0.9y-7

[manisandher]

THANK YOU Peter! This is superb!!!

In 'Adaptive Mode', I'm getting most (if not all) of the naturalness of 'Special Mode' (zero digital 'glare') but without the hiccups.

With my modest Atom-based PC (the C.A.P.S.), I have the buffer set to 128 and Q1=2 (with Q1=1, I still get the odd click and pop, which also shows up in the log file).

This evening, I'll check your 'new' thoughts on the Adaptive Mode and QAI combo...

THANK YOU once again!

Mani.

[Telstar]

As said in other thread, on my studio pc it works just great on the lowly x-fi (yeah, i should get a better card). Preferred without upsampling.

[listening]

I looked forward to hearing "Rene Clemencic and his flutes" this evening and read the message about the new version concurrently. My first impressions after a short music session: I was able to find the optimal working point immediately in adaptive mode. The different ancient flutes sound for the first time without that little acute sound - really impressive. Addionally I feel that especially with high definition records  (I tried with 2L) string instruments sound very expressive but without agression.

Congratulations Peter!

Georg

[ivo]

Hi Mani,

I have the buffer set to 128 and Q1=2

How is your CPU feeling at the above settings?

Ivo

[Leon]

Hi Peter,

Just wanted to say I love the detailed and yet full sound of 09y-7 KS in adaptive mode, I just cannot get enough of listening to the music especially songs with a strong emphasize on focals and analogue guitar play. I am not going to start superlatives or try to describe what it is which makes it so much more than it already was. Just wanted you to know I think you did a great job.

ciao
Leon

[PeterSt]

Thank you Leon. I must say, just today I'm listening to it "in full" by myself (as just explained the other topic, Quad Anti Imaging).
So, I just started "We want Miles" CD-2, and ... I can only agree (and not because it is something which I "created" -> always keep in mind the help of you all out there !! ).

Shivers ...

[Marcin_gps]

I like it very much Peter! Adaptive mode Q1=1, buffer size=128. I like it so much, that I'll probably give up from CMP on dedicated server.

[PeterSt]

Thank you for the nice words Marcin. But keep on being critical where needed !

Peter

[GerardA]

But keep on being critical where needed !

Ok, You asked for it...
I tried 9y-7 and it worked ok in adaptive mode with 256 and 88 samples in the terratec.
It sounds very detailed, BUT there is a high frequency ringing to voices (transients) that I think will make my ears bleed if I turn up the volume.
Details are etched out, but there is still the smell of acid around.
Maybe it is my OS-DAC that is unmasked now, or is this version perfect only for NOS-DAC's?
So I went back to 9-y5-8 (?) and the sound was smooth again although with less detail, but more analog like.
Only I did not manage to get it back in special mode yet.
So is it just me complaining? Or maybe some better settings? (DAP, DAI, NOS all the same for this problem)
Sorry to be critical!

[PeterSt]

Ok Gerard, good, and thanks.

I tried 9y-7 and it worked ok in adaptive mode with 256 and 88 samples in the terratec.

What is 88 ? 88 does not exist ...

On a side note, there is no need to go back to an earlier KS version, because in here (0.9y-7) you can do the same, if you're only using Normal Mode or Special Mode (it's probably the latter you used).
But let's start with my question above ...

Peter

[GerardA]

88 samples is the lowest value I can set in the terratec software ,which says 3.6 (or 3.2? (at work now)) seconds latency.
This is for asio, but it makes a difference when I change it so I guess it's just general latency.

Is the previous version exactly the same in normal mode? (Then I'll have to listen better tonight)
(Special mode did not work for me in the last two versions, but maybe I have to try more for that?)

[manisandher]

I have the buffer set to 128 and Q1=2

How is your CPU feeling at the above settings?

It's feeling on top of the world!

I can't quite get total stability with Q1=1@128, irrespective of core appointment (this still makes a difference even though the Atom N270 processor is single core - it has two threads which seem to act it the same way as cores).

But with Q1=2@128, everything really is cool.

Mani.

[manisandher]

Peter, I'm very happy with the sound that I'm getting with Q1=2@128 (Adaptive, of course).

BUT...

I can also get a stable sound (no errors in log file) with Q1=5@32. (Interestingly, setting the buffer to 64 samples in XXHE and soundcard doesn't work as well as 32 or 128.)

I will take a closer listen this evening to see which I prefer. But from a technical perspective, should there be much of a difference?

Mani.

[PeterSt]

Mani - Yes, because (when you change the driver's buffer length) you change the interaction of the driver with the soundcard/DAC. And as I said elsewhere, that is a first phenomenon which changes sound. It always did.

