HQ Player

[Yviena]

1 hour ago, Miska said:

 

How much do you think would be nice? If five minute song takes ten minutes to process, waiting 10 minutes for music playback to start is not that nice.

 

Yeah I didn't think about that... I was thinking more like 10-30 seconds, with HQplayer working on the next 10-30 secs while the previous  upsampled content is playing, something similar to what MP4 streaming chunks does if that is even possible, probably not...

[rando]

Still curious how the new EC modulators are used to make downloadable 512 files to be used with non-EC modulators.  Because nobody can play 512 using them.  

 

The bigger question here is what the developer is up to.  I'm fully behind outpacing the slow march of consumer tech if that is what occurred.  Catching his breath more than 4 hours a night sounds humane enough.  No way of knowing so I have to wonder.

 

Are forthcoming modifications going to be refinements or a capitulation allowing EC  512 without needing to keep up with "churning" hardware cycles?  

[Miska]

2 hours ago, rando said:

Still curious how the new EC modulators are used to make downloadable 512 files to be used with non-EC modulators.  Because nobody can play 512 using them.  

 

With Pro you can do offline conversion from a file to a file, it can take what ever amount of time needed to do the conversion. You can then play it with Direct SDM enabled, if you don't need any other processing.

 

2 hours ago, rando said:

Are forthcoming modifications going to be refinements or a capitulation allowing EC  512 without needing to keep up with "churning" hardware cycles?

 

I'm not having too much expectations about 512x rates, even less 1024x rates. I'm happy if I can make 256x more reliable and accessible with wider variety of hardware.

 

I will continue trying to figure out how to better distribute the load and optimize the algorithms so that CPUs are operating as efficiently as possible.

 

To some extent, going further either requires CPUs to become more efficient in terms of total clock cycles consumed to perform certain set of operations. Or get the clock frequencies up higher.

 

In a way this reminds me of old days with Microsoft Flight Simulator, where it would always take couple of years after release of a new version before hardware became available that could run it smoothly with full detail settings.

 

[Miska]

3 hours ago, Yviena said:

Yeah I didn't think about that... I was thinking more like 10-30 seconds, with HQplayer working on the next 10-30 secs while the previous  upsampled content is playing, something similar to what MP4 streaming chunks does if that is even possible, probably not...

 

If the processing is running slower than playback speed, playback will eventually catch up with processing and then you will start having the dropouts. Only solution is to have the processing running faster than playback speed.

 

And if I pre-process first 30 seconds and that takes 60 seconds to do, there will be lot of complaints about having to wait for a long time when starting playback.

 

[Yviena]

39 minutes ago, Miska said:

 

If the processing is running slower than playback speed, playback will eventually catch up with processing and then you will start having the dropouts. Only solution is to have the processing running faster than playback speed.

 

And if I pre-process first 30 seconds and that takes 60 seconds to do, there will be lot of complaints about having to wait for a long time when starting playback.

 

Hmm yeah your right, it will probably be better/easier for you to just focus on trying to Parallelize the EC modulators instead.

[rando]

31 minutes ago, Miska said:

I'm not having too much expectations about 512x rates, even less 1024x rates. I'm happy if I can make 256x more reliable and accessible with wider variety of hardware.

 

Thank you, I realize how brusquely the question was framed.  

[k6davis]

1 hour ago, Miska said:

I'm not having too much expectations about 512x rates, even less 1024x rates. I'm happy if I can make 256x more reliable and accessible with wider variety of hardware.

 

I will continue trying to figure out how to better distribute the load and optimize the algorithms so that CPUs are operating as efficiently as possible.

 

The silver lining is that, even now, the hardware to get to EC DSD256 isn't very expensive or complex, and that investment is not likely to be made obsolete for a while.

 

EC really does sound great Miska. I've been focused on testing and configuring hardware and only recently have sat back and listened for enjoyment. The improvement EC brings is not subtle. It's much more apparent to me than when I made the jump from (non EC) DSD256 to DSD512.

 

And for anyone who's server is "only" able to upsample to EC DSD128, the essential character of EC can still be heard at that rate.

[elan120]

5 minutes ago, k6davis said:

And for anyone who's server is "only" able to upsample to EC DSD128, the essential character of EC can still be heard at that rate.

This is good to know.  I also have AMD CPU and have been upsampling to non-EC DSD512 with non-2s filters for a long time, but switching to EC DSD128 has a big jump in SQ improvement.  Hopefully it can do EC256 with more optimized algorithms...soon.

