HQ Player

[Iain]

RedBook CD doesn't contain any metadata. Internet metadata access such for commercial players, such as Gracenote, cost money. How much more you would be ready to pay for HQPlayer license to get metadata downloaded from the internet for RBCD?

 

You could use freedb.

 

Which is what Foobar2000 uses, mainly.

[Miska]

You could use freedb.

 

IIRC, last time I checked the terms it didn't seem feasible, but I can check again...

 

...still not feasible:

"All commercial and non-commercial licenses of the existing software components and the database are only available under the terms and conditions of the GPL."

[mansr]

...still not feasible:

"All commercial and non-commercial licenses of the existing software components and the database are only available under the terms and conditions of the GPL."

Do you need to use their software? Last I checked, it was a simple HTTP interface. Since you wouldn't be distributing the database, I don't see how the GPL would apply. Then again, I'm not a lawyer and could be missing something, and either way it would be rude to use the service against their wishes.

[Miska]

Do you need to use their software? Last I checked, it was a simple HTTP interface. Since you wouldn't be distributing the database, I don't see how the GPL would apply. Then again, I'm not a lawyer and could be missing something, and either way it would be rude to use the service against their wishes.

 

I don't think I'd need their software. But it the database is considered GPL and if that would make software using the database a derivative work. As I don't for sure know how GPL works in context of cloud-accessed database service, I consider better stay off...

[tsaett]

Miska,

in the context of upsampling everything to DSD512, what would be the appropriate sampling frequency for room correction filters? I'm using acourate to generate them but the limit is 192khz, I could ask the developer for higher rates though.

192khz seems quite lowish to me.

 

Thanks!

Christoph

[scintilla]

...what would be the appropriate sampling frequency for room correction filters? I'm using acourate to generate them but the limit is 192khz, I could ask the developer for higher rates though.

192khz seems quite lowish to me.

 

While I am not Jussi, I am using acourate filter with HQPlayer. There is a school of thought on filter bandwidth that says you should limit it to that which is necessary the bandwidth of the actual correction. If you are correcting to 24khz, then 48 khz filters and let the software scale the filter. Because HQP uses high-quality scaling for non-integer sample-rate conversions, you can get away with one rate. Personally, I do use 192k filters because I have a lot of those files and I stay in PCM through to the DAC, so it minimizes conversions. But for upsampling to high-rate DSD, I see no advantage in using anything more than nyquist. Jussi has stated that the actual rate of the filter is content-dependent and if you have 192k filter and feed it a 44.1 signal, the filter is scales to the incoming rate to save procesing power.

[tsaett]

Thanks scintilla,

I had in mind Jussi mentioned to use the highest rate possible but your explanation sounds just right.

That brought up another question:

I already use 192khz filters and I am upsampling everything to dsd128 which works a treat.

When playing native dsd128 recordings, I get constant dropouts. This does only happen with enabled matrix processing.

I checked for CPU issues but the load is at roughly 50%. This led me to the thought, the 192khz might be the bottleneck but your explanation is kind of a contradiction to this. Any thoughts on this specific topic?

 

Thanks!

Christoph

[scintilla]

I think Jussi will have to answer that, since I don't have a DSD-capable DAC and do not use the upsampler.

[Solstice380]

Thanks scintilla,

I had in mind Jussi mentioned to use the highest rate possible but your explanation sounds just right.

That brought up another question:

I already use 192khz filters and I am upsampling everything to dsd128 which works a treat.

When playing native dsd128 recordings, I get constant dropouts. This does only happen with enabled matrix processing.

I checked for CPU issues but the load is at roughly 50%. This led me to the thought, the 192khz might be the bottleneck but your explanation is kind of a contradiction to this. Any thoughts on this specific topic?

 

Thanks!

Christoph

 

Are you using a Mac or a Win computer?

 

On my Mac Mini with OS X I would get a random dropout regardless of filter or upsampled rate. Some times it would be in the first song, sometimes an hour into playing and sometimes never. I talked to a couple other guys who experienced it also. I didn't have anything else plugged in, but I think it turned out to be a USB buffer issue. I bootcampedmthe Mini with Win 10 and was able to increase the buffer to 50 msec and all stuttering issues went away. No buffer adjustment for OS X, though.

[tsaett]

I'm on WinServer2012. I get dropouts only with DSD128 of which I have 2 or 3 albums, so basically I couldn't care less.

Every other format and combination works perfectly.

It seems to me that 128->convolution->128 is particularly cpu-intensive. I simulated Redbook to DSD512 including convolution which gave me similar cpu loads than 128->128. (50%, below that with -2s family)

In the future I plan to upgrade my DAC to 512 capability, my concern now is that I could get dropouts there too and cannot test it before paying for the new board... Of course only if this is actually a cpu issue which seems odd to me giving the numbers...

 

Christoph

[Solstice380]

I'm pretty sure it's a USB issue. My 2012 i7 Mac mini hums along beautifully at low CPU % for an hour and then oops the sound goes out for a second and then back. The CPU Load never changed and there was nothing in the event logs happening at that time. Kind of a mystery.

[Miska]

It seems to me that 128->convolution->128 is particularly cpu-intensive. I simulated Redbook to DSD512 including convolution which gave me similar cpu loads than 128->128. (50%, below that with -2s family)

 

One thing where CUDA offload helps a lot is running convolution engine. For stereo use case, GeForce GTX 980 can do massive amounts of processing for sane price tag.

 

During H2/16 - H1/17 the upcoming 1000-series (Pascal architecture) nVidia is supposed to be up to about 16x faster than the current generation on HQPlayer-type loads.

[Miska]

When playing native dsd128 recordings, I get constant dropouts. This does only happen with enabled matrix processing.

I checked for CPU issues but the load is at roughly 50%.

 

This is a heavy use case. I assume you have Pipeline SDM enabled? The total CPU load can be misleading, especially for CPUs that have HyperThreading which virtually doubles number of CPU cores but doesn't really add computing resources. For example quad-core CPU with HyperThreding looks like 8-core CPU. But when four cores are fully loaded with application like HQPlayer, the CPU is practically 100% loaded, although OS considers still the remaining 4 cores being on low utilization and thus total load being 50%. For such system it is hard to reach 100% load with any practical application. HyperThreading still helps, but the load figures are not accurate...

 

It is best to check the per-core CPU load graphs to see if any core is peaking close to max.

[tsaett]

Hi Miska,

that is a very helpful information, I just checked and indeed while some cores are busy at around 50%, others run almost at full throttle. There is however a wide fluctuation within the cores. (CPU being Xeon e3 1265Lv3)

 

With DSD512 as my goal, I get stranger results though (I can only simulate, so no sound check)

Three cores are almost at 100% while the other 5 are idling at maybe 10-20%.

 

So what to do? Get a new CPU or is there a solution that spreads the load more evenly across the cores? (PipelineSDM is activated)

 

Christoph

[Miska]

that is a very helpful information, I just checked and indeed while some cores are busy at around 50%, others run almost at full throttle. There is however a wide fluctuation within the cores. (CPU being Xeon e3 1265Lv3)

 

Since that CPU is quad-core, it is unlikely to go much above 50% load figure (four physical cores out of eight virtual fully loaded).

 

With DSD512 as my goal, I get stranger results though (I can only simulate, so no sound check)

Three cores are almost at 100% while the other 5 are idling at maybe 10-20%.

 

So what to do? Get a new CPU or is there a solution that spreads the load more evenly across the cores? (PipelineSDM is activated)

 

With -2s filter variants I'd expect a bit less than 100% load, although it is hard to estimate.

 

With stereo content and Pipeline SDM you can get pretty much four cores fully loaded and two others partially, if you have 6-core CPU.

 

Bottleneck is most likely CPU clock speed in this case, base frequency being 2.5 GHz. For normal stereo playback case, the highest performance is pretty much max frequency quad-core combined with fast nVidia GPU for CUDA offload. So something like Core i7 6700K (4 GHz base frequency) combined with for example GeForce GTX 980 (Ti). Or alternatively Xeon E5-2643 v4 with 4-channel DDR4-2400 memory that has 2x the memory bandwidth compared to Core i7....

[Superdad]

I'm pretty sure it's a USB issue. My 2012 i7 Mac mini hums along beautifully at low CPU % for an hour and then oops the sound goes out for a second and then back. The CPU Load never changed and there was nothing in the event logs happening at that time. Kind of a mystery.

 

That exact thing happens to me too! (Same machine as yours.)

[kaka]

Just wondering if HQP can be made to decide what DSD setup to use based on whether the source material is PCM or DSD?

 

- I'm after DSD being dealt with using closedform/ASDM5/11.2

- and PCM converted to DSD using minringFIR/ASDM7/11.2

 

PCM is not heading through closedform cleanly (buffer time 100) but minringFIR sounds good. DSD is sounding better via closedform

I have set my DSD defaults to what I want for DSD, but I need at adjust them within HQP if I am playing PCM (from Roon)

 

Thanks

[tsaett]

Jussi,

thanks for your thoughts on this.

 

That's too bad, I just finished this built half a year ago and now it already seems obsolete for my new needs.

