Split file size and volume

[PeterSt]

I think Boleary AKA SFS moniker is right ... sort of. It is volume related ... kind of.

I tried to mimic something of what you all claim here, and at some stage I had something working. I continued the Dread Zeppelin album where I quit the day before, and immedietaley noticed

a. an emphazation of distortion (guitars hehe);
b. some kind of crazy dynamics for the more rough parts. Like hitting a snare (drum) without the snare activated.

Next I ran that Album from Prince of which I never know the title (I call it "P" because it's in cold (a P) on the sleeve), and of which the first track is called "My name is Prince". It is no coincidence that I wanted to try this album, because it has been one of those 5 CDs I never could play without distortion for 10 years long. This was history already for a year or so, but still.
This first track never sounded so mean to my ears as ever before, but for the first time I could hear why that is;

First there's a bass synth which forms out of very fast off/on and which is very difficult to notice because if the (bass part of) a system isn't fast enough, it isn't off/on at all, but just a mere continues tone (things can't follow). Btw, the 2nd track on Musicology contains such a bass which may excurse your woofer to tearing dimensions when your system can't follow (and which may be a matter of the proper Q1 setting !).

Now, with this "My name is Prince", this time, I also noticed that they applied the same on/off thing to his voice. Actually the effect was very similar to the distortion guitars and I appreciated it as distortion. Very strange, because only his voice exhibited it. Also, all of the next tracks didn't have it, and didn't have any of the emphazation effect at all.

To me it looks like that a very "square" wave now can be followed better. For Dread Zeppelin I didn't like it, and the whole lot received a flare hence character. Worse, I heard that character back in the first track of P. But then listening carefully, I coulld only appreciate it, because it just shows something which seems not to be there otherwise. But still. And it didn't bother me at all anymore after the first track.
3 tracks before the end I got an an out of memory, changed the program for that, ran a nect album with smaller tracks, but kept on having the out of memory. So I reinstalled the original program to be able to play along.

If this is emphasizing the more square stuff, the first to notice will be sibilance in women's voices. You may like this better because it will show better the unique character of the voice. I mean, if it's in there it just is, and then better hear it. But if you play this louder, things may go very wrong, because this is so square that things won't be able to follow, and you'll have distortion instead. So this *is* volume related ...

I'm not sure what to do with it;
So far I couldn't find any means to make it decently working in the program, and with my 3GB of memory there's hardly any headroom to play with. So, this needs more memory, but it just isn't there. I must add to this that all is right in the middle of my mangled "Dot-net" memory management, which is fragile to begin with, because not official(ly used). So if I tweak only a little, I'll get an out of memory. And like I just implied, with the small tracks ! Two years back I have been working on this for two months or so, and I better leave it alone ...

But it sure looks there is this other dimension. Now how to normally utilize it.

[manisandher]

I tried to mimic something of what you all claim here...

Peter, what SFS were you using during your listening?

What's of particular interest to me is whether changing the SFS causes a consistent change in SQ. For me, starting at 12MB and increasing the SFS makes the sound fuller, but perhaps too woolly if increased too much. Do other people experience the same?

What I am now convinced of though is that we CANNOT come up with a universal 'best SFS' number for everyone. When I had my CPU rate set to 9x, I preferred a SFS of ~60-70 (for 16/44.1 material). But set at 21x, I'm perfectly happy with a SFS of 12MB.

Mani.

[PeterSt]

My SFS has been 70MB for ages. CPU is at 9x btw.

During my testing yesterday we can't speak of "split File size" as such, because I mangled the program.

[Jack]

Mani
Sorry about that new 'can of worms' I just gave you!
However, did you up the voltage at each change, or just once, or not at all? Just to complete the picture for me.
Thanks
Jack

[Marcin_gps]

I think again, its all about speed,

Did anyone proved that underclocking is better vs more RFI ?!
The faster a pc can handle things, the more time there is to handle the music.

Speed, speed, speed.

Roy

and the more distortions goes to an audio interface 

[PeterSt]

... but it is more likely it goes the other way around, though the distortions will be of a differend kind then (think like skips).

As far as I'm concerned, it must be concidered a myth that undervolting etc. will really help in a "controllable" fashion. *If* something is interfering with sound at all, it should be removed and not attenuated. Thus by more physical means like better grounding.

I think I have seen how such things can change the jitter level impeeded by something we call noise, but in the end is always some (recognizeable) signal coming through which shouldn't. For example, I can let the "clock" run faster because of it. So, if the clock signal coming from the clock (steadily running by itself) gets polluted with another nice signal, once in the x cycles of the both signals another nice "top" of the wave comes up, which is seen as the proper rising (or falling) edge within the DAC. By this means 1000Hz can become 1030Hz easily !

This was an example of the most sound influencing kind, and it is totally unrelated to something like CPU speed "hence" RFI or anything. It is wrong design and out of your control.

The (for me) strange thing is, that while I never saw prooven that RFI does something to a signal we are talking about here, the "grounding wire" itself *does*. So, when there is this unwanted signal (could be the clock visible on the RCA output) and all is officially grounded already, connecting a next ground wire to some point (you expect to be good), moving it around changes the amplitude of that unwanted signal all over. Move your hand around that wire also shows big changes. Connecting it to one side of the same cabinet may remove the signal, while connecting it to the other will amplify it.

