HQ Player

[louisxiawei]

42 minutes ago, Sam Lord said:

Just stopping by, please forgive my ignorance: Why is anyone trying to change the base clock, i.e. 44.1k to 48k or vice-versa?  Is it because:

 

A) People are listening to files with different base clocks (like 24/96 and 24/88.2), and any change in the base clock screws up playback for some reason?  IIRC, lots of dropouts occurred in many systems for this reason.

B) Are some of the DACs only using one base clock rate, so HQP (or another software renderer) needs to convert to that for best DAC performance, i.e. minimize any computation at the DAC?

 

Both reasons?

Thanks!

Not ignorant at all.

 

I first found out upsampling 44.1 → 48 base has quite significant SQ improvement is the time when I use exasound e12 with HQplayer. 44.1 → 48 x 256 upsampling has more airy, better image and smoother sounding compared to 44.1 → 44.1 x 256, that was one of the main reasons I purchased the T+A DAC8 for DSD512, because such little rate change made such obvious SQ difference.

 

Then I discussed this with Miska, he explained that there might be a few reasons why 48K upsampling is better than 44.1.

 

1. 48K clock performance is better than 44.1 in the dac.

2. 48k-base has higher sampling rate. Higher the rate multiplier, bigger the difference. At DSD512 the rates are 22.5792 MHz and 24.576 MHz , which is 1.9968 MHz difference, more than half of the rate of DSD64.

 

IRCC, my previous exasound e12 only has one clock for both 44.1 and 48K files, so I think the higher 48K-base is the reason which made me feel the SQ is improved. At least, I feel it as an improvement. Other people might have different opinions.

 

There is another funny thing I sense is that: Although the difference between DSD256 and DSD512 on dac8 is significant, the SQ difference between (44.1 → 44.1 x 512 vs 44.1 → 48 x 512) is smaller than the (44.1 → 44.1 x 256 vs 44.1 → 48 x 256), if you know what I mean.

 

For Holo spring DAC, it's another case because the dac cannot support 48K file on windows system due to XMOS firmware, so he has no choice but to upsample 48K to 44.1 based files.

 

Hope it helps a little.

 

 

 

 

[Sevenfeet]

On 1/15/2018 at 1:23 PM, Miska said:

 

Yes, and no, but it depends on both hardware in question, HQPlayer configuration and type of content you are playing... It is unlikely that you would save any money by doing so. But for example in this convolution offload case it helps if I use poly-sinc-xtr to DSD512 on the i7 6950X and want room correction too with a bit more complex filters (I have GTX 1080 on that machine). I still cannot get 5.0 channel DSD256 to stereo DSD512 downmix in realtime. So I need more CPU or GPU power or more clever way to utilize both CPU and GPU in parallel. I'm expecting quite a lot from the new Volta-generation Titan V, although it is quite expensive. So in some cases it helps you keep CPU clock frequencies higher because you can move part of the work to GPU.

 

 

I'm talking about physical cores. Hyper-threading doesn't really help HQPlayer, but it can help the OS to some extent. My guesstimate is 10 - 15% performance increase.

 

For those CPUs, cache sizes probably makes bigger difference in this case.

 

 

6700K is just quad core. But why so old CPU instead of 7700K?

 

 

I don't have any machine with DDR3 running HQPlayer anymore, so I cannot comment on that. But check out RAM with high clock speed and low CL values.

 

For example this Kingston HyperX Predator DDR4-2400 kit offers quite nice XMP profile with CL12:

https://www.kingston.com/dataSheets/HX424C12PB3K4_32.pdf

 

 

One thing I think we need to keep in mind is that Miska is a one man shop so it's unrealistic for him to have a matrix of a bunch of different motherboards, CPUs and graphics cards along with the filters they can perform without drops/distortion.

 

That being said, there are a bunch of clues based on his intimate knowledge with the source code.  If I were building a custom HQPlayer machine, I think it would be much like a video editing machine since both are dependent on processing a lot of data in a timely manner....video editors want high resolution/color mapping without frame drops...we want the ability to process music files effectively in real time.