[k6davis]

25 minutes ago, elan120 said:

This is good to know.  I also have AMD CPU and have been upsampling to non-EC DSD512 with non-2s filters for a long time, but switching to EC DSD128 has a big jump in SQ improvement.  Hopefully it can do EC256 with more optimized algorithms...soon.

 

Definitely. I have both an AMD and an Intel server on hand right now. I just boxed the Intel up because I'm ultimately going with a different Intel system. So I've gone from listening to EC DSD256 to EC DSD128 for now and it still sounds fantastic.

[LoryWiv]

3 hours ago, Miska said:

 

I'm not having too much expectations about 512x rates, even less 1024x rates. I'm happy if I can make 256x more reliable and accessible with wider variety of hardware.

 

Many replies here indicate that we can do EC modulators at 128x with most hardward but unable to do 256x. That is my result as well. I would welcome optimizations that allow the new modulators to work for 256x without stutter or such high cpu demand as a next step. Iterative progress is real progress. Thanks Jussi!

[Yviena]

@Miska oops i mean't to say that there is no sound when sending PCM upsampled to DSD when directSDM is selected when using 4x version not DSD to PCM, seems that my phone was autocorrecting the words.

[DancingSea]

On 7/25/2019 at 11:35 PM, Le Concombre Masqué said:

"soup nazi" ? is it a quote from surf nazis must die ?

 

Its a Seinfeld reference.  Its a joke...

[JTS]

So my home system uses an old HPZ420 video workstation-in-a-closet which has a Xeon E5-1620 in it (3.6-3.9 ghz) with an old GTX980 in it. It runs DSD512 fine with most filters to a W10 mac mini NAA into my Holo Spring.

 

I just sparked up the EC modulators and I get a good solid DSD128 going with ASDM7EC. It sound fabulous. While it is early days, I can say that the difference is marked to the previous setup in DSD512. I'll do more testing tomorrow.

 

I am curious, as it has been mentioned here, if DSD128 is preferable to the previous modulators @ DSD512. Anyone else making this comparison?

[asdf1000]

6 hours ago, k6davis said:

that investment is not likely to be made obsolete for a while.

 

That's what  people that upgraded PC's to finally get xtr filter working at DSD512 without -2s thought.

 

And then EC modulators came 😁

 

I don't mind though, so this is not a complaint. Jussi isn't forcing people to use EC modulators or use ext2.

 

ASDM7/5 didn't suddenly become crap and obsolete. They remain high quality modulators.

 

[Le Concombre Masqué]

1 hour ago, Em2016 said:

 

That's what  people that upgraded PC's to finally get xtr filter working at DSD512 without -2s thought.

 

And then EC modulators came 😁

 

I don't mind though, so this is not a complaint. Jussi isn't forcing people to use EC modulators or use ext2.

 

ASDM7/5 didn't suddenly become crap and obsolete. They remain high quality modulators.

 

True but there's no coming back. In my case it goes as far as reverting to Desktop from Embedded for latest E iteration stutters with my hardware.

Very good descriptions have been made above ; k6davis nails the main point very well with his "clarity on transients with absolutely zero harshness"

I could add that recordings sounds newer fresher "livier". You could ie focus on the speed and delineation of a kick drum : mean lean deep as it should when, in comparison it would have been slow fat boomy

[asdf1000]

Hi @Miska

 

Have you seen this (in particular highlighted section at the bottom).

 

DSD Direct supported, like RME ADI-2 and your Teac.

 

Only RCA analogue outputs  though....

 

[Miska]

11 hours ago, Em2016 said:

Have you seen this (in particular highlighted section at the bottom).

 

DSD Direct supported, like RME ADI-2 and your Teac.

 

Only RCA analogue outputs  though....

 

Looks like yet another alternative. Nice to see to have various alternatives.

 

If I can find one at nice price, I could get one for testing.

 

[fred_com]

To all my fellow Threadripper users - I seem to have found a way to enjoy EC filters* into DSD256 with little effort.

*up to ASDM5EC - ASDM7EC is still out of our league, at least for my machine.

 

So all it takes is to disable two out of four of the dies of our CPU, and set the memory mode to "local".

I've done this in BIOS, although this should be the same as setting a "Game Mode" in Ryzen Master software.

 

There is an "Advanced/AMD CBS/Zen Common Options" menu in my Taichi X399M BIOS - I guess it should be in all other X399 chipset motherboards. Under this menu there is "Core/Thread Enablement" item.

