With 1/2 the incoming rate, you have 1/2 the work processing incoming samples, but the same work (roughly speaking) processing the output samples. Of course @jussi_laako can correct me, but my hunch (from my very basic understanding of DSP) is that the number of calculations with 1/2 the incoming rate can be lower, even if the outgoing rate is the same.
It is very possible indeed.
To know as a percentage the difference in demand for a chip like the Intel Core i9 12900K between a non-integer sampling rate and an integer sampling rate i.e. at 192Khz 24bit with poly-sinc-gauss-xla filter would be interesting.
Merging hapi allows me to use only 44.1k-based rates.
@jussi_laako Let’s hope Jussi will be willing to answer this question. 
Technically, same amount of calculations are performed regardless if the conversion is integer or non-integer ratio. But the non-integer ratio is more demanding for RAM access because there is more data to access.
For poly-sinc-gauss-xla, being two stage, when the source rate increases, so does the processing demand, because it also means that the output rate of the first stage increases too. So doubling input rate doubles the load for the first stage.
1 Like
Thanks for your answer.
I would take a liberty to slightly amend and state “some DACs are more discriminating than others” in general. 
In my own experience I see rather different behaviour and receptiveness of different DAC’s to mathematical upsampling and digital corrections (like PEQ for example = Parametric EQ).
Part of my job involves quite a lot of travelling. My HiFi travelling rig is
a) two headphones:
- good old Bose QC25 with noise cancelling (this is for the airplane, public areas and conferencing)
- HiFiMan Sundara (this is for the privte hotel room).
b) As a source I use simple iPhone + Roon ARC enhanced by Oratory PEQ’s suggesed for these two headphone models.
c) As a playback device I have a small quiver of dongles (AKM, ESS, CL chips). Actually when I go on a long trip I take all of them little toys with me.
My “favourite” prefence changing with time and I beleive largely with my mood, this is very subjective and depens on how tired I am.
What can I tell…
The PEQ !!! - the equalizer - actually making the main (and phenomenal) difference in sound. The rest is much less impactful. My ESS dongle is more receptive to upsampling in ARC by giving more “airy stage”, AKM is less so, while CL dongles sound awesome, they really show no difference in flipping upsampling in ARC on or off. So to my experience it is all about engeneering and DAC design. And personal preference.
Why we do what we do? I think it is all about the art of sound and music at the end. As long as you love what you (you, yourself and only you) hear - it is good ! This is most important part of it all.
My own experience with a bit of experimentation is still a “path of least resistance”. Explanation - you probably have noticed the amount of small improvements to HQPlayer universe over the course of time. Simply put - is totally insane. Almost every month there’s a new release of Desktop, Embedded and OS versions for several different platforms. And these are not merely bug fixes, these are enhancements of functionality, one after another. Ofcourse Signalyst (Jussi) tries to cover wide range of devices but the array or volume to cover is, simply put, - enormous. Some time ago I just concluded, since I loved HQP experience so much, to follow Jussi’s reccomendation as close as possible (that always ensures best in class experience) with a bit of detours left and right for experementations that my pocket allows me. And so far this strategy is the best of what I ever could get in audio-enthousiast wold. I love your word - aficionado! Hehehe.
This is by no means by the way to say that your current set up needs any change - not at all, as long as you are happy in your journey this is the best set up in class ! untill you crave for something different that is…
1 Like
We are here in another framework.
I wanted to talk about our home hi-fi listening.
Today I listen very little with headphones except sometimes to spare my wife’s ears when listening which does not suit her tastes.
In addition, the only headphone I still use is an AKG K240DF (purchased in 1987…) whose impedance is 600 ohms and therefore completely unsuitable to work with a travel DAC and an iPhone.
I also use at home the Innerfidelity PEQ (provided by Jaakko Pasanen) suggested for this headset.
Coming back to the subject of our listening at home.
Do you use Parametric EQ or convolution in Roon or in HQPlayer convolution or pipeline setup?
Luckily, I don’t use any of this at home, the geometry of my living room and its environment (carpets, curtains, bookcases, etc.) as well as the positioning of my speakers do what is necessary for a plausible truth of the restitution of instruments and music.
That said, live music remains incomparable.
I clearly specified discriminating in the use of HQPlayer and not in absolute terms.
If this were not the case, I would not understand that the new Jussi’s DAC correction support does not integrate the DAC (Merging hapi) model that I use.
It seems to me that any user (customer) of HQPlayer should be able to benefit from all the technical improvements of the product when these improvements are relevant.
This is why I prefer to think that this improvement is not relevant for Merging hapi.
@jussi_laako But perhaps Jussi may tell us a word about it, I mean about this revelance or not.
I completely agree.
@IgorSki Thanks again for this discussion.
It is not a question of knowing who is right or wrong, but simply of discussing without being peremptory and trying to understand things since music and audio reproduction are our pleasure (with a little computer science along the way 
).
That’s what you do.
It would be, but I don’t have Hapi anymore, so I cannot make correction profiles for it either…
It is indeed difficult for you to have all the existing DACs at home.
If by chance you were able to recover the one you had, just long enough to carry out your test measurements (correction profile) and integrate it into the DAC correction support, I would of course appreciate it.
I have many more, but of course I cannot have all the DACs. I will make more available when I get the cloud things done and find time to make those profiles.
And I prefer to have constant access to the devices I make corrections for.
This point is a bit turned on me, and I would not like to have to buy a DAC just to benefit from your latest developments in your software even if I understand your point of view.