Peter

[PeterSt]

88 samples is the lowest value I can set in the terratec software ,which says 3.6 (or 3.2? (at work now)) seconds latency.
This is for asio, but it makes a difference when I change it so I guess it's just general latency.

Is the previous version exactly the same in normal mode? (Then I'll have to listen better tonight)
(Special mode did not work for me in the last two versions, but maybe I have to try more for that?)

Hi Gerard,

I'm afraid you mess things up a bit (I didn't say you make a mess of it, haha);
If you have a setting of 88 samples in the Terratec, you can not use that for adaptive mode. Why ? well, because I did not take that into account, and there is no matching setting for it in XXHighEnd (I could add it though).

Now, 88 samples would be a very strange value (but it could be). Similarly, 3.6 or whatever seconds latency is a. way too high to exist in reality (I mean, what would the poor Pro guys doing with that) and b. never matches your 88 number. I am not sure which of the two to believe, but I tend to believe neither.

That this leads to unintended sound, seems a "fact" to me ...

Is the previous version exactly the same in normal mode?

Together with your other post (all implying you never got around to Special Mode (working)), to me this tells that somehow you(r system) may not be able to cope with "better detail", so to say. I mean, it seems that we all benefit from the ultra low latency, and assumed (!) that we all are not into "adding a pile of bass" or anything which would be far from neutral/intended, it should be so that the ultra low latency gives us a better representation of what's really in the material. Keep in mind : it is important to keep on assuming that in here we are all like that, which is a tad different from the car radio SPL world record leage. Thus :

That assumed, your system would be emphasizing the wrong things in it.
Notice this is similar to my SSD experience at first, while
a. the SSD seems to make things better, but
b. it just emphasizes (somehow).
and in my system it emphasized something which was wrong. So, the same could be happening in your system.

But before blahblahing around further about this, please first use Adpative Mode as intended, and this means matching the buffer size setting in XXHighEnd with reality ! (and funnily enough, at not doing so, XX would be making sound worse on purpose ... indirectly, but it would).

Let me know what happens !
Peter


PS: No, Normal Mode is not 100% the same, and although I can not imagine it can make a difference, there's at least one person around here to whom I had to tell the same (so, he too perceived a difference which I couldn't explain other than that small change I applied).
But this doesn't seem what you are talking about, and merely compare apples and organges. On this matter, do note that Normal Mode, Special Mode and Adaptive mode are there to have a different sound (explicitly) ! Just like Engine#3 is another beast all together.

[GerardA]

Sorry, Milliseconds of course.
I'll type over the complete list tonight for your info/fun.
Adaptive mode works like you explained with lowering the buffersize (in my case to 256).
Special mode worked before, I'll try to find the old log!
And of course I'll try to make it work again and write down all the settings.
After that the Wife Acceptance test!

[PeterSt]

Ok. But make that milliseconds microseconds, and I will think of believing it. Haha.

[GerardA]

OK, I'll make it microseconds, but that's as far as I'll go!

[PeterSt]

Just for fun :

1 / 176400 = 0.00000566893. This times 88 = 0.00049886584.

I guess that is 498 microseconds.
But we'll see when you get home.

[GerardA]

Maybe: 1 / 48000 * 88 samples * 2 channels = 3,67 ms ? 

[GerardA]

OK, here are the results:

        3.6 ms  5.6 ms  9.6 ms 13.6 ms 17.6 ms  25.6 ms  37.6 ms
44      88       176     352      529       705       1058      1587
48      96       192     384      576       768       1152      1728
88     176       352     705    1058      1411       2116     3175
96     192       384     768    1152      1536       2304     3456
192    384      768    1536    2304      3072        313      2617

So these are the samples.
What is the buffersize with  these????

[PeterSt]

Ok Gerard - Try 48 and 96, and anything else that shows up in XXHE. Ok ?

[GerardA]

Hi Peter,

I checked the old log files and on 21/2/2010 all logs said special mode.
Does the data give any clue to what my settings were?


Name wanted  : PHASE 24 FW (0) MC Out >> #spdifpcmout01  Number of KS devices found : 1
o
Filter found : \\?\tt1394_enum#tt1394_avs_000aac0400#000aac0700cd6917#{6994ad04-93ef-11d0-a3cc-00a0c9223196}\spdifpcmout01
Name attempt : PHASE 24 FW (0) MC Out ** Found ** ** Succeeded ! **
No ambiguous devices found.
-13
-14
-15
-16
Data-1 : 0.34  Data-2 : 34.1  Data-3 : 8173  Data-4 : 1025  Data-5 : 1024
-17
-18
-19
-28
-29 (Start Playback)
Special Mode !

