The optimal PC-setup

[Flecko]

Hi all,
as I mentioned in an other thread, I like to build a dedicated PC-system for audioplayback. So now I have to decide how this systems has to be build.
One theory I read about is, to tune the system to a very low power consumption (the CMD2 guys). They advice some parts they think are very good for a pc system. If theire theory is right, a very simple Intel Atom based PC should be ideal (but they do not advise such a system..). The whole system of Atom Netbook has a powerconsumtion of 10W! Thats incredibly low. On the other hand there is XXHE, using a lot of ram and need some computing power to run. So as I like XXHE more than the CMD stuff I should of course build a system that is capable to run it. What is your opinion? How should the system be build? I probably do not need the computing power for upsampling because I eventually will buy the ART Legato, which is designed to 44.1khz only. And BTW: Peter, what are you cooking in your lab? Should I wait for it?
Thanks and Greetings
Adrian

[Flecko]

PS:
There are two options to get from the PC to my DAC.
1: Art Legato, a high end usb/spdif converter costing 500€
2: Hiface, you all know it...costing 140€

The plus of the Legato is, it is build with its own PSU and with a super high quality clock, which is designed to 44,1khz only. This is better if you just play 44,1khz files but XXHE supports upsampling technics, that as far as I remember sounded alway better than the orignal. At the moment I can not use it because my usb/spdif converter just can 16/48 maximal. So why spending so much mony if probably the sound with the highface could be equal or better than the Legato. I read, Legato is better if you use it with 44,1khz but the distance is not so big they say. Any experiences or opinions?

[PeterSt]

Hey Adrian - my 2c on this :

I think it is known by now that I don't do much about the PC side of things. Maybe I should, but there is no way I will. It is just too inconvenient, like where to put all the disks otherwise, knowing that one (2TB) copies in 4 hours, which will be 2 days via USB, to name something. So, SATAII connected, and all to the mobo. They draw power, so that power (hence PSU) has to be there, as is the cabinet to store them in (I just counted, and there are 8 permanently connected).

So, where this puts constraints on the mobo (must have all the SATA connections), the only remainder is a good processor. It must have large 2nd level cache, two real cores, and be able to swap between cores fast (copying of the cache). A fast FSB is what is left, I think.

I have the cabinet just open because the HDDs stay cool anyway, and this way the processor keeps cool enough not to let spinup the fan. I know, this is not the recommended way (air flow), but it still works the best.
This disks make no real noise, and the PSU and cabinet fan don't either, as long as you keep everything clean from dust (the more dust, the more hot the mobo and everything gets).

Do not put in any WiFi stuff and don't use an IR keyboard/mouse. An IR dongle for remote control is perfectly ok.

Of course this is not 100% noise free, but I have the lot in a cabinet with closed door at the front (open at the back).

As said, just 2c, and my personal likings. But it may give you some ideas.
Peter

[Telstar]

Well, a NAS would be a perfect solution, especially since it can be located in another room. This is what i will do when i'll need more than 2TB of space, which for my speed is not very soon

Do you have any arguments against NAS about SQ, Peter?

Also, I have to figure out the thing about wireless keyboard/mouse (i.e trying without)

[PeterSt]

Nothing against a NAS for SQ. But, it will depend largely on the interface and the virtual environment which is (or may be) created. If that piece of driver stuff is constantly looking at changes on the NAS, it *will* influence SQ.

No NAS for me. There's just no value in them, and consume more disks generally (because of the desired RAID setup). They have no value for backups either, unless you think a disk may fail, and the other is still there (legit by itself). But keep in mind, in my case it would take 16 disks at least, so again it doesn't work. And it will always be 3-4 times more slow.

[GerardA]

About swapping between the cores fast:
The memory can be set in ganged mode or unganged, will this make a difference for XXHE?

[PeterSt]

You really shouldn't be asking question to me nobody has the answer to anyway.

I would say Unganged is the best but YMMV.
It is not something an application (like XX) can explicitly benefit from, but I can imagine specific tasks can (by accident). So, Ganged would be faster if coincidentally things have to be done which can be dealt with with one command. But if that is not the case, with Ganged you'll loose half of the bandwidth (must wait for a next command and each command can deal with half of the bandwidth). With Unganged each command is efficient, but no two commands can be performed at the same time. So the two commands (which possibly could be combined) now have to wait for eachother.

The above is my take from it all, and no way I can map it to the working of XX.

This is one of the most understandable versions of what is expressed about this all : http://gathering.tweakers.net/forum/list_message/30362099#30362099 (the english part of this text, "The Barcelona ...").

[PeterSt]

And BTW: Peter, what are you cooking in your lab? Should I wait for it?

Adrian - I only now realize that I never responded to this. So, not meant as a commercial, but merely to possibly prevent you from spending money on something which turns out not to be "the best" ... what I am cooking should be the best (ever).

I am not sure why you ask the question, since it should be commonly known about the Phasure NOS1 DAC, but let's say that I may have the idea that you just jumped a hurdle you couldn't or didn't the past 3 (and more) years. And so *now* you are interested ...