Maybe a more understandable example :
Connect your interlink to your preamp (or DAC) and at the other end connect the scope (all signals off). The cable has two wires, one for ground and one for signal. Both are connected to the plugs. Now connect another single wire to the outside of the plug (which is connected to the ground) and also connect it to the plug at the scope's side. The noise will be totally different from this, and while you had some "hum" before, now it may be totally gone.

So, no matter this is beyond our comprehension most of the time, it is these things which matter. All that other hocus pocus ? I can't see it on a scope ...

All together, if you find that undervolting etc. really makes a difference, better investigate whether you did all to avoid the influences. This is where I started with a network player to be the "soundcard" and some later the FireFace (USB should do the same, but it never cut it for me).

That software can still influence SQ is quite another matter, for the largest parts a secret for myself too. But it influences in the DAC, not in the PC.

Peter

[Marcin_gps]

True, but my sound card works as a DAC. For digital transport the influence is smaller and proper ground should do it.

[PeterSt]

For digital transport the influence is larger ...

[Marcin_gps]

Peter, so what you're implying is that it should be easier to get a decent sound from a sound card (analog outs) than from a transport+DAC combo assuming that both sound card and transport are fed from PC? This is sth new to me, I always thought that it's the other way...

Regarding our topic, after few sessions and multiple combinations I came to a conclusion that 'file size=split size' rule applies for 16/44 upsampled tracks as well. There is no need to increase it further.

[PeterSt]

after few sessions and multiple combinations I came to a conclusion that 'file size=split size' rule applies for 16/44 upsampled tracks as well. There is no need to increase it further.

Hi Marcin - I guess I must read back some posts to understand this exactly. So, what do you mean "increase it further" ?

On the soundcard ...

Peter, so what you're implying is that it should be easier to get a decent sound from a sound card (analog outs) than from a transport+DAC combo assuming that both sound card and transport are fed from PC?

No, that is not what I wish to imply. Both will be subject to the same sh*t, assumed the s* is there, but the influence on digital data is more devistating.
Notice that both digital and alogue data are as analogue as can be, and thus both are subject to the same influences. But, for example, if a most significant bit becomes 1 while it's intended as 0, the magnitude is 32768 (for 16 bit data) "wrong" for the analogue result of it. And while this is about a "data" error, it can also be about clock errors. This is all when the data gets too bad to be processed digitally, and depending on where it goes wrong you may not even have a lock. So *this* won't happen so easily, or it must be very very bad. But :

[and here I just gave the finest description how jitter emerges, but I decided to scratch it away; I just realized that even the best "jitter engineers" here in Holland don't realize that it works like that and I started to think a bit commercially; my DAC should remain the best there is, and I just layed out a large deal of how it happens. Sorry !]

But ... Jitter emerges under the influence of noise, which changes the square wave implying the clock signal. So, the falling (or rising) edges which trigger the samples to (and out of) the DAC will change per "clock cycle" and and the timing of that trigger will vary because of it. This is jitter. So, the less noise, the better it is.

When that same noise would be in the analogue signal, it would be inaudible.

Peter

[Marcin_gps]

after few sessions and multiple combinations I came to a conclusion that 'file size=split size' rule applies for 16/44 upsampled tracks as well. There is no need to increase it further.

Hi Marcin - I guess I must read back some posts to understand this exactly. So, what do you mean "increase it further" ?

I meant that in my opinion there is no need to increase the split size over the track's file size, even when we play 16/44 track which is upsampled (QAP).

[Marcin_gps]

An easy test would be for me to set the split file size and then increase the CPU clock rate and/or CPU voltage. I'll do this and report back...

I've just tried this, and am quite surprised by the results - not what I expected at all...

Just to give some background, I can vary the base CPU rate of 133MHz by between 9x-21x using the mobo's BIOS utility. I normally have it set to 9x. But I've just tried 15x and 21x also, keeping the split file size fixed at 12MB. I used a single track for evaluation throughout; David Sylvian's 'When Poets Dream of Angels', which is beautifully recorded and starts with three acoustic guitars at left, centre and right.

9x 133MHz = 1.2GHz
The sound is as I described it before - very detailed, but edgy and fatiguing after a while.

15x 133MHz = 2.0GHz
The sound is fuller. It's almost like the split file size has been increased.

21x 133MHz = 2.8GHz
The sound is fuller still. Rich and warm. Again, it's like the split file size has been increased further.

So, increasing the clock rate has a similar effect to increasing the the split file size!!! This is not what I expected at all. Certainly, with the CPU rate set to 21x, I have no problems listening with a split file size of 12MB - it sounds so much more refined than with the CPU rate set to 9x.

The only explanation I can offer is this: With a smaller split file size, data is transferred more often (though in smaller chunks). Maybe more CPU power allows things to settle down more quickly, well before the next data chunk is sent.

Of course, God knows what happens to other things, such as RFI, with increasing CPU rates...

Peter?

Mani.