[PeterSt]

No, sorry. Only that you applied 1024 samples, but not with which buffer. And you seem to have a few there.
Of course it must have been one which allows the 1024 (which the lower buffer sizes won't I think). But even if you would have found the in XXHighEnd set buffer size, it still wouldn't say much about the driver's buffer size. Notice though that only a few match the XX suggestions, and maybe you recall to have chosen matching ones (which is what I would have done). This makes the list of possibilities smaller ...

[boleary]

Finally had an opportunity to hear Adaptive Mode yesterday. My experience is similar to GerardA's: Adaptive mode presents on my system with too much detail at the expense of the more "organic" sound that I get with 9y-5.06. ( BTW normal mode with 9y-5.06 sounds better here than normal mode on all updates since the "Peak Extension" checkbox was created.)

Obviously, as Peter continues to hone in on the best sound for his NOS dac, folks like myself, using old school stuff, my OS MSB dac was new in 2002, will not be able to experience all the refinements. This is not to say that my experience of XX isn't "worthwhile", cause the sound that I'm getting here with 9y-5.06, given my equipment, is really wonderful.

I have a hunch that one significant reason that Adaptive mode sounds too "detailed" and less "organic" is that my means of usb/spdif conversion just adds to much jitter which then becomes amplified with "more detail".  Will be able to test this theory when I finally receive that frickin' Hiface gizmo that, as it turns out, once was lost but now is found!

Question for GerardA and Peter :

I'm afraid you mess things up a bit (I didn't say you make a mess of it, haha);
If you have a setting of 88 samples in the Terratec, you can not use that for adaptive mode. Why ? well, because I did not take that into account, and there is no matching setting for it in XXHighEnd (I could add it though).

Now, 88 samples would be a very strange value (but it could be). Similarly, 3.6 or whatever seconds latency is a. way too high to exist in reality (I mean, what would the poor Pro guys doing with that) and b. never matches your 88 number. I am not sure which of the two to believe, but I tend to believe neither./quote]

What are you guys talking about here? Isn't the sampling frequency necessarily 88.2 when using 2x Arc Prediction or are you referring to something else? Not being quite the GEEK as others around here, I'm sorely challenged trying to follow these conversations.....

[PeterSt]

Well, you could be right, but remember, it is my own idea about it, and this is not proven right, yet .

Another thing : I wouldn't be so sure your MSB isn't oversampling, although it depends on the model if and how much (no, don't let me look it up, because if their website is as it ever was ... the worst).

IMHHHO you have one of the best DACs around ... I probably would have chosen that, weren't it that, well, today they oversample (a little only).

What are you guys talking about here? Isn't the sampling frequency necessarily 88.2 when using 2x Arc Prediction or are you referring to something else?

Uhm, what you say is quite correct, but this is about 88200 samples, while I (and Gerard's driver) talk about 88 ...

Peter

[boleary]

  Well, I spoke with the MSB folks a while ago and they confirmed that my DAC is a delta-sigma, so I have presumed that it must be oversampling. It does sound very good. The original op-amps were replaced with B.B. 627's.

[GerardA]

Hi Boleary,

One way I'm glad you confirm my problem...
I hoped to receive my Hiface yesterday, but somewhere a delay happened.
With the Hiface at least we're sure it's not the jitter that gives us the jitters.

The 88 means that for the terratec 88 samples are used in the buffer which is an equivalent of 3.6 milliseconds.
If one sample has 2 bytes of 8 bits and two channels then I guess the buffersize is 2*2*8*88 = 2816 but I'm not sure of that, let alone if this can be handled the right way, and what setting would be best.

I hoped to get some quidance, because I really want the special-mode sound back that I (we) had with 9y-5-6!
Maybe 9y-8 can do the trick, gonna try that now.

[GerardA]

Well 9y8 is really stable, no cpu-problems.
Soundwise it's the same like before.
Like a resonance somewhere between 2 and 8 kHz which especially vibrates in high female voices.
I think now it's not really in the transients but more in the vowels (klinkers) and long notes of violins and brass.
Maybe its a cone breakup of my metal cones...
Hope you can find something Peter!

[boleary]

Thanks GerardA, guess i know what to expect with 9y-8. Will be anxious to hear your experience with the Hiface. I think I'll have mine by the end of this week, hopefully. One thing I've noticed with Adaptive mode is that it sounds better with orchestral classical music, whereas it definitely has issues with the higher frequencies of vocal singer/songwriter stuff (my favorite).