Allright. It may be interesting for others too, to put all in a somewhat other perspective, also looking at the title of this topic. Here goes :

By now, audio forums are full with PC playback. And, although I don't recognize much in "the best sound ever" from all those forums, in here we explicitly do, and on that matter it is not much different from the old days at Bert's forum. And as you may remember, he may have been the first to explicitly be (open) about PC playback, and this at a level of which we both know it is about "the best" only. This was back then, and this still is the case. On Bert's forum we indeed don't see that much anymore, but I can tell you that both him and me still see eachother once per 4-6 weeks or so, and when that happens it is about long nights of listening, comparing, and draw out of equipment what's in there. The most important here is, that both Bert and me never sit back when something can audibly be improved, and whether this is in the speaker range, the software or as how it is now, the DAC ... it will be done.

What on other forums (apparently) doesn't work (at all) is the making clear of the progression which is made in only a few years time. I recall - and this really is less than two years ago - that for the first time a hi-hat started to be expressing through my speakers. Keep in mind, this was after working on XXHighEnd for two years already, with which I only want to say : this is not about software only. It is about the whole chain used. Today (thus less than two years later) I always must smile with the hi-hats all over, and just everything being there as, say, should. Similarly I recall the day that we heard a noise through the (horn) speaker, and with our ear in the horn we identified it as a brush. Today ? it seems totally impossible that this brush sounded so faint, because now it almost blasts through the room.

It goes by bits and pieces, but all together it is an unbelieveable blast compared to a few years back. Many of us here will remember the unbelieveably suddenly coming forward of background noises (a brush belongs to that in my ideas) only because of Arc Prediction Upsampling. Today we are used to it and we proceeded with Kernel Streaming, the Adaptive Mode version of it, and later (this is only about steps taking a few weeks) the Ultra Low Latency Special Mode blahblah. But again it went better because of it. Yes, it seems unbelieveable that it is not a half year ago that we had Engine#3 ... which nobody listens to anymore, I'm quite sure.

The real message is, that because of having it all so much better today, it has become enormously more easy to proceed. So, it goes faster and faster, although that will end at some stage, I'm sure.


With the above as a background, I started working on a DAC, and it should have the base principles I had in mind for a longer time, which would be non oversampling, and no filter. The latter because I had in mind to improve the filtering plus I thought it could be done in software (thus not in-DAC as usual). Now, while at some stage I managed to create that filter (Arc Prediction), it had to be consistent with the design of the DAC *and* its environment;

One of the key design principles is that the harmonic distortion would be shifted out of any danger zone by applying 8 x upsampling. Notice that "upsampling" is a filter means by itself, and no digital filter can work without it. But, since the filter is in the player software (XXHighEnd), the DAC has to accept 24/384. This by itself is a nice task to begin with, knowing that no computer interface allows for that. So, this is an example of what I mean by consistency with the environment. No downsides were and are to be accepted, and when I say that 384 is needed, well, this is what will be created.

I won't eleborate on the many other aspects, but all of them are approached the same : when (my) theory says what is the best, it *will* be created like that, no matter there's no one around to tell me how. And this is how the DAC ends up with *each* of the elements in there being not applied before. So, not only one, but all of it.
You can call this crazy, but if you want to get the best out of it, compromises are not allowed. And thus again, when the environment has to change in order to get there, it will be done (and it has been done). The total effect is really drastic, because this time we have an NOS DAC that measures better than -98dB THD+N for all the frequencies covered for, and mind you, this is 10Hz - 192KHz, would 192KHz be audible. Now look at the other NOS DACs, and try to find one which measures better than -70dB THD+N ... while what you will find will be about 1KHz as a test tone only. And oh, I measure at the end of the interlink.

While you could say that many DACs have better figures, I will say this is not so to begin with because the measurement will not be about 10Hz - 192KHz, and a 1KHz test tone is dead easy ! - but far more important is that this will be about oversampling DACs. And now the (very wrong) filtering of those comes into play, and it is exactly this why I started this DAC project as I did. Thus, an NOS DAC with the best figures ever, and I succeeded in that.

While before NOS DACs might have been a choice for many because they sound better than OS DACs, those people (and that included me a few years ago) must realize that they listened to huge amounts of harmonic distortion. On the other hand, what actually nobody thinks about in a decent way, is that the OS DAC measures good, but only because of the ringing the filter incurs for can't be measured for THD. The ringing itself can be measured, but not the distortion coming from that. This is exactly why an NOS / Filterless DAC sounds so much better to many, with the further perception "yeah, it measures the worst, but apparently our brain can filter that ?".

The latter appears to be non-sense, because the two (OS / NOS) just can't be compared, and nobody could ever listen to an NOS DAC with a for that type of DAC justified filter. So, engage Arc Prediction, and there is the sound never heard before ... It filters as should, with zero ringing.

So, should you wait for it ? I think so, yes.
Peter

[manisandher]

Peter, I find the prospect of the NOS1 really exciting.

As I've mentioned before, I'm looking forward to comparing its SQ with that of my PM Model Two - easily the best DAC I've ever heard in my system... I mean, it's in another league to pretty much anything else out there... certainly all the delta-sigma designs that I've heard, which just sound 'broken' to me...