 

At the core of all of this is...well, the cores.  You want something with a high enough clock speed balanced with enough cores to do the job.  However, going to 18 slower cores in the newer Intel designs may not be the proper choice according to Miska.  This would be a departure from some video editing suites you might build.  But then after that, everything is about feeding those cores....bus speeds, memory speeds and storage.  Don't scrimp if you can.

 

I get the whole hyperthreading thing.  Some programs really like it and some don't see as much bang for buck.  I remember when hyperthreading was a new thing many years ago and the first version literally made my company's enterprise software package slow down...we had our customers turn it off entirely.  That's not the case anymore, but don't expect to see 2X speeds...it's unrealistic even in the best circumstances.

 

The AMD Threadripper possibility is intriguing to say the least...and may be a cheaper machine to build.

 

For the rest of us on a budget, the -2 filters and DSD256 are a good compromise.

 

Also, Miska is trying to deal with upsampling multichannel DSD to DSD512...not easy but also not the needs of all of his customers (or even most).

 

One more note...I think the next major performance advance will be PCIe 4.0 which we should hopefully see later this year.  The faster bus speed should allow cores to be fed faster and further eliminate bottlenecks.

[seatrope]

@louisxiawei @Miska

Hi Louis,

 

i fully stand corrected. You are absolutely correct, and the info from the Headfi thread is also good.

My Threadripper 1950x either on windows or Linux cannot upsample poly-sinc-xtr non-2s pcm 44.1 to dsd512 @ 48khz without stuttering. Just checked it today and no go, the initializing bar goes forever, maybe 4-5 minutes then playback starts with a lot of stutter.

 

I am sorry that I gave wrong information. I must have had it on poly-sinc or some other easier filter.

Yeang

[Miska]

Performance test results for the new Nvidia's Volta-series Titan V show such huge performance step in scientific applications, that I'm hopeful that it could actually handle that case...

[shadowlight]

I have Lampi B7 with chipless DSD512 and I am able to play both 44.1k and 48k base files up converted to DSD512.  My system is not powerful enough to do xtr but xtr-2s works just fine.  My configuration uses auto.

[seatrope]

@shadowlight thanks for the info!

 

i have a Lampi gg2 and am having trouble with NAA streaming of DSD512. Can you enlighten me as to what you exact setup is?

Windows or Linux? Usb direct or NAA/SMS-200/microrendu?

what buffer settings do you use?

which firmware?

 

thanks!!

yeang

[shadowlight]

10 hours ago, seatrope said:

@shadowlight thanks for the info!

 

i have a Lampi gg2 and am having trouble with NAA streaming of DSD512. Can you enlighten me as to what you exact setup is?

Windows or Linux? Usb direct or NAA/SMS-200/microrendu?

what buffer settings do you use?

which firmware?

 

thanks!!

yeang

Yeang,

 

I will send you the screen shot when I get home.  My setup is as follow

 

Windows 2012 Server running Minimserver, LMS, Music Storage --> HQPlayer Embedded running under Ubuntu --> Windows 10 NAA (stripped down based on the optimization thread on CA) --> Lampi B7

 

I believe the Amanero is running firmware 1096 (I am waiting for Amanero to release code that is supported both under Linux and Windows for DSD512).  As soon as they do that and Lukasz and team, certify it for release I plan to switch over to Linux based NAA.

 

My HQPe system is Lenovo ThinkServer with Xeon processor with clock speed of 3.4, 16gb of memory (I think), with SSD drive running Ubuntu with Jussi's customized kernel.  The NAA is Lenovo Thinkcentre i3 with either 4 or 8gb of memory and ssd drive.  Everything plugged into wired Gig switches.  I will most likely flip the NAA over to wireless when I switch over to Linux based NAA.

 

What are the issues that you are running into with NAA?  I believe you might be using Linux based NAA, which I have not tested.  In my current setup I am only able to do DoP which is restricted to DSD128.

