Too bad Roon still displays “filter” incorrectly for DSD sources… 
It should be saying “XFi” as rate converter there. If you compare to what HQPlayer Client says.
Surprisingly low load figures still, and temps are not bad at all either!
I have replaced Noctua NH-D15 Chromax (air cooling) with Arctic Liquid Freezer II - 360 - (AIO Cooling). At 100 bucks Freezer is just a beast and kicks a…! My jaw dropped. Sadly Noctua will have to go. a) at DSD1024 noise levels are in favour of Freezer, even though I had to open additional air pathes in my case to have this installed b) the Freezer performance is increadibly sustainable, this temperatures stays like this after hours of DSD1024
Anyway, I will post more tech observations @CA “Best CPU thread” in coming days.
But the point I continue to make - the whole 12900K set up did not break 1000 bucks, Ok depends on location and market, but I’m based in Suisse, not the cheapest market of all - the whole thing is very affordable and results are just insane.
2 Likes
Theoretically it could be also possible with AMD CPU, or maybe next gen at least.
Because looking at loads of 5800X running at DSD512, it would not be horribly far from DSD1024 (double loads):
3 Likes
Based on Ares II measurement, -3 dB is pretty accurate. I got -3.1789 dB difference through measurement. I will add -3.2 dB or -3.18 as gain compensation figure to the manual for next release. (re 1 kHz sine wave)
2 Likes
Hello, wish you all a nice Sunday morning 
Since a few days I own an Audio-GD R28, and since then my ifi stack collect dust…
The device is the best in conjunction with HQPlayer - for me! - what I was allowed to hear so far.
I only use PCM upsampling, with these filters:
I’m seriously considering buying an R27, or for my R28 the DI-20HE as a DDC, we’ll see what time brings
I use an Asus Tinkerboard as NAA Bridge, and think that with a good DDC and transmission via I2s the quality would improve again.
Any opinions or experiences from you?
Best regards from Germany
Thank you for taking the time to measure the difference in levels.
Holo may , dac bits 20 , sinc M / sinc Ll / sinc Lm, LNS15 1.5m.
how does the new sinc Ll and Lm compare to M mechanically?
sinc-Lx family are non-apodizing and have just average stop-band attenuation close to Nyquist. These also look quite similar to Chord’s filters. If you’d like something similar.
sinc-S and sinc-M(x) family are apodizing and have very high stop-band attenuation already close to Nyquist.
How does going for short to long with these filters change the sound (in rough terms)? The manual recommends short for electronic/rock/etc, the medium lengths with any genre, and the xl for jazz and classical. Why is this?
For me, the longer filters appear to have more depth of soundstage, whereas the shorter ones sound flatter. As a result I’m inclined towards using xl(a) for all genres, but wondering if there’s something about the shorter filters that I just haven’t caught on and might be missing out on.
Shorter filters have shorter time domain response, which is good for transient reproduction, such as cymbals and such. Longer filters are not as good in this respect, but have steeper cut-off and tend to be better in such aspects as space.
Electronic and rock are typically multi-track studio productions, not recorded in real acoustic spaces, but rely heavily on transients such as frequent cymbal use.
While classical is typically recorded with fewer microphones in a real acoustic space and the acoustics of the space play more important role. And classical music rarely contains such transients and has much less heavy reliance on such.
This is the personal aspect that is also important in filter selection. Different people are sensitive to different aspects in the sound.
For example when I was developing filters such as poly-sinc-short-mp, I was using for example Pink Floyd’s Meddle album a lot, and particularly listening for things like snappiness of the percussions on San Tropez track. While that track doesn’t have particularly strong reliance on depth of soundstage - for me at least.
There are some rough suggestions in the manual (as requested) when certain filters may be most suitable. But the final decision what suits best is always a personal choice. It may be different from those suggestions.
10 Likes
Thank you. In a similar vein, how would you characterize the differences between lp vs. ip vs. mp for poly-sinc-gauss-hires?
In this case the manual does not provide any differences in “special focus” or genre recommendation.
It is about trade-off between pre-ringing and phase response. It is separate thing from filter length, while both affect each other. It has similar effect as described about filter lengths.
Minimum-phase filters don’t have pre-ringing, but have effect on phase response. Linear-phase filters have 50% of the length pre-ringing. Intermediate-phase filters are between the two.
You can try to listen for the leading edges of high frequency transients, such as cymbal snapping on the track I was talking about earlier. And compare how much you hear difference in transient definition and rise time speed vs space (soundstage). If you cannot decide, go for the linear phase variant. This is likely also thing that varies between genres, or albums and maybe sometimes even between tracks.
6 Likes
One question the 
on navigation bar is an “inverting attenuator”?
I’m trying to understand the basics of digital audio, and there’s lots of technical jargon. My question is what’s the function on using the inverting attenuators on HQPlayer.
Many thanks
That is convolution engine on/off switch.
Next to it, there’s the phase invert, single track repeat, all track repeat and random playback.
1 Like
Thanks, Jussi!
Thanks Jussi! Fast reply after our forum blackout yesterday!
Hi, I’m loving HQPlayer so far and it’s really spurred me to dive deeper into digital audio and filter design. Very illuminating! I have one thing I don’t quite understand and I was hoping someone may shed some light on it.
My current understanding that the tradeoff between short and long filters is short filters cause the least pre/post ringing giving ideal transient response, while a longer one will be better at rejecting aliases and out of band noise leading to better stage and timbre.
For me there is an apparent paradox to this: The Nyquist filter.
The Nyquist interpolation filter is an “ideal” lowpass filter that we cannot construct because it is infinite in length. Because it is infinite in length, it thus also has infinite pre and post ringing. This filter however is mathematically proven to perfectly reconstruct the analog waveform which also means perfect transient response, despite it having pre and post ringing.
Now the closest approximation to this filter is an extremely long Sinc filter which is windowed appropriately, which is also the philosophy behind Rob Watt’s filter designs.
So if we take Sinc-L (which I hope is a windowed sinc? documentation is sparse on this) and upsample x16 we’ve got a 2 million tap filter and I would think that the difference between the ideal sinc and this windowed sinc is vanishingly small. If we then have an almost perfect approximation of the ideal filter does this not almost give us an almost perfect transient response?
I know that if you take an unwindowed sinc and simply truncate it after 2 million taps you still get lots of filter ripple but that’s what the windowing function is for. I don’t really see an alternative explanation for less-than-ideal transient response.
Any explanation would be greatly appreciated!
It would also require infinitely long signal. And in addition infinite sample precision. For perfect reconstruction of such infinite signal. But it would be also infinitely bad on infinitely short transient. Real physical world doesn’t have infinite things though.
Main problem here is that the theory says that signal can be perfectly reconstructed if the sampling rate is sufficiently high to contain all frequency components. However, this is root of the problem, especially RedBook is severely band-limited that modifies the original signal by band-limiting, which introduces ringing.
It is windowed yes.
Such filter would be almost ideal in frequency domain, but completely opposite in time domain.
Ideal filter would be 1 tap long, since it wouldn’t modify the original data at all. It is so called all-pass filter. However, it doesn’t filter anything at all either, so it does nothing. But it doesn’t have any ringing at all either.
Longer the filter, more it looks into future and history, and more the future and past affect also the present. Ultimately, for transient performance, what we are interested at is step response of a single step.
See for example also here:
Let’s say you have a single step event in signal which is otherwise DC. With a million tap linear phase filter, the step already begins to affect output 500 000 samples before the step is going to happen. This is especially unnatural, because in real physical world, you won’t have something affecting your perception before it has even happened.
3 Likes