@Stephane, yes I use the digital room correction. Not really sophisticated. I make a moderate correction in low frequences range 40Hz - 1000Hz. This correction is fixing couple of artifacts, I have few nasty bass spikes and drops around my listening position (sofa that is) and this correction helps to bring system to mellow full flavour experience. As a result I don’t need to use the “sub”. I used to have this correction in HQPlayer convolution section. Now I have moved it to Matrix profiles. Also my room in not symmetrical and I use speaker distance correction in HQPlayer as well.
Yes of course, you are correcting the headphones flaws (and/or changing to more your taste in a much bigger way than other DSP).
Like room correction too
Even though the latest hqp version fixes the memory errors, the dac correction feature is still very demanding in terms of processing power. It does fix the memory usage problem, but I desperately hope the author continues to optimize hqplayer, which is software I can’t live without every day.
@Stephane Hi Stephane
The combo you demanded is very CPU task intensive. I have a Merging Hapi Mk2. My system could play 192KHz contents using Nx: gauss-xla + ASDM5EC-super + CPU-only without stuttering.
On my special-tuned 7950x server you can see the high CPU utilization rate without GPU:
If you insist to play 192 content using the combo you demanded especially without OC, a single 12900K probably would not support that anymore. Get an NVIDIA 3080ti GPU or higher would help. Or switch platform to Linux and try HQPlayer Embedded (HQPe can squeeze more CPU raw power, but this part you need to know how to build the Ravenna/AES67 ALSA LKM).
2 Likes
@Chunhao_Lee Hi Chunhao,
Thank you for your reply.
My request is a bit like that of a spoiled child, there are so many other possibilities with HQPlayer.
With my current computer I can play 96kHz using Nx: gauss-xla + ASDM5EC-xxx + CPU-only without stuttering.
I don’t want to change OS like Linux or use HQPlayer embedded for now.
The solution would indeed be to have a more powerful GPU but for the moment I play 192 kHz with another filter like the default poly-sync-gauss-hires-lp.
What I regret is that Jussi does not offer a DAC correction support for Merging hapi, and I would be curious to know the improvement that this would bring to listening in these conditions.
Maybe a big difference or maybe not much…???
I also do not understand what Jussi’s methodology was for carrying out these DAC corrections.
@jussi_laako Could you please explain your methodology, if possible. It may be secret.
Hi,
continue this thread instead of starting a new with basically similar question about performance. Today I manage with an I9-10900, offering 10 cores and 20 threads. Newer CPU’s have this hard-to-understand performance and energy cores.
I will likely get hold of a motherboard with a LGA1700 socket. It will open options for a better CPU, and perhaps a bit quieter operation.
But, what does I9-12900K mean really? It lacks two “performance” cores compared to the one I use now, but the total cores and threads are larger than mine.
Is there any gain from this swap at all, or is it just make beleive?
btw I will move over my nVidia RTX3060Ti graphics adapter off course.
The computer as is today, is a bit loud, but is stable as a rock and no glitches. Perhaps I am just creating potential problems?
@Stefan_Jorma what is your goal or expectation in making the swap to newer CPU or hardware ? Quiter build, higher upsampling rates, heavier filters ?
PS: in two channels listening, in particular, the HQPlayer performance (in terms of ability to upsample to higher DSD rates for example, and running demanding modulators because these can not be offloaded to GPU) depends on single core performance of CPU. Intel is still getting higher and higher single core performance from one generation to another. “K” models are “unlocked” so you can squeeze even more, and “KS” are specially binned, normally getting even slighly more performance per single p-core. This helps to move from SDM@256, to 512 to 1024 and 1024x48, etc etc… wheather one needs that is another question.
1 Like
E-cores can offer an “alternative” for GPU. Especially on processors like i9-14900K(S) that has 16 E-cores. For cases when GPU is not available. Also processors like i9-13900T/14900T can provide rather impressive performance without GPU and lower TDP. My 13900T can do DSD1024 with default filters and ASDM7EC-light. Difference is that 14900K can do the same with ASDM7EC-super without even breaking a sweat.
One area I’m still going to improve is combining E-cores with GPU.
Instead of rather old i9-12900, I would rather go with i9-14900.
From HQPlayer perspective, E-core is maybe half the performance of P-core. But combination of 8 P-cores and 16 E-cores is close to having 16 P-cores for certain processing cases, while having lower power consumption. This is because each E-core still sports a complete AVX2 block.
3 Likes
I am looking for quiter build and some future proof level/performance overhead. Do not need more than DSD256, I do not plan to replace my RME DAC, as it is still the best I have heard in that price level. Further it immediatly responds to all my efforts, trials and errors when getting the network and other digital transmission better and better isolated from noise and ground issues. But I would like to go for a Windows application next update and plan accordingly. That means a portion of performance overhead compensating the ongoing Windows processes. Made my friends pc ready today after completely falling flat with the Ubuntu/Roon/HQPlayer for a month. The Windows installation with Roon and HQP took a couple of hours, nice chat, coffee&cake and a few laughs.
So, just planning ahead
1 Like
Thanks! Good explanation. The price level of the I9-14900K is for me a bit expensive, nothing I take on a salary. When I look at benchmark performance between I9-13900K and I9-14900K they seem almost alike, but perhaps the CPU architecture is different and more beneficial for HQPlyers performance?
Further the I9-12900K is low on E-cores, but offer 8 performance cores also, even if I fully understand they are not same action per core as 13 or 14900. It is as always, is it good enough?
Do HQP Desktop run smoothly on Windows 11 Pro? With let say an I9-12900K, 32GB RAM@4000 Mhz? Will all filters/modulators probably run @ DSD256?