hold of the latest build (4.33.0) via the SonicOribiter update and was hoping this would fix my matrix problem. Unfortunately it is still the same - sound from left channel only (mono?) for half a minute or so, followed by the addition of the right channel, heavily distorted and starting the track from near the beginning so the two channels are heavily out of time alignment.
OK, this seems to be somehow related to sonicOrbiter. It is already a difficult build because the base OS is a bit old and tricky because it needs to run from Atom CPUs to latest Core, so cannot be built like the AVX2 build on other distributions. Bug either in build tooling or some of the OS libraries.
Obsoleted build tooling on obsoleted base OS makes life a bit difficult.
Then next question is how much performance hit people are ready to take to get this behavior fixed.
Thanks Jussi. Regarding “how much performance hit people are ready to take to get this behavior fixed” i would not want to see any reduction in performance from the present levels. My i9 can run 7ECv2 at DSD 256 and I wouldn’t want to go backwards from there to accomodate having matrix room correction added in. As I mentioned before I can use these filter settings ok whilst also running convolution within Roon, so I am basically happy with the setup as is.
I will keep trying to figure out what goes wrong and how to work around it. So please check with each subsequent release if it begins to work and let me know if it gets fixed.
Hi Jussi, after rounds and rounds of testing I stepped down to 141avx release. In short 12900KS is a King and 141avx is its GOLDEN throne for stable 7ECv2@1024 performance (no CUDA)
I have posted very detailed write up on my set up and tests here, important to mention that contrary to ordinary 12900K, i did not have luck with your custom kernels with 12900KS.
IMPORTANT EDIT
After a little bit of careful settings release - hqplayerd_4.33.0-145avx2_amd64.deb is stable as a brick and runs non stop for 12+ hours with mixed playback on 7ECv2 @ SDM 1024 / 1x: ext3 / Nx: hires-mp and FIR2/Xfi for SDM with exact BIOS OC settings mentioned in my post in CA forum above. No issues!
This is possible with manual configuration in hqplayerd.xml file for “engine”: - multicore=“1” - (manually add) nblocks=“8”
I say in combination with 12900KS - we have HQPe 7ECv2 @ 1024 golden standard from @jussi_laako !
Thanks for the tips.
My low end i5 10400 desktop running Ubuntu Jammy server with 12 gb ram is now doing 7EC (not v2) @ SDM 512 / 1x: gauss-long.
Not yet tried Nx, It’s near max at 1x but working…so far… It never was able to do this before.
Update:
1x: gauss-long / ASDM5EC @ SDM 512x48
Nx: gauss-hires-lp / ASDM5EC @ SDM 512x48
Both are stable on my meager machine.
I think Jussi just bumped hqplayer up another step in processing ability.
@jussi_laako work on HQP performance optimisation is
a) underestimated by many users
b) totally unexpected at times, but comes as sweetest cherry on the cake!
7EC @ 512 works well, 7ECv2 works too at 1x. I am testing the limits of my stock cpu fan, I’ve ordered a Noctua Cpu cooling fan to see if i can make 7ECv2 run cooler.
I actually like the sound of ASDM5ECV2 and ASDM5EC @ 512 with my dac, both work quite well so far.
Will try 7EC 512+fs, forgot about that one.
It was very much capable for DSD512 tasks and actually reasonably silent. For DSD1024 it would be pushed to the max , was loud and generally was not enough for 7ECv2 at this level.
The main thing is, previous to the latest version of hqplayer embedded, I wasn’t even in the conversation of running and trying out the EC modulators @ SDM 512 with my i5.
Great work Jussi.
I am so pleased with HQPlayer Embedded, so before going on, I just wanted to say I am blown away by the SQ improvement!!!
Now, there is an issue I have, I cannot have the HQPe computer to play FLAC 24/192 from Qobuz, through Roon (separate pc), with the ASDM7ECv2, DSD256 over PCM and filter poly-sinc-gauss-xla without short pauses/interruptions in playback. And some of my favourites are only available in that format unless 16/44.1. I have an Intel I9 10900 and fast memories and buying a new computer is not an option (again). Depending in master quality of course, some of the 24/192 display an exquisite SQ and I am sad not being able to playback those files. Is it the pc performance, you think, guys or is it some optimization that I might perform myself? I DO want to use the filter. The combination of the utmost silence in my rig and this filter pitch black background play dynamics on a scary scale …
There are many small or big factors that may cause drop outs, hardware, firmware, software and hqp settings at the end.
If I understood you correctly, let me rephrase - your source is Qobuz via Roon, that is running on separate PC. You stream 192/24 PCM source to HQP embedded machine, and you have drop outs with ASDM7ECv2/DSD256 and poly-sinc-gauss-xla filter?
Do you have GPU and do you use CUDA? In general gauss-xla or ext3 filters are a bit on a heavy side when used as Nx: filter.
Are other filters like poly-sinc-gauss-hires-lp, for example, drop out as well?
Yes, you’ve nailed it. Further to your questions … I will try the hires-lp tomorrow. I have built a quite competent pc, I know about bios, chipset, RAM choice and other factors that become vital for functionality and speed at the bottom line. So the MB is ASUS, would not consider any else except perhaps MSI, the chipset is Z590 to have full customability, but also the widest range of compatibility for any peripheral hardware, ie RAM, etc.
I have everything set to auto, since I have selected the low energy version of the CPU and the auto choices seem to be better if nothing needs to be locked-up. I prefer the 65 watt CPU, as it will not cause any heat issues, the difference is more in the idle/normal speed rather than max speed, so if the software (HQPe) manage the CPU turbo boost functionality and is able on demand to crank up the CPU to max turbo speed, there is not much of a difference in performance between 10900 and its K-version, as far as I have read and understood (?).
I use a peripheral hifi USB controller, Matrix Audio Element H, but that should not make any difference for the ability for sufficient speed, as far as I understand. It is capable of USB 3 interface, far more than required, I suppose? In the PCIE port.
I have no GPU off-load. CUDA that is. I would assume that the included Intel GPU’s on the CPU chip itself, could participate in the heavy floating point maths? OK, it is not the most capable GPU in terms of gaming, but it is not the worst either … I have not, however, allocated any large memory size to onboard graphics, though. Could that be an issue?
I’ve had a period recently without the HQPe PC and there was not any problems for the ROON PC to unpack FLAC 24/192 from Qobuz to PCM in real time and from that ROON PC, I went via the Matrix USB card to my RME DAC, however without any resample. But that proves the ROON PC are able to unpack and playback 24/192 FLAC to PCM in real time.
24/96 or any other 24 bit highres version is no problems and I have not encountered any problems playing back any DSD files from my NAS when received and re-moduled by HQPe to DSD256 over PCM. Will come back tomorrow when I have tried the gauss-hires-lp…