[Sevenfeet]

2 hours ago, shadowlight said:

Yeang,

 

I will send you the screen shot when I get home.  My setup is as follow

 

Windows 2012 Server running Minimserver, LMS, Music Storage --> HQPlayer Embedded running under Ubuntu --> Windows 10 NAA (stripped down based on the optimization thread on CA) --> Lampi B7

 

I believe the Amanero is running firmware 1096 (I am waiting for Amanero to release code that is supported both under Linux and Windows for DSD512).  As soon as they do that and Lukasz and team, certify it for release I plan to switch over to Linux based NAA.

 

My HQPe system is Lenovo ThinkServer with Xeon processor with clock speed of 3.4, 16gb of memory (I think), with SSD drive running Ubuntu with Jussi's customized kernel.  The NAA is Lenovo Thinkcentre i3 with either 4 or 8gb of memory and ssd drive.  Everything plugged into wired Gig switches.  I will most likely flip the NAA over to wireless when I switch over to Linux based NAA.

 

What are the issues that you are running into with NAA?  I believe you might be using Linux based NAA, which I have not tested.  In my current setup I am only able to do DoP which is restricted to DSD128.

 

Just to be clear what you're doing...

 

You're using Minimserver for DLNA and Logitech Media Server to handle your music streaming off a Windows 2012 server?  And HQPlayer Embedded is running off a different machine and you have a Windows 10-based NAA on a third machine?  Did I miss anything?  How is HQPlayer getting access to the music files?  Just curious...I've been trying to figure out what my own final HQPlayer setup will look like.

[shadowlight]

1 hour ago, Sevenfeet said:

 

Just to be clear what you're doing...

 

You're using Minimserver for DLNA and Logitech Media Server to handle your music streaming off a Windows 2012 server?  And HQPlayer Embedded is running off a different machine and you have a Windows 10-based NAA on a third machine?  Did I miss anything?  How is HQPlayer getting access to the music files?  Just curious...I've been trying to figure out what my own final HQPlayer setup will look like.

 

Correct.  Using BubbleUpNP to send stream over to HQPlayer Embedded.  The 2012 server is also my NAS server.  I still have older Squeezeboxes around the house and I am trying to get Deezer streaming working with LMS UpNP Bridge and Ickstream plugin so I can dump Tidal.  Over the next couple weeks I plan to add JRiver and BubbleUpNP Server on the same 2012 system so I can use JRemote/EoS (Nicer interface compared to BubbleUpNP for ripped music) remote on tablet and Lumin/Linn remote apps to send stream to HQP embedded.

 

If you are using using beta 16 or higher of embedded it how has library function similar to desktop and you can map a drive from your NAS/Windows server to embedded system and use that mount point for you ripped music to play.  I have not tried that functionality yet, but I believe you will need the recently release Android app to play files.

 

In my setup the only thing that is in the music room is the low powered NAA and everything else is outside of the room.

 

Setup guide.

On 10/26/2017 at 9:12 PM, shadowlight said:

Guide with information updated for the 4.0 beta including configuration using the web interface that was introduced late in the alpha stage.  Please let me know if you find any error or I missed something that is crucial.

 

HQPlayer Embedded Install Guide v.0.5.pdf

hqpe-startup.tgz

 

Edit: Forgot to ask are you using HQPlayer Desktop or Embedded and which OS are you using?  If you post information on your ultimate goal and maybe we can make some suggestions. 

[Sam Lord]

Thanks very much!  And your point about the base rate change being less critical at DSD512 than 256 makes sense, obviously when one can compare.  That said, I don't see the logic in Miska's point #2.  22.5792 MHz vs. 24.576 MHz is just Godzilla versus his twin.  Okay, I do see it, but note the computation required to change base rates without audible rounding errors, as we have to consider a $3k Volta GPU card to get it done along with 8x oversampling with the best filters (and no EQ)!  Geez, even the Nvidia Titan won't do that, it seems.  I just think someone would need God's ears to hear that ~10% rate increment.

 

On 1/17/2018 at 6:48 AM, louisxiawei said:

Not ignorant at all.