BUT... the PM Model Two was never designed as a DAC. Its real strength is in its recording capability, with its full ladder ADC chips (IMHO, still better than anything available today, with the top-end Lavry stuff coming second), and also its rate conversion capability (which admittedly has probably been superceded now with software-based converters) - the DAC is there just for monitoring purposes. And this is what is really exciting about the NOS1 for me. To finally be able to hear how good recordings made on the PM Model Two really are. Intuitively, I feel PM Model Two recordings played back on the NOS1 will be something very special.

But this leads me to one question for you. When using the Model Two, should I record at 16/44.1 or 24/176.4? The former, I know, will work with Arc Prediction. But in a future release of XX, will I be able to upsample 24/176.4 by 2x or 4x and feed into the NOS1? I'd rather reocrd at the higher rates - the Model Two works internally at either 176.4KHz or 192KHz, and when recording at these rates, no decimation is necessary, which I'm assuming is a good thing.

Mani.

[PeterSt]

Yes, before the NOS1 ships, doubling (and quadrupulling) those rates will be there. Notice that the biggest hurdle for this so far, has been the upsampling with 24 bit files as a base (meaning, the structures in the software were based upon 16 bit files to upsample). But, since I thought that HDCD has to be upsampled just the same - while as you know the decoded output of HDCD is 24/44.1 - the upsampling from 24 bit files had to be in there.

... Ok, since HDCD decoding is supported in 0.9z that is.

Thus, the hard part has been done already, and now it is merely about finding some time to do it.

Thanks,
Peter

[manisandher]

Thanks Peter, that's really helpful.

Mani.

[manisandher]

Peter, as a reference for your NOS1 DAC, here are some test results of my particular PM Model Two (provided to me by Dave Peck, the only person qualified to service the Model Two):

1) DAC THD test at -10dBFS
2) ADC/DAC test – analogue signal, converted it to digital, looped from the digital out to the digital in, and converted back to analogue – the entire A/D-D/A chain end to end.

Would be interested in your thoughts...

Mani.

[PeterSt]

Hehe, nice ...

About the first picture, THD+N would be at -115dBFS, but notice I can't see the real output voltage. And only if you know it can do 2VRMS we can assume the -10dBFS to really be there. Also notice that what the analyser says in the left pane "-10dBFS" is just the analyser signal generator setting. So, in reality the output can be less, and THD+N is (or must be) related to the real output signal.

Then, as him to shut off the band limiting. He may say "impossible because then I can't produce an FFT". In that case, bad luck.
I think the "22KHz" in the left pane indicates it is on.

Lastly, the figures in general don't look as filterless to me. But this is exactly why the band limiting in the analyser should be shut off; what is
"lacking" in the DAC is now applied by the analyser !
But maybe the DAC is not filterless in the first place and then it all doesn't say much. Why ? again, because the filter will be ringing, and this can't be measured into THD+N ...

I'll stand corrected including apologizes when I'm completely off here !
Peter


PS: For comparison, look at the last picture in this post : World's first NOS 24/192 filterless DAC - Latest Graphs;
Here the output is at -40dBFS, THD+N (because of that) showing at -135dB or so. The difference between -40 and -135 is (-)95.
At that stage in time Arc Prediction wasn't there yet, and instead AI was used. Beware : to playback a file. So, letting the signal generator provide the test tone will never work, because indeed *some* filtering has to be applied in order to make THD good. If it's in the DAC, ok, but then it's in the DAC and rings. If it's done before the DAC, again ok, but now shut off the filtering in the analyser. The latter (filtering before the DAC) is not likely to be anticipated upon for the M2, so or it shows the worst figures when no filter is in there (and use 5KHz instead of 1KHz !!) or it shows nice figures in that case because of the analyser being applied wrongly.
And keep in mind my due apologizes.
 

[manisandher]

Hey Peter, thanks for your thoughts - I know you're busy working on 0.9z.

I've sent an email to Dave Peck at Euphonix (now part of Avid), so let's see what he comes back with. I would be amazed if the Model Two is oversampling at 176.4/192KHz rates. Firstly, what filter would it be using? It only has a PMD-100 HDCD filter, which just isn't capable of these rates. Also, there is such a big improvement in SQ between using its on-board 4x upsampling and using QAP in XX - I can't see how this could be possible with a filter in play.

I think the first graph may well be at 1fs, in which case the PMD-100 would certainly be engaged. But the AD/DA graph looks like its at 192KHz, if I'm reading the left-hand pane correctly. But you're right, this is probably still band-limited, so of limited use.

In any event, hopefully Dave Peck will reply and we'll know for sure. I know no-one else is likely interested in this, but I certainly am. If nothing else, I want to know why recordings made on the Model Two sound so different to others... they have an immediately identifiable SQ.

Mani.

[PeterSt]

Thanks Mani. Notice that I didn't talk about the AD-DA (second picture). Also I must admit that I didn't think about the 1fs when writing, but I saw it too. Of course everything will be different at playing hires files (so, I did NOT think about that during my writing), but still the filter in the analyser should not be engaged.

So, that was apology nr.1. Haha.