Questions Y7

[Gerard]

Peter

I have a view questions.

How come that KS now does work with my EMU where it before did not. KS trough special mode 4096 with q1 at 1204. Adaptive mode play's with a buffer mode at 728. (All SPDIF)

Trough USB much lower (SM 5)(AM 256)

With the DAC connected via USB indeed the errors buffers size come's up at a setting. Connected trough SPDIF (EMU) it does not. Only cracks shows me i do not must go lower. 

When those error's do not apear i cant be sure that i am in the right setting or? (meaning Adaptive mode)

Anti Image takes much more time to start playback.

Grtz

[PeterSt]

When those error's do not apear i cant be sure that i am in the right setting or?

Yes ... or sort of. If you hear glitches without buffer errors, they have to come from a "stall of everything" (which is a kind of severe), and even the driver stops working. So, technically nothing is missing, except from some sound from the speakers ...

Anti Image takes much more time to start playback.

That is correct. This is a tradeoff for now with more robustness (it wasn't quite).

The other things ... I don't know at the moment. But I guess this is more for you to find out, while I didn't change much ... ?

Later,
Peter

[Telstar]

Y7 works in adaptive mode with my x-fi (1024 buffer, maybe also lower).

Kudos to Peter

[PeterSt]

Hi Gerard,

How come that KS now does work with my EMU where it before did not. KS trough special mode 4096 with q1 at 1204. Adaptive mode play's with a buffer mode at 728. (All SPDIF)

Ok, I don't know. But what means "buffer mode at 728" ? ... 728 does not exist ...

?

[Gerard]

Hi Gerard,

How come that KS now does work with my EMU where it before did not. KS trough special mode 4096 with q1 at 1204. Adaptive mode play's with a buffer mode at 728. (All SPDIF)

Ok, I don't know. But what means "buffer mode at 728" ? ... 728 does not exist ...

?

Peter,

Sorry i meant 768. But i am back to USB that plays fine with a buffer size 256 (AM).

Another one : Does the latency that we have learned in AM also count for SM? Or can we say in SM that the lower the latency the better.

[PeterSt]

To be honest ... I'm not sure now. What I perceive from the Adaptive Mode low latency is something which seems to sound more stable. "Stiff" if you like. Something like less flutter in a turn table.
But please keep in mind that this time this is as new to you as it is to me. Or maybe you (all) had already much more time to judge it as I had so far. Usually this is the other way around when something new(ish) is introduced.

Tonight I a.o. played Paranoid from Black Sabbath (which was IIRC my very first LP), and out of everything it sounds completely from today for cymbals and all (meaning : very wel recorded). But the point is (I guess), why did I play this one ? I think I liked to hear the stiff/deep involving dark basses, which indeed just "work".

I need some more days myself ...

[Gerard]

Peter,

What about the buffer error apearing after a few minutes. During playback. Does that man that i have to go one step higher in the buffer size? Or do i have to fine tune with Q1?

 

[ivo]

Peter,

By using your method I found that I can live with buffer from 128 - 512 in Adaptive mode with Q1=1. With all buffer values (128,256,512) I occasionally get buffer errors in log file, but very rare. BTW, I understood if buffer size is set wrong, CPU gets utilized a lot more, if it gets set right, CPU usage drops significantly.

Question#1:
Since I found the buffer, can I use it in Special mode then? What is the difference between the Adaptive and Special modes if I know the buffer?

Question#2:
Can you tell us the measurement unit of Q1? If buffer number is in samples, then what is Q1?
Also, I feel good in Adaptive mode with Q1=1, seems like I can also hear good sound with Q1=30, then what is the rule? I assume Q1=1 is kind of better for SQ than Q1=30 in Adaptive mode?

Question#3:
What about Q2-Q5 in Adaptive mode?

Ivo

[PeterSt]

Hi Gerard. I too saw that buffer errors can occur, and they are always related to other activities. But, at least in my system they are inaudible. But keep in mind, you will be seeing 3 of them or so, and 3 is really nothing. So, I did *not* go a level higher in order to avoid them, because I'm more "afraid" to loose the SQ than that I'm disturbed by those missing few samples which I don't hear anyway. Maybe this is not the best approach, but it suits me.

The better apporach would be to find an "external" setting which avoids the buffer errors at all. One of them would be (as it seems) to shut off the Wallpaper. But then of course the again better solution will be that I make the Wallpaper stuff so that it doesn't influence.

Please notice that I myself have to learn how to deal with this all (and what actually can happen) just the same. It is already exciting enough to (be able to) creating something like this which suits everybody with those various soundcards, types (like Firewire vs. USB), systems and settings.
And therefore I can't emphasize enough on how important it is to have the feedback from you all.

But Gerard, to answer your question if still needed : use Q1. Increase the sound device's buffer size probably doesn't even help. But, you may find out for yourself, and I guess it depends on the buffer size in the first place. Thus, with a buffer size of 48 it probably won't help to set it to 96 with the same Q1, while it (I think) will help to set Q1 to 2 (etc.). This is *unless* 48 and Q1=30 also doesn't help, which will indicate that your system just can't deal with the low latency on the driver's side.

I know, this may be all quite tough for you out there, and I hope I'm not overdoing things. But hey, the good cause ...

Peter

[PeterSt]

Wow Ivo, tough questions ... (or smart maybe haha) ...

Ok, the general approach is to let the driver work more efficiently. This, derived from what I see happening with the cpu is nothing is done about it (shown by Special Mode the best, because that goes into ultra low latency as well (and even more, as you know). And, in the very end you can notice this yourself by being able to work without buffer errors, while not matching with the driver behaviour will get you buffer errors with much much higher latency (see Normal Mode and a too low Q1, with my promise that latency is around a 100 times higher there). So :

Since I found the buffer, can I use it in Special mode then? What is the difference between the Adaptive and Special modes if I know the buffer?

The general answer is No because it works technically very different. What I do myself with Special Mode is finding a buffer size which shows more or less normal (cpu), and within that just use 22 samples (I don't know why not lower, but it seems enough to me). But, this is completely depending on the driver, and where I may see strange spikes, you may see just nothing. Now, IMO the right part of the cpu graph is the most important (with 2 cores and Scheme-3) Why ? well, the left part is the driver itself, while the right part suggests to be the audio engine, and although not true at all, it *is* the net effect of it all. Thus, if the right part shows strange behaviour, or I must choose another Q1, or another buffer size. Also notice that with Scheme-3 it is the right part (Core 2) which produces the sound, so I don't want that to have spikes "in the middle of my sound". Those spikes should directly influence the audio engine itself, because remember, those spikes just are not from the audio engine. They will be from the OS somehow, and will represent a kind of sibilance in task switching (second level cache copying), or something like that. Also important might be the fact that I am looking at this under W7, which may be very different from Vista (which is actually what I expect).
So, YMMV, and as long as at least I don't know better it will stay a geeky thing.

Can you tell us the measurement unit of Q1? If buffer number is in samples, then what is Q1?

Undefined. Of course I could explain the process, but I won't. But, the higher the number, the higher the NET latency. This latter may be important to know, because with Special Mode this is undefined, but generally there it is the internal latency.
Internal latency : the latency between the program and the driver;
Net latency : the latency between the program and the DAC (like what musicians need to know).

Also, I feel good in Adaptive mode with Q1=1, seems like I can also hear good sound with Q1=30, then what is the rule? I assume Q1=1 is kind of better for SQ than Q1=30 in Adaptive mode?

It would mean that the driver latency is determining the sound (for you).
Driver latency : the latency between the filling of the driver's buffer and the DAC (thus : device's buffer length).
Notice this explanation is a tad too simple, because more buffers playing a role here, but I guess (!) my first sentence above is true. However, in that case another device buffer size must change your sound (and ever back I didn't know better than that was the only thing changing sound -> before XXHighEnd, so this sure is a factor too).

Keyword is and remains sibilance (which I explained about a year or so ago).

What about Q2-Q5 in Adaptive mode?

No difference from before, and they remain totally unrelated, while they keep on doing the exact same job. One thing, now you force me to think about it :
With the just mentioned sibilance in mind, the ultra low latency kind of forces to spread possible sibilance (and I'm talking about mechanical sibilance) over several groups of "latency samples", while a high latency will more spread *within* the "latency samples group". At least that is how I feel it. Along with this feeling by itself, next goes the feeling that with ulytra low latency the impact of Q2-Q5 may be larger, to the "anomaly" extend. Thus, if you see the group of latency samples (which is the "per amount stuffing to") as the foremost important influence, and see each of Q2, Q3, Q4, Q5 as an impact onto that (each Q with a different means), you may be able to agree with me that the smaller the latency group of samples is, the more unbalanced things may become when *another* influence is there (Q2-Q5).

I better stop now, because I feel dr. Spock already, no matter how serious I am.
Peter

[ivo]

Peter,

OK, thanks for explanations, but still:

Let's suppose i have the same buffer size and Q1 settings in Special and Adaptive modes, then net latency will be lower in the latter, ah?

I have come to conclusion that Scheme-2 works better for me (AMD Dual core) than Scheme-3. If I am not mistaken Scheme-2 has cpu affinity settings for engine3.exe on both cores, so when I use Scheme-3 (engine3.exe) is only in right core and I get regular pops and sound clicks. In other words engine3.exe in Scheme-2 seems like less sensitive to other activities going on in CPU than Scheme-3.

Ivo

[PeterSt]

Hi Ivo,

Your thinking on the cpu schemes is good. Once you see through these kind of things, you can start calibrate everything to your needs (well, they are the same for everybody) and to your further environment (which is different for everybody).

Let's suppose i have the same buffer size and Q1 settings in Special and Adaptive modes, then net latency will be lower in the latter, ah?

No sorry, not correct. This really can't be said, but generally will be true because the Special Mode latency most probably will be lower (smaller) than the device's buffer size. But what if it's larger (which you can see by the number of samples presented to you) ? ... now you won't know anymore. And besides, Q1 settings can't be the same, because it's apples and oranges.

Peter