 

I first found out upsampling 44.1 → 48 base has quite significant SQ improvement is the time when I use exasound e12 with HQplayer. 44.1 → 48 x 256 upsampling has more airy, better image and smoother sounding compared to 44.1 → 44.1 x 256, that was one of the main reasons I purchased the T+A DAC8 for DSD512, because such little rate change made such obvious SQ difference.

Then I discussed this with Miska, he explained that there might be a few reasons why 48K upsampling is better than 44.1.

 

1. 48K clock performance is better than 44.1 in the dac.

2. 48k-base has higher sampling rate. Higher the rate multiplier, bigger the difference. At DSD512 the rates are 22.5792 MHz and 24.576 MHz , which is 1.9968 MHz difference, more than half of the rate of DSD64.

 

IRCC, my previous exasound e12 only has one clock for both 44.1 and 48K files, so I think the higher 48K-base is the reason which made me feel the SQ is improved. At least, I feel it as an improvement. Other people might have different opinions.

There is another funny thing I sense is that: Although the difference between DSD256 and DSD512 on dac8 is significant, the SQ difference between (44.1 → 44.1 x 512 vs 44.1 → 48 x 512) is smaller than the (44.1 → 44.1 x 256 vs 44.1 → 48 x 256), if you know what I mean.

For Holo spring DAC, it's another case because the dac cannot support 48K file on windows system due to XMOS firmware, so he has no choice but to upsample 48K to 44.1 based files.

Hope it helps a little.

 

[shadowlight]

Here are my settings for HQPe.

[shadowlight]

HQPe Server is running Intel E3-1225 @ 3.2GHz and 20Gb of memory.

 

[louisxiawei]

2 hours ago, Sam Lord said:

Thanks very much!  And your point about the base rate change being less critical at DSD512 than 256 makes sense, obviously when one can compare.  

 

 

I think Miska is also aware of that well. As he said during the discussion:  the practical benefit is relatively more significant at those lower DSD rates because there even smaller rate increase makes difference, which perfectly matchs my listening experience about rate changes at DSD256 vs DSD512.

 

Quote

Okay, I do see it, but note the computation required to change base rates without audible rounding errors, as we have to consider a $3k Volta GPU card to get it done along with 8x oversampling with the best filters (and no EQ)!  Geez, even the Nvidia Titan won't do that, it seems.  I just think someone would need God's ears to hear that ~10% rate increment.

 

From 44.1 upsampling to 48 x 512 is still audibly different compared to 44.1 x512, not that night and day different but it's not subtle, no need god's ears or some people think I have placebo effect .

 

I agree with you regarding the cost for running HQplayer. For xtr filter, my titan Xp can do nothing for the offload since it cannot catch up the speed of the CPU. At the moment,  all heavy load filters are run by CPU alone, thus the reason I purchased Titan Xp for HQ computation is being wasted, I should have purchased GTX 1080 Ti just for gaming. But the effort of extra £1000 of Titan V for running 44.1 → 48 x 512 is too much for my wallet now and whether it can be done or not still remains unknown. However, I have no regret for purchasing 7980XE for running non-2s filter at DSD512 in terms of SQ improvement. It can run Xtr DSD512 while many other tasks can be done smoothly including playing graphic-demanding games at highest graphic settings.

 

But hey, computer stuff is not like Hifi gadgets that remain the high-price tag all the time. In a few years, powerful CPU and Volta series gaming graphic cards that can run xtr filter will be wide-spread and affordable. Maybe then Miska will create more powerful filter. 

 

 

[louisxiawei]

deleted.

[Miska]

7 hours ago, Sam Lord said:

Thanks very much!  And your point about the base rate change being less critical at DSD512 than 256 makes sense, obviously when one can compare.  That said, I don't see the logic in Miska's point #2.  22.5792 MHz vs. 24.576 MHz is just Godzilla versus his twin.  Okay, I do see it, but note the computation required to change base rates without audible rounding errors, as we have to consider a $3k Volta GPU card to get it done along with 8x oversampling with the best filters (and no EQ)!  Geez, even the Nvidia Titan won't do that, it seems.  I just think someone would need God's ears to hear that ~10% rate increment.