[PeterSt]

Gerard - 0.9y-8 doesn't change a bit or byte.

And, I just don't understand that table you presented.

FYI : One sample is always indicated as an "x channel sample" where x in our case is 2. So, one 16/44100 sample comprises of :
2 bytes (= 16 bits) per channel = 32 bits = 4 bytes for 2 channels = 44100 * 4 = 176400 bytes per second @ 16/44100.

1 sample thus (regardless of bit depth) takes 1/44100 = 0.0002267573 seconds. E.g. 88 samples take 88 times of that = 0.00199546424 seconds. This is 2ms.
At 176400 samples per second this is 1/176400 = 0.00000566893 for 1 sample, and 88 samples take 0.00049886584 seconds, or 499 microseconds (half a millisecond).

Maybe I am wrong somewhere, but if not, I don't understand that table.

Of course this has nothing to do with the (not so good) sound you perceive from it all.

[GerardA]

      3.6 ms  5.6 ms  9.6 ms 13.6 ms 17.6 ms  25.6 ms  37.6 ms
44      88       176     352      529       705       1058      1587
48      96       192     384      576       768       1152      1728
88     176       352     705    1058      1411       2116     3175
96     192       384     768    1152      1536       2304     3456
192    384      768    1536    2304      3072        313      2617

The first column is the samplerate, the other 7 columns are the 7 settings  I can do in the software.
Like at 44 kHz and a setting of 3.6 ms the display says this is 88 samples.
If this is true I dont know.

BTW. Apart from the one complaint it sounds super!

[PeterSt]

Yeah, but the "sibilance" in women voices seems a clear case of emphasizing the wrong. It's one of the most difficult things anyway (ever tried Diana Krall ?).

Let's say we're not done with this one, okay ?

[GerardA]

No not done yet, are you ever done?

But it is not the sibilance, it's more the aaaaa, eeeee, and iiiiiiii!!
Like a nylon string starts to sound like a metal one.
I have no musical education, so difficult to explain!

[xp9433]

GerardA

You said "Like a resonance somewhere between 2 and 8 kHz which especially vibrates in high female voices.
I think now it's not really in the transients but more in the vowels (klinkers) and long notes of violins and brass.
Maybe its a cone breakup of my metal cones..."

As a matter of interest have you tried the same track with reversed absolute phase to see if this makes any difference to the above sound?

Frank

[ivo]

Hello Peter,

Just installed yesterday the latest 0.9y-8, but this is also with 0.9y-7 using engine#4, no up-sampling, just going straight 16/44:

So, it seems that *on some albums (FLAC)* there are little holes during playback - kind of very short drop outs in sound, however there are no buffer errors in log file and there are no CPU spikes as well. Average CPU usage = 5-8 %. nothing else being done on machine. DAC is connected via USB, see my sig.

Setups problem seen with: Adaptive and Normal modes, Q1=1-4, Device buffer 128-256, Schemes 2 and 3. There are albums when I can easily go with Q1=1 and Buffer=128, but there are albums where Q1=4, buffer=256 and still occasionally small dropouts. BTW, tried the same tracks via engine#3 - all goes smooth.

Question1: Does *Adaptive* also means there is something dynamically being adjusted with buffers during playback?

Question2: Should I go and increase Q1 even more up? I guess higher Q1 is not that good for SQ in KS mode?

Question3: Earlier you mentioned that Q1's unit is undefined, but to my mind XX' s code contains either function or object, where Q1 serves as an input value. If that is so, then what is that function's/object's output then? What is the unit? Is it time or sample or offset or coefficient? I am not trying to understand your code, just want to understand what is the slider I move so frequently!

Otherwise when no drop-outs occur, playback is just amazing, faboulus, fantastic and whatever else great!

Thanks,
Ivo

[Marcin_gps]

Gerard - 0.9y-8 doesn't change a bit or byte.

And, I just don't understand that table you presented.

FYI : One sample is always indicated as an "x channel sample" where x in our case is 2. So, one 16/44100 sample comprises of :
2 bytes (= 16 bits) per channel = 32 bits = 4 bytes for 2 channels = 44100 * 4 = 176400 bytes per second @ 16/44100.

1 sample thus (regardless of bit depth) takes 1/44100 = 0.0002267573 seconds. E.g. 88 samples take 88 times of that = 0.00199546424 seconds. This is 2ms.
At 176400 samples per second this is 1/176400 = 0.00000566893 for 1 sample, and 88 samples take 0.00049886584 seconds, or 499 microseconds (half a millisecond).