Now try the same with your Weiss interface I think that maybe your motherboard is 'prepared' for higher clocks, providing more stable voltages with default CPU speeds (which now are 2.5-3 GHz) rather than underclocked and undervolted. I'm just thinking out loud, I'm not an electronics engineer.

[PeterSt]

Without meaning to stir the pot, I hope people realise how important this topic is.

Allright. A week back or so, I set my Split File size (SFS) to 12. I don't know anymore whether I did that to just listen and judge it, or that I started to work on the ticks at track part boundaries, but in either case it combined with having another amp to test, different LS cables and more confusing stuff. Next I got a rather severe cold and my ears wouldn't do what they're supposed to.
In the mean time, I received complaint over complaint that I played too loud, which by itself could have been related to the higher gain I have in the DAC which also happened somewhere last week, and I lost my volume reference a bit (including mis behaving ears that is).

Even yesterday, at testing some not more than 20 seconds pieces, I was asked why this all must be that loud. Why ? otherwise I can't hear the ticks.

Last night, those "you play too loud" pestering was out, so I grabbed the opportunity to play without the nagging. Still being bugged by stucked ears, I measured a 90dB as minimum level (@4m), so that should be quite normal.
I played a familiar album, but could not much recognize it; tried to hear through the stucked ears, but couldn't imagine much other than "this is different".
I set down behind the PC to select another album, which was just too late for a new track (not) to start. It started with some deeper drums, and it not only sounded strange to me, but also felt strange (hands on the desk). And *then* I thought, wait a minute, this won't be that SFS of 12, will it ?
So I set it back to my normal 70, and indeed it appeared that the drum now was normal again. Switched back and forth a couple of times, ending up with "70 is better".
I selected Yello - Flag, and went on with what I was doing (which was cooking).

Really within seconds I received such a silky sound from not such an at all silky performer, that I thought my improvements from the last months must have been vast, knowing that I hadn't listened to this album for quite some time. I must add that I was totally ignorant here, and never thought of that SFS = 12, now changed to 70 setting. I was merely overwhelmed by this beautiful sliky sound, which reminded me of switching from SPDIF to i2s.
I listened and listened, and even more ignorantly I started to think what could be wrong that my wife says it sounds harsh and too loud; what can't I hear that she can, despite my little hearing problem ?

Believe it or not, but it really took 30 minutes before I thought of the SFS setting again. It took another 5 because I wouldn't let go of the "beauty", but then I set it back to 12 ...

Oh man ...
This is much much worse than you can imagine ... I think. But the quote from Mani above is there for a reason of course. And yes, I recall his referring to this topic in between the lines a few days back ....
At 12, at least here, the sound becomes totally harsh, things come forward which shouldn't, and there is NO normal bass anynore. Haha, NOW I knew what my wife was bothering this last week. But was I right ?
Put SFS back to 70, and right in the last track of Flag she came home. Uhohh ... not the best example track ...
But nothing happened.

Moved over to a next album, and I noticed she was rather shouting towards me in order to be able to make her words rise above the music level. Uh-ohh.
But still nothing happened.
Then already within the next album, I dared to ask : "So you still hear harshness or anything ?" and the answer was "no, nothing so far".

Allright, I always present you with these kind of "ABX" stories, because they work the best for me, but also hopefully show the realism / facts of matters.

In the mean time, I now know this is so severe that I will go to the bottom of it, because indeed it is so important. It is more important than any Q setting, and it even plainly destroys a well behaving DAC.
At this moment I have no clue as to what can be happening that this influences so much, but regarding the character of it (frequencies come forward which shouldn't) I can even imagine it can me measured.
So, nothing else to do than to "solve" this. If I can, beware. Ok, most probably we have listened to the optimum already, assuming it is some border which is crossed somewhere. But if it goes further and really would be related to the track size or anything (how ???) a real vast improvement must be possible again.
You know I already tried some things, and alhough audible, it wouldn't work stable. Also, this was without clue what to actually achieve. So I guess that's first now.

Help ...

Peter


PS: I have read about the suggestion of the 2nd level cache being 12MB, but I haven't and most people haven't. Still it may be in that area somewhere.

[AUDIODIDAKT]

Did we know that we can control the L2 Cache Size in Windows

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management

SecondLevelDataCache = 0 on standard

256KB: 100 hex
512KB: 200 hex
1024KB: 400 hex
2048KB: 800 hex
etc

You can speed up the PC giving it a value, well I think, didnt try it yet.

Roy

PS: probably it will be set automaticly already by the HAL.

[manisandher]

... indeed it is so important. It is more important than any Q setting, and it even plainly destroys a well behaving DAC.

It's great when 'our' developer finally starts listening to what we're saying

I think this is important:

When I had my CPU rate set to 9x, I preferred a SFS of ~60-70 (for 16/44.1 material). But set at 21x, I'm perfectly happy with a SFS of 12MB.

But let me be clear - when I say "perfectly happy", I'm really only talking about 16/44.1 material with no upsampling in XX. Hires and 4x upsampled material seems to require a larger SFS to sound right. (BTW, I wrote this before I knew what SFS Peter uses.)

The minimum SFS seems linked with the CPU power in some way...

Mani.