 

You can do the same with -2s filters on much lighter hardware to compare and hear the relative difference it makes for you (if any)...

 

[Sam Lord]

10 hours ago, louisxiawei said:

I think Miska is also aware of that well. As he said during the discussion:  the practical benefit is relatively more significant at those lower DSD rates because there even smaller rate increase makes difference, which perfectly matchs my listening experience about rate changes at DSD256 vs DSD512.

 

 

From 44.1 upsampling to 48 x 512 is still audibly different compared to 44.1 x512, not that night and day different but it's not subtle, no need god's ears or some people think I have placebo effect .

 

I agree with you regarding the cost for running HQplayer. For xtr filter, my titan Xp can do nothing for the offload since it cannot catch up the speed of the CPU. At the moment,  all heavy load filters are run by CPU alone, thus the reason I purchased Titan Xp for HQ computation is being wasted, I should have purchased GTX 1080 Ti just for gaming. But the effort of extra £1000 of Titan V for running 44.1 → 48 x 512 is too much for my wallet now and whether it can be done or not still remains unknown. However, I have no regret for purchasing 7980XE for running non-2s filter at DSD512 in terms of SQ improvement. It can run Xtr DSD512 while many other tasks can be done smoothly including playing graphic-demanding games at highest graphic settings.

 

But hey, computer stuff is not like Hifi gadgets that remain the high-price tag all the time. In a few years, powerful CPU and Volta series gaming graphic cards that can run xtr filter will be wide-spread and affordable. Maybe then Miska will create more powerful filter. 

 

 

 

Thanks for your impressions Louis.  I've had ear problems for months, and in that time my preamp and whole audio computing system (including NAA) has been kaput.  But the ears are better, so I need to make decisions; right now I'm using an Oppo 105D as both dig preamp and DAC, not ideal.  I have a preamp now, will install pretty soon.  I expect I'll end up near where you are, running DSD512 with a DAC that can change base rates.  The Schiit Yggy beckons, but I love lots of DSD titles and by all accounts the DSD-PCM conversion has not been a beneficial one if it can be avoided.  Kinda wish I had jumped on the Exasound deal earlier, but I think DSD512 isn't available on the E32(?).  On the third hand, some of the top innovative labels report best results by recording at DXD and converting to DSDx2, 4, etc.    

 

5 hours ago, Miska said:

 

You can do the same with -2s filters on much lighter hardware to compare and hear the relative difference it makes for you (if any)...

 

Indeed.  I've learned to believe some pretty extraordinary things, based on the huge gains I got from using HQP to convert redbook to DSD128 using some -2s filters, so I'll accept your experience!  Probably will do the usual, make some tests like that, then try a few recommended settings that my new streamer/renderer could handle, and call it good.  Determined to get to DSD512 in any case. 

[57gold]

 

Mac Mini late 2012, 2.6 GHz Quad Core i7, 16GB RAM.  Files on a 4TB La CIe HD attached to Thunderbolt bus.  Mini is attached to Mytek Brooklyn DAC+ with an AQ Coffee USB cable.

 

Can't up convert 44.1 PCM to DSD128 with xtr filter even though Activity Monitor shows that I'm only using @30% of CPU.  It starts playing, sounds sweet for 5 seconds, then trails into stuttering garble.  What am I doing wrong?

 

[louisxiawei]

47 minutes ago, Sam Lord said:

The Schiit Yggy beckons, but I love lots of DSD titles and by all accounts the DSD-PCM conversion has not been a beneficial one if it can be avoided.  Kinda wish I had jumped on the Exasound deal earlier, but I think DSD512 isn't available on the E32(?).  On the third hand, some of the top innovative labels report best results by recording at DXD and converting to DSDx2, 4, etc. 

Consider the top three DAC recommendations on the HQplayer website. exasound is a wonderful DAC with HQplayer while T+A DAC8 DSD is the commercial product closest to the DSC1 based on the same principle 1-bit converter , so if you are determined to get DSD512, your choice is simple. 

 

Quote

some of the top innovative labels report best results by recording at DXD and converting to DSDx2, 4, etc. 

 

Sounds like what  "spirit of turtle" said. Recordings from that website are stunning. I talked to Bert van der Wolf from spirit of turtle via email a while ago and he told me that all their recordings since 2005 are DXD origin either with dCS or Merging ADC, so DXD is the original in all cases and they do further convert DXD back to DSD64/128/256 for sales because its claimed reason: https://www.spiritofturtle.com/what-is-high-resolution-audio/?v=69e1aafeccc5

 

For HQplayer upsampling, you'd better go for the DXD version since it can take advantage of the SDM filter while DSD files only use modulator. 

 

and don't forget website like nativeDSD, they have album recorded in DSD256. In general, try best to get the original sampling rate/bit depth recording if possible. 

 

 

[lmitche]

3 hours ago, Sam Lord said:

 

Thanks for your impressions Louis.  I've had ear problems for months, and in that time my preamp and whole audio computing system (including NAA) has been kaput.  But the ears are better, so I need to make decisions; right now I'm using an Oppo 105D as both dig preamp and DAC, not ideal.  I have a preamp now, will install pretty soon.  I expect I'll end up near where you are, running DSD512 with a DAC that can change base rates.  The Schiit Yggy beckons, but I love lots of DSD titles and by all accounts the DSD-PCM conversion has not been a beneficial one if it can be avoided.  Kinda wish I had jumped on the Exasound deal earlier, but I think DSD512 isn't available on the E32(?).  On the third hand, some of the top innovative labels report best results by recording at DXD and converting to DSDx2, 4, etc.    

 

Indeed.  I've learned to believe some pretty extraordinary things, based on the huge gains I got from using HQP to convert redbook to DSD128 using some -2s filters, so I'll accept your experience!  Probably will do the usual, make some tests like that, then try a few recommended settings that my new streamer/renderer could handle, and call it good.  Determined to get to DSD512 in any case. 

Sam,

 

Don't forget the IFI microiDSD DACs, including both the original and black label versions. These dacs are one of the least expensive dsd512 capable DACs out there.  Used versions are available for as little as $250. 

 

This is a great way to get into the dsd512 upsampling game without breaking the bank.  This DAC hits it's stride at dsd512, and with a well configured upsampling PC delivers jaw dropping sound quality with zero compromise.

 

Larry

[gdpr]

With all do respect, but I am not really appreciating the sound quality of the originaliteit IFI microiDSD dac, even (or especially) at DSD512.

My T+A 8 DSD blogs the iFI completely away, even at DSD128. (Still waiting for THE long promosed Amanero firmware that Will support DSD512 using Linux)

My IFI is for sale. If you are living in EU and are interested, leave me a message.

I will make you a nice deal!

Dirk

 

[jimdukey]

T+A is $4000, Ifi probably less than $400.

Not a fair comparison.

The T+A Should blow it away.

I'd be shocked of it didn't.

[lmitche]

13 minutes ago, jimdukey said:

T+A is $4000, Ifi probably less than $400.

Not a fair comparison.

The T+A Should blow it away.

I'd be shocked of it didn't.

Nevertheless the IFI is jaw dropping good when driven properly. I've had both here and I disagree completely.  They are closer then you think. We will have to agree to disagree.

[jimdukey]

I didn't post about the sound of either, never heard them.

Only commenting on Price difference.

What is" Driven Properly?"

You Had both, but what did you get instead?

[lmitche]

I use the IFI microIDSD BL daily. SQ is spine tingling.

 

Properly is a long list, just look at my recent posts.  It's in there.

[jimdukey]

I gather you're on a PC.

I just have an iMac.

I really have no idea what all the Initials and numbers are about with PC or Linux Systems.

Seems like another world of Computer Knowledge.

I'm happy w/DSD 128, on Mytek, Lampizator and especially the Mac D 150.

Actually, I like my Sony HAP Z1 better than any of my Dacs/Imac.

I'd get a Z2 if there ever was one...