9Y-5-06 special mode

[PeterSt]

Thank you all. But I'm afraid you ain't seen nothing yet ... As a matter of fact, I unveiled a new dimension where all stopped with WASAPI. Oh, I think I still can improve on that, but here ... haha ... wait.



I have turned the Special Mode into standard way of working (although it will be an option by itself), meaning that Q1 will do its work and all the range is Special Mode now (until it breaks because there is a bandwidth for it). Now, I have been listening and listening and listening yesterday, and from the first second till the last my jaw has been hanging on my shoes somewhere. Not only because of the sound itself and inmense details, but merely because of the constant thinking "how can this ever be possible while all is still 'bit perfect' ?". So, what is actually infulenced here ? is it something we can't measure ? does it impeed for not bit perfectness at all - but farther away than were we measure (like the loop back and recording could be "too close") ?

Yeah, and you thought I was talking about what you can perceive yourself in Special Mode, right ? haha, no ...

We all may take our Guinness Books of Records, and look for the lowest latency possible. So yes, I am back on that (low latency) track for when the best sound happens.
Remember, MS always said 20ms for Kernel Streaming. Btw, I have found why and it would actually be 16ms.

So ... 16ms. Let's see. For 32 bits and 176400 samples per second this is 8 x 176400 = 1411200 bytes per second (2 channel). Per ms this is 1411.2 bytes. Notice that for WASAPI this is the best I can achieve currently, BUT it will merely come down to this same 16ms I mentioned. This is related to the accuracy of internal timers. Thus, 16 x 1411.2 = 22579 bytes, which is 2822 samples. This is the actual guaranteed latency which can be achieved on a Windows system, for two channels of 32 bits at 176.4KHz sample rate.

Summarized : 2822 samples is the guaranteed latency for a Windows system. This is equal to 16ms.

FYI, this is the latency for the communication with the driver. The driver itself may communicate with the hardware at 48 samples or even 32 (which I have seen from IIRC an RME Hammerfall). Such a low latency is meant for the Pro guys, and more common would be something like 1024 or 2048 samples, which is what we would be using for playing music (and which actually doesn't need any latency as such).

So, our Windows program can sustain 2822 samples latency (16ms) and the soundcards can do 2048 at worst and say 48 samples at best.

But here is Kernel Streaming Special Mode ...

Currently (still 32/176400) I am at 176 BYTES of genuine latency. This is 22 samples. This is 0.125ms.
And this is where I just stopped because I didn't want to know the limits yesterday. I played with it for hours and hours without a single tick heard, and with the soundcard at 64 samples (so net latency would be ~0.375ms).


So here is my new dimension. Or actually half of it. The other half I won't tell about, and this is what I am going to exploit further now.
No, you ain't seen nothing yet.

Peter

[Gerard]

How can it be... Hahahaha!! 

 

[boleary]

Been experimenting with 5-06 all morning. Here's what I find:

1. Unable to get into special mode,
2. At Q1=5 distortion is profound and X3PB file shows normal mode.
3. First time kernal streaming sounds REALLY good on my system.
4. Settings for REALLY good: Q1=7, Q2+3= 4, others 0, 2x arc prediction. (these are vastly different from engine 3)

The sound is different, suddenly engine 3 sounds more digital and engine 4 vocals no longer sound recessed in a hall. I need more listening to know if one is preferable, could be a more refined level of sound options have finally sprouted here. 

Might I be in special mode though normal is what the log shows?

[Telstar]

I finally did some listening comparison on the dedicated pc.
y4 with engine 3 Q1=4 and all rest to 0, no upsampling (y5-6 does the same with this configuration).
This is the best sound i have ever got from this system. It has PRAT, it doesnt fatigue me.

y5-4 with engine #4 KS. With the same settings i get the special mode, I had to raise to Q1 to 19 to go back to normal mode. But the problem is that i dont hear a damn difference between normal mode and special mode
So i settled for Q1=18. I tried to raise the other sliders and didnt like or didnt change much.

So, the comparison was between wasapi Q1=4 and KS Q1=18, all attended and without upsampling.