Here you can select number of cores to disable via the "Downcore control" item. The options of this item are a little confusing, but "4+0" should mean that only 8 cores out of 16 should be used.

 

Memory mode is found under "Advanced/AMD CBS/DF Common Options", item is "Memory interleaving" - it should be set to "Channel".

 

All other CPU options are left to default, meaning that Boost and all other stuff should be working as usual. 

 

There's just one caveat - Hibernate ceases to work when you disable the cores. There is no way to enable "Suspend to RAM" in the corresponding ACPI settings in BIOS. (You can enable it, but it will fall back to disabled after reboot)

 

Of course, cores can be enabled back by setting the corresponding item to "Auto", so there is no harm.

 

And YMMV of course  

[elan120]

1 hour ago, fred_com said:

To all my fellow Threadripper users - I seem to have found a way to enjoy EC filters* into DSD256 with little effort.

*up to ASDM5EC - ASDM7EC is still out of our league, at least for my machine.

 

So all it takes is to disable two out of four of the dies of our CPU, and set the memory mode to "local".

I've done this in BIOS, although this should be the same as setting a "Game Mode" in Ryzen Master software.

 

There is an "Advanced/AMD CBS/Zen Common Options" menu in my Taichi X399M BIOS - I guess it should be in all other X399 chipset motherboards. Under this menu there is "Core/Thread Enablement" item.

Here you can select number of cores to disable via the "Downcore control" item. The options of this item are a little confusing, but "4+0" should mean that only 8 cores out of 16 should be used.

 

Memory mode is found under "Advanced/AMD CBS/DF Common Options", item is "Memory interleaving" - it should be set to "Channel".

 

All other CPU options are left to default, meaning that Boost and all other stuff should be working as usual. 

 

There's just one caveat - Hibernate ceases to work when you disable the cores. There is no way to enable "Suspend to RAM" in the corresponding ACPI settings in BIOS. (You can enable it, but it will fall back to disabled after reboot)

 

Of course, cores can be enabled back by setting the corresponding item to "Auto", so there is no harm.

 

And YMMV of course  

Thank you, Fred,

 

I will give this a try since I also have Taichi X399 Mobo and hopefully it will go from DSD128 to DSD256  😊

[rando]

That's using your noodle @fred_com  

 

PM'd CC about this broad topic as a means of educating while sifting out the "gamey" ideas, in an article.  fred_com reinforced the overall structure I suggested almost exactly.  A HQP specific thread may be required in the meantime if AMD manipulations take root.  

[fred_com]

Forgot to mention - those results are with CUDA offload to 1050 Ti - so all the filters work, except for poly-sync-xtr.

 

@rando - CPU-specific optimizations thread is a must, for sure. But just want to add that, by turning off half of the CPU, we essentially turn our super-fancy gorgeous CPU into mere Ryzen 2700, or similar  

So I wouldn't call it a true AMD CPU optimization, just a specific case for Threadripper.

[StreamFidelity]

On 8/2/2019 at 4:07 AM, JTS said:

I am curious, as it has been mentioned here, if DSD128 is preferable to the previous modulators @ DSD512. Anyone else making this comparison?

For me the EC modulator ASDM7EC is only really good with DSD 256. With DSD 128 I found it too lean.

[rando]

8 hours ago, fred_com said:

 

@rando - CPU-specific optimizations thread is a must, for sure.

 

I'm sure Miska would appreciate a separate discussion as well as pool of results such as you've provided.

[bmoura]

On 7/31/2019 at 9:04 AM, ted_b said:

if you are a classical fan, realize that our NativeDSD site now offers 512 examples done with Jussi's EC modulators, so if you are well versed in what the lower DSD rates sound like in those recordings you can give them a try.

 

 

And not just classical fans.  The NativeDSD site also has some Jazz in DSD 512 with Jussi's modulators from Sound Liaison, Groove Note and Yarlung.  Very nice!   (15% Off thru 8/5/19 with code SUMMER19). 

 

https://www.nativedsd.com/homepage/dsd512_music

[Miska]

21 hours ago, fred_com said:

To all my fellow Threadripper users - I seem to have found a way to enjoy EC filters* into DSD256 with little effort.

*up to ASDM5EC - ASDM7EC is still out of our league, at least for my machine.

 

Have you checked if this essentially leads to higher frequency boosts on the remaining cores?