Maybe I am wrong somewhere, but if not, I don't understand that table.

Of course this has nothing to do with the (not so good) sound you perceive from it all.

yyyy, does it mean that the higher sample rate, the lower latency is possible? or am I missing sth oO
because basically I can get lower latencies with my m-audio at 96kHz than at 44kHz.

[PeterSt]

In theory : yes. But then this counts when you reached 1 sample. About the more normal situation, I really don't know ! To me it seems it doesn't matter (or shouldn't) because it is about the response of the system to the driver needing samples, and how fast they are consumed at that other side (from the driver towards the DAC) is unrelated (sort of, but notice that for that too system resources are needed). My feeling though says that with the higher sample rate a lower latency can be achieved, yes.

[PeterSt]

Hello Peter,

Just installed yesterday the latest 0.9y-8, but this is also with 0.9y-7 using engine#4, no up-sampling, just going straight 16/44:

So, it seems that *on some albums (FLAC)* there are little holes during playback - kind of very short drop outs in sound, however there are no buffer errors in log file and there are no CPU spikes as well. Average CPU usage = 5-8 %. nothing else being done on machine. DAC is connected via USB, see my sig.

Setups problem seen with: Adaptive and Normal modes, Q1=1-4, Device buffer 128-256, Schemes 2 and 3. There are albums when I can easily go with Q1=1 and Buffer=128, but there are albums where Q1=4, buffer=256 and still occasionally small dropouts. BTW, tried the same tracks via engine#3 - all goes smooth.

Question1: Does *Adaptive* also means there is something dynamically being adjusted with buffers during playback?

Question2: Should I go and increase Q1 even more up? I guess higher Q1 is not that good for SQ in KS mode?

Question3: Earlier you mentioned that Q1's unit is undefined, but to my mind XX' s code contains either function or object, where Q1 serves as an input value. If that is so, then what is that function's/object's output then? What is the unit? Is it time or sample or offset or coefficient? I am not trying to understand your code, just want to understand what is the slider I move so frequently!

Otherwise when no drop-outs occur, playback is just amazing, faboulus, fantastic and whatever else great!

Thanks,
Ivo

Pwew, I don't know what to think of dropouts despending on the album, except from "can't be so". But I know, you won't be helped by that.
Are you sure this isn't the case for all albums, but that only a certain type of music shows it to you ?
In any case I would try another PC. I mean, drop outs without a spur of extra CPU (and thus player IO because that always goes along with CPU right after it), while towards the program also nothing is shown from buffer errors, must be a PC thing. Don't forget the driver though.


Q1 : No.

Q2 : No. But this makes me think ... It looks like you may be doing something wrong. I mean, how can it be that you work in different buffer sizes, while your USB device doesn't allow you to change the buffer size. Hmm ... I guess you didn't follow the procedure (for Adaptive Mode) to get to the buffer size and use that. So, that is 128 or 256 or whatever, but not 128 *and* 256, as you said (anyway, that is how I read it). So, when you found your buffer at 128, I think you should stay there, and increase Q1 when needed. You may even accept buffer errors (to test) but avoid the dropouts with it (??).
But also keep in mind what I said earlier in this post.

Q3: No, is not so. Specially for you : Let's say that sliding up Q1 (this is Adaptive Mode only !!) gives more headroom for other tasks to do their work, oriented towards the driver task. In the end sliding up means higher latency, so if it is that you wanted to know, now you do. But to keep in mind (I told it before, somewhere), this is net latency. Thus, this "slider you use so frequently" this time changes net latency (which is what musicians would observe).
So, you get that ?
Haha, right. Normal Mode does that too, but the KS means used is very different, and therefore the latency is way more high to begin with. This is still no normal Kernel Streaming because it allows a latency of minutes for certain devices !! (the first two KS versions used that). Currently this is not supported anymore, because not all devices support it (like USB).

Be warned : the more times you ask about this, the more you will get confused.

Peter

[ivo]

Thanks Peter for explanations.
I am actually waiting on HiFace arrival, let's see how it goes with it. I hope it will perform better as it has custom drivers and it works in async mode ove USB.
At least I know now what Q1 is doing

Ivo

[Telstar]

IMHHHO you have one of the best DACs around ... I probably would have chosen that, weren't it that, well, today they oversample (a little only).

Peter, note that the Gold link is NOT the DAC III Platinum, which I finally heard a few months ago and it sounds REALLY good indeed.