KS is less reliable, sometimes xxhe crash, while wasapi is finally stable.
The sound of KS is like wadding cotton. the midrange is super smooth, while the upper end is rolled off. the voices SEEM more natural but the instruments like cymbals, percussions loose the sparkle and dont ring right to me. I started to experience listening fatigue after about half an hour, whuich is a bad, bad symptom.

It's like the sound is more natural but less realistic.

With my current setup (remember that i am forced to use 2048 samples and that the equalization adds latency), i definitely prefer engine #3.

[Marcin_gps]

Peter, I can't wait to hear this revelations your're talking about, espacially that my new speakers are coming 

http://www.aurumcantus.com/aurumcantu-loudspeaker/loudspeaker_v3m.htm - sweeeet 

[boleary]

Hey Gerard A, Han, Marcin and Tony S, what mode are you in, normal or special?

[ivo]

Hi,

Did try 5-06 engine#4 on laptop via USB. So, I have a problem because my USB DAC is 24/96, but the only way I can get sound out is when I set DAC is 16/44. So, I cannot use DAP or QAP, but so far the sound is really interesting. I would say the sound is really analog and sweet. Need more listening still.
About Qs: So, the above listening was performed with this setting: -4/0/0/0/0. Log file shows "Normal mode". BTW, can someone explain what is Normal and what is Special modes? Which is the one?
I was experimenting with Qs with engine#4 and found correlation: -4/0/0/0/0 gives the most precise sound - meaning that I hear things more clearly and can better recognize instrument sound, however when moving Q1 up (till value 25) the sound gets kind of fuller, but individual things get harder to recognize. As a result I stay with lowest.
Based on above findings I am starting to think (also based on Peter's earlier posts) that by using more Qs (sliding them up) we kind of start loosing audio samples which are sent to play. Maybe I am wrong, but if we listen to lossless music then we should be hearing all audio samples which are in the audio file. Otherwise we move towards lossy music, because lossy music has less info inside. I think by using Qs we fool ourselves because wider soundstage or other things we can achieve with Qs does not mean it is better, we just think it is better however we listen to less samples than there are in the file. So, the above correlates with my listening experience because at -4/0/0/0/0 I hear things more precisely which altogether forms a wonderful music experience.
Peter, your signature says you also play with almost all Qs to 0, so is there at least some part of truth in my theory I described above?

Thanks,

Ivo

[glynnw]

I am still unaware of what is meant by Special Mode vs Normal mode, but I think this version using Engine 4 is the best yet on my system.  I seem to be able to follow individual instrumental lines much better than with Engine 3 (which a month ago was the best I had ever heard).  I just don't want to stop listening.

[earflappin]

I installed 9Y-05-06 today and can report that on my system engine 4 is easily superior to the previous version.  Like the other poster, I'm not sure what Special mode is and how one knows if it is invoked, but I really dig what I'm hearing. Thanks and congrats Peter.

[boleary]

The release notes of 9y--5-05 & 6 describe how to reach "special mode". This is for listening to engine 4, kernel streaming only. Chances are if you didn't do the procedure described by Peter and your using engine 4 you are in normal mode. To know which mode you are in you must tick the log activities checkbox at the very bottom of the settings tab. When you do that a temp data file is created in your 9y--5-06 directory. Each time you start a listening session an X3PB file is created in that temp directory. Go in that directory after starting a session and open the x3PB file. When the file opens look towards the bottom, you will see what mode you are in.

According to the release notes, you should start your Q1 setting high, around 30. Play a song and check the mode, if it is Normal then stop playing and reduce the Q1 setting. Repeat this until you reach the first instance of Special mode when you play a track. According to Peter, Q1 settings below the initial instance of Special Mode will cause distortion. I think one of the things Peter is trying to determine is whether all of us are able to initiate special mode.

When I did this my music distorted when Q1 reached 5 0r 6 so I set it to seven and it sounds great but the log file still shows normal not special mode.  So I guess I haven't heard the best of kernel streaming. This doesn't surprise me given my system.   However, I've been listening to engine 4 all day and it is VERY pleasing.

[PeterSt]

Ok, I'll try again : Kernel Streaming Special Mode is a means of sending the samples to the driver, allowing for ultra low latency. So, even way more low than you can currently use with 5-06. What I exactly do to get there remains a .

