XXHighEnd

Hi,

I read the description of cpu core appointment schemes and I see that it is based on dual core architecture. How does it work with tricores, quads and up? I currently use AMD X3 CPU and am quite sure that Scheme-2 sounds better than Scheme-3. Is there any rule that applies to multicore systems?

Cheers,
Marcin

I remember when i had a dual cpu i used scheme 2 as well. With a single dual core, I use schme 3.

I've just unlocked forth core and it's the same - Scheme-2 > Scheme-3. Less agressive, better low extension. Also CPU Usage diagram looks different depending on Scheme selecion. Have a look:

This is crayz. One track is subjectively better with Scheme-2, another played via Scheme-3 :D I put back my previous statement about Scheme-2 beeing better than Scheme-3. Further comