HQPlayer Embedded Discussion [2021-2022]

ok, understood your strategy on CPU. there certainly would be a big difference with “K” model, but it would depend how far you’d like to push the HQP… at less demanding settings should not be noticable.

general comment on CUDA. only Nvidia type GPU with properly installled drivers would contribute to HQP performance when enabled in HQP settings via CUDA technology. Other GPUs would play no role.

I dont know Matrix USB, but looking at description, yeah - does not look like bottle neck.

And your architecture is Roon PC → ethernet → HQPe → USB → DAC?

In what you described - “OK for 96/24 and not OK for 192/24” I would suspect your set-up hits CPU limit when using gauss-xla filter settings for Nx filter. But even if so, few other things to check…

  • what is your OS / kernel / HQPe version?
  • what is your HQP settings
    a) in the web interface ?
    b) in xml file - did you try multicore=“1”

Do you also use convolution/speakers/matrix?

1 Like

Would you consider the K modell to be superior even if the non-K-model is not far behind at full throttle? Is there any other things onboard that K CPU that makes the whole difference? Could the K model be set to auto levels without tuning and still be so far infront of the std CPU? Do not have so good experience with tuning hardware, since it every now and then usually result in “blue screen”. Also tried the various auto tune modes on several MB’s- But it ends up so often with me not listening to music and instead fiddling with the PC settings. That is not where I should be if my family want to mett my better part…
Ok, so CUDA or nothing … why I ask myself, computing heavy math is performed in any GPU, I would guess? Why not use the one’s present?
Matrix is OK, I think too


Yepp; ROON R.O.C.K. Linux image, but not on Nucleaus, but instead on a powerful silent PC → Ethernet (local enterprise switch, short cables, QoS compliant) → HQPembedded 4.33.0 → USB Card out → short USB cable → RME DAC using DSD Direct mode DSD256 over PCM.
No convolution. Did not think multicore was of any use nowadays, now since the core management is upgraded since quite a few HQPe versions? The SW would activate cores/manage in and out as it see’s fit for the task, is at least how I have understood Jussi’s explanations.
Will try 24/192 with hires-lp for Nx speed before sending this answer:
Yepp, works with hires-lp for this specific sampel rate and bit depth without play back interruptions. Nope, not as good SQ as -xla.

Now, I could try to find faster CPU (I9-10900K), got the Z590, shouold be compatible, have to enforce cooling, but how should I set up the faster K model to in realworld get better performance? Is auto setting in BIOS sufficient? I don’t like the idea with tuning above reference spec, as I mentioned before, that usually result in many other things to do, than just playing music. What is the specific features? I have read that 10900 is in auto mode stepped up to 5,2 Ghz in max turbo power 3,0, while 10900K is 5,3 Ghz in same mode. Is that 100 Mhz that critical? The base speed is far faster with the K, but why should not the HQPe manage to throttle the std 10900 to turbo speed, where the difference is not that much?
Lot of questions, sorry for that, but there are a lot of things not clear to me, when it comes to computer management and I have to have additional info to what I read before understanding the background. Whe reading the spec’s it seems all so clear … :wink:

Because the software needs to interface with the GPU. And not all GPUs and their application interfaces are geared towards general purpose math computations, but instead just for graphics.

Generally, in BIOS:

  1. Use “Load optimized defaults”
  2. Select XMP Profile 1 for your RAM

This should make things perform nicely.

Non-K CPUs are TDP limited to 65W while K-models are 125W or more. Thus, with non-K models you hit the TDP limit pretty quickly. Of course this means that you need at least twice as much cooling capacity too for K models.

If your desired modulators already run fine, but you need more demanding filters, using CUDA offload may improve overall performance without replacing the CPU. However, the GPU needs to be powerful enough compared to the CPU so that CPU doesn’t need to sit waiting for the GPU to complete it’s tasks. As that would be wasted time.

2 Likes

Hi @Lars-Goran_Andersson

changing multicore from auto to “1” could make change on some systems. in your case it seems Nx: gauss-xla is a border line of your system capacity, i would give it a try…

I understand you connect to DAC directly - what is your linux version and linux kernel? If not already, you may want to try:
a) stock lowlatency kernel /or
b) custom Jussi kernel

HQP settings look alright, however two questions

  1. ADI-2 RME supports DSDx48?
  2. Why DoP? Can’t ADI-2 RME do native on ALSA? - although this may not relate to drop outs.

As for 10900 model, I understand your two concerns are:

  • BIOS on Auto, to have all intel automated potential.
  • TDP…

hm, with this you will certainly have more limits compared with K as explained by Jussi above. If I’m not mistaken you can not overclock 10900 with ‘standard’ x-multiplier method, but can’t you remove or change TDP cap in BIOS leaving everything else on auto? This could be indirect overclocking. Well - cooling solution applies.

Also you can try to search/ask on Audiophilestyle forum in HQP threads if anyone have 10900 (not K) and what it can achive to have alternative benchmark for your system to see if anything can be adjusted before you think of changing cpu.

1 Like

Yes…

No, ADI-2 supports only DoP. Not that it matters because it supports max DSD256 and max 705.6/768k PCM (which is needed for DSD256 over DoP). Advantage is that all it’s rates are available on all platforms the same way, including macOS. Disadvantage is that DSD support cannot be auto-detected.

2 Likes

Currently running hq player desktop through roon directly on a mac mini. I’m wondering if there’s a way to repurpose a pi2aes with hq player embedded and place the upsampling responsibility further up in my audio chain. Reason I’d prefer this is because I use a minidsp studio for dirac live room correction but it’s internal sample rate is capped at 96khz. Is it possible to place an hq player embedded pi2aes after the minidsp and before my dac so I can do higher sample rates?

Below is a diagram of my current setup

Hi Vincent, Could you not take the miniDSP completely out of the chain and just use the Convolution options within HQPlayer?

Where are you running this Dirac Live now?

As I use the HQPe (Embedded) it is all Jussi’s HQP and OS in the same image. As per now, 4.33.0 is all new OS base and I assume an all streamlined HQPe on top of this new OS.

Do you think I will make the config, as shown previously above, work as is (with also the Nx setting as gauss-xla) by replacing the I9-10900 to a I9-10900K equipped with a better cooling in a state of just basic MB BOIS settings taking advantage of the K-versions better performance?

Hello

why not use a Linux distribution OS Audiolinux 4.0 with optimized Kernel x86_64 v3 UPMPDCLI + MPD + Diretta ALSA drivers

Sincerely

Ah, HQP OS! Probably the best strategy - system on defaults; HQP OS on a USB stick; No fiddling. Just music and fun :slight_smile:

@ffk , hey, hello again :slight_smile: do you use 10900k in your system with GPU? It could do some rather cool stuff, couldn’t it?

1 Like

The minidsp shd studio does the dirac processing (https://www.minidsp.com/products/streaming-hd-series/shd-studio) and sends out aes connections to my holo may for my mains as well as a topping d90 to run my subwoofers. The minidsp functions as my volume control/dirac dsp/crossover. Maybe I don’t understand how hqplayer embedded works. I was under the impression I’d be able to use just the upscaling functions in a dedicated box and place that between my minidsp and my holo may. The goal being to input aes at 96khz and output i2s at 192khz+. Another question would be, do you know of a ddc that does aes>i2s/usb that could also be flashed with hqplayer os. Sorry if I sound like an idiot and my question doesn’t actually make sense or the way I’m thinking about the configuration is simply flawed.

Here’s a diagram of what I’m envisioning the new setup to be like.

If it comes down to it I might do that but I’m pretty used to the way dirac functions and it’s integrated into my chain nicely already. I also like the fact it runs multiple simultaneous digital outputs so I can send one of them to a cheap dac then to a spectrum analyzer. After testing hq player on my headphone setup I’ve heard a big enough difference I’d be willing to sacrifice dirac if push came to shove. This is not my ideal configuration though. It should also be noted the minidsp acts as my volume control/crossover as well so removing it entirely from my chain would require me rethinking some things and likely buying new hardware.

I have a 10900k but it’s in a hackintosh at the moment. HQP is on a 9900k with an old GTX Titan X, Ubuntu 22.04 (HQP Desktop though, I don’t have an embedded lic). It will do DSD512 with 7ECv2 with all but a few combinations. It is however more capable than a the MacStudio I have. :grin: And alot less expensive.

1 Like

I had thought of using the minidsp myself a few times but it was the samplerate that always made the decision for me.
I’m not sure if its possible but will hqplayer allow outputs to multiple endpoints? If it could you could place an extra cheap endpoint on that feeds your cheaper dac. If its for a spectrum analyser, I’m sure timing would not be that important?
Also you can use the volume control on roon.
Just my 2cent :slight_smile: .

Yes, you can easily input 96k AES to HQPlayer. But I would output at least DSD256 or 768k PCM to Holo May. And you get better results with USB than with I2S.

But your subwoofer feed will be an issue with your current setup.

Some suitable PC with for example RME ADI-2 Pro or RME HDSPe AIO for input.

No, since different DACs typically use separate clocks, these would be drifting vs each other. It could be solved with ASRC, but I’m not fond of using such. But maybe this will be added as a feature in future.

Better option is to use a multichannel DAC for output, such as the new RME ADI-2/4 Pro that has 4 channel output.

Or for more channels, something like Merging NADAC (audiophile) or Hapi (professional) for 8 channels or more.

It seems my current audio chain just doesn’t support hqplayer up to the sample rates I want to achieve. Would you be able to suggest a configuration which keeps my holo may dac, benchmark ahb2 amp, kef ls50 meta speakers, and dual svs sb-2000 pro subwoofers while also maintaining a hardware volume control, a crossover between my subs/mains, and a digital room correction (doesn’t have to be dirac live as it is currently). If possible upsampling to PCM 1.536MHz. If it’s too much of a pain in the ass to think of a viable configuration which features all the above, or if it’s simply impossible don’t worry about it. I’ll continue to use hqplayer with my susvaras regardless of the outcome of my speaker setup. Thank you so much for your responses thus far. Your work has been a blessing to me personally as well as the rest of the community.

Does the Holo May output on both balance and unbalanced simultaneously? If so, you could use the unbalanced output for your spectrum analyser.
For feeding the subs my thoughts could be an analogue active crossover of some sort, This ‘could’ be linked onto the same input as the spectrum analyser? I use one of these myself for my subs.
That would mean just one dac doing the work and not 3.
With the volume control i’m not sure, a passive stepped attenuator would be a good choice, but this is going down the DIY route very quickly.

Again, just throwing out some ideas for you.