[PeterSt]

The sound of KS is like wadding cotton. the midrange is super smooth, while the upper end is rolled off. the voices SEEM more natural but the instruments like cymbals, percussions loose the sparkle and dont ring right to me. I started to experience listening fatigue after about half an hour, whuich is a bad, bad symptom.

It's like the sound is more natural but less realistic.

This description would fairly meet my own, BUT, for the earlier KS versions. What I have here now seems Engine#3 to the extreme, but with a little flavor of "this can't be right".
I am rather sure the trick is in the length of the driver's buffer, and Telstar, notice that with the long buffer you can't achieve the low latency towards the driver. I know that now. The burst the driver needs, takes away the (ultra short) response needed towards the driver (but I'm not complete satisfied with my own description here). So I'm afraid you can't compare.

Again :

It's like the sound is more natural but less realistic.

We could say that this is a contradiction and therefore wrong. But IMO it just describes what is going on with KS and the higher latency operation.

[PeterSt]

I was experimenting with Qs with engine#4 and found correlation: -4/0/0/0/0 gives the most precise sound - meaning that I hear things more clearly and can better recognize instrument sound, however when moving Q1 up (till value 25) the sound gets kind of fuller, but individual things get harder to recognize. As a result I stay with lowest.

Well, this prooves that my simulation of Q1 (normal mode) works, hence does similar as what it does with Engine#3. I didn't really try it myself yet, because I am at, say Q1 = -450 or so. Haha. But towards there the story is the same, but a little bit outrageously more.
Anyway Ivo, this is exactly how people -explicitly testing- described Q1 from Engine#3 in the beginning. And, if you'd make this a psychological thing, it becomes more easy to understand :

When we spit out the individual samples (or close to that) they sound as individuals. When we spit them out by thousands a time, they sound like a mass.

I think very few people will disagree with this, if one at all. The problem is though, from the technical point of view this is complete BS. Still it works like this for KS at least, and I'd say, watch the next version. Thus :

Based on above findings I am starting to think (also based on Peter's earlier posts) that by using more Qs (sliding them up) we kind of start loosing audio samples which are sent to play. Maybe I am wrong, but if we listen to lossless music then we should be hearing all audio samples which are in the audio file. Otherwise we move towards lossy music, because lossy music has less info inside. I think by using Qs we fool ourselves because wider soundstage or other things we can achieve with Qs does not mean it is better, we just think it is better however we listen to less samples than there are in the file. So, the above correlates with my listening experience because at -4/0/0/0/0 I hear things more precisely which altogether forms a wonderful music experience.

I wish that were true, because it would be the technical explanation. But it isn't ... (test for bit perfect and you'll know).
I even would say it is rather the other way around (incurred by the technique again) : the lower the latency the harder it will be to get all the samples "there" in time, the bigger the chance we loose samples, the higher (faked) dynamics get. But this too isn't true. As long as we can check for bit perfectness, it isn't.

All 'n all it is becoming a nice discussion, which btw always has been there, but is today somewhat more extreme because the differences are more extreme.
Wait till the next version ... I guess you don't know what to say anymore, apart from something definitely being very wrong in our thinking.

Peter

[ivo]

Peter,

Still about those 2 modes. It seems I tried Q1 from 30 down to -4 and cannot get "Special mode" entry. Does it mean if I have "Normal mode" then I have higher latency than if I were in "Special mode"?
Also, if there is "Normal mode" at Q1=30 and Q1=-4, then I guess better to use -4, ah?

Ivo

[PeterSt]

Does it mean if I have "Normal mode" then I have higher latency than if I were in "Special mode"?

No. You, apparantly, can go so low in Normal Mode already. But for sure somewhere this will stop; Q1 in 5-06 just doesn't go low enough. Notice it may be USB allowing for this.
So, I just finished the code that will go as low as 0.05ms in Special Mode (and I know I can reach 0.125ms), while 5-06 goes to 1ms as lowest (Q1 = -4).

Keep in mind : in the next version Special Mode is just a choice in XXHighEnd, so once chosen, you will always be in it.

Also, if there is "Normal mode" at Q1=30 and Q1=-4, then I guess better to use -4, ah?

Yes, I think so. But it may be personal.