I hope I can consult you one more time on this topic. I added a screenshot of my cpu loads in the link below, left is 44.1 to 512 with matrix, right without (both with -2s). Bild: 1265lv3pfu37.png - abload.de

 

In my humble understanding the load still seems very uneven. Is this a normal behavior or is there a tool which fixes that at least in parts? (again sorry for my lack of computer-knowledge)

 

If this is a hopeless case, I'm going to try the CUDA offload. However, because I'm using a CAPS with a server motherboard, graphic card choice is limited (only x8). What would be your recommendation with this limitation?

 

I know, lots of questions but I would really like to get this 512 thing going.

Besides, I'm looking forward meeting you at the Sound Galleries meeting in Munich, I'll be there too!

 

Christoph

[Solstice380]

That exact thing happens to me too! (Same machine as yours.)

 

You'll just have to kick OS X to the curb! [emoji15] Not having ASIO is a major shortcoming. I Bootcamped my Mini with Win10. I got some stuttering initially but bumping up the buffer to just 50 msec alleviated that and I have never heard a dropout since. Longest session was 5 hours the other day, so I think it's good. Oh, and I got a solid bump in SQ with Win 10 and only the obvious processes and stuff shut down... No AO, ProcessLasso or Fidelizer or anything, yet.

 

Now I'm going to turn it into a NAA. I'll be trying both the router route (hehehe it was funny typing that, for those of us in the US who know the company Roto-Rooter!) and you're direct Ethernet connection. I've heard good things about the EMO Systems 70 network isolator so may get one to test. They cost more than a REGEN! Can John whip one of those up in his spare time? LOL

[Miska]

If this is a hopeless case, I'm going to try the CUDA offload. However, because I'm using a CAPS with a server motherboard, graphic card choice is limited (only x8). What would be your recommendation with this limitation?

 

What motherboard is it? Is the slot 16x size, with just 8x speed? If it is, then possibly something like the GTX 980 would work on it.

 

There are some parts of the processing that require just plain raw clock speed, so it is hard to say what exact hardware can do what without trying.

[tsaett]

It's the Supermicro X10 SLH-F, I think it is slot size 16x with 8x max speed.

My picture now gets clearer, thanks for that.

 

I did some research today and came to the conclusion that possibly the easiest solution would be to change the CPU, as sad as it is for me. Within my limitations (MB, power supply, cost, heat) the Xeons E3-1276v3 and E3-1271v3 seems like the best solution. 3,6GHz, Quad Core, 80W/84W TDP.

 

Jussi, knowing that in your last answer you said that a prediction is hard without trying, my last question for you would be your opinion on this CPU. I read about some guys getting great results from the 6700K doing 512 without CUDA, which albeit running at 4Ghz is just a tad above the Xeons in benchmark tests.

I know it's not your job to be every man's hardware consultant, but your opinion would be much appreciated!

 

Thanks

Christoph

 

 

Edit: For everybody that can use xeon e5's, there are incredible deals on the e5 2670 8core on ebay, selling them for about 5%-10% of the retail price (way above 1000$)

Use two and you have 16 cores available :-)

[sydlow]

Can Auralic Vega do DSD128 with WASAPI?

 

My Vega can do PCM to DSD128 conversion fine with the Auralic ASIO driver, but I wanted to test it with WASAPI.

However, it doesn't seem to want to do so with the WASAPI interface, showing 3.0M in the UI interface.

 

Is this a limitation of the WASAPI/Vega?

[Miska]

Can Auralic Vega do DSD128 with WASAPI?

 

My Vega can do PCM to DSD128 conversion fine with the Auralic ASIO driver, but I wanted to test it with WASAPI.

However, it doesn't seem to want to do so with the WASAPI interface, showing 3.0M in the UI interface.

 

Is this a limitation of the WASAPI/Vega?

 

If the WASAPI driver exposes PCM only up to 384k sampling rate, then it is WASAPI/Vega limitation due to DoP. With ASIO you can get raw DSD with DoP and thus different and unrelated set of sampling rates compared to PCM.

[Theobetley]

ok I want to update to latest level how do I do? I went to Signalyst site and tried downloading the latest version which was easy but it stopped during installation and I had to abort. What did I do wrong?

 

Closed Hqplayer down and tried again. Wanted to abort because I could not install the pdf file so I said ignore. Did I do it right?

[ericuco]

ok I want to update to latest level how do I do? I went to Signalyst site and tried downloading the latest version which was easy but it stopped during installation and I had to abort. What did I do wrong?

 

Closed Hqplayer down and tried again. Wanted to abort because I could not install the pdf file so I said ignore. Did I do it right?

 

It is usually helpful to include which OS (Win, Mac, Linux) you are using. The installations/upgrade procedures are different for each.

 

If you have HQP open and are trying to upgrade at the same time, I would close out HQP before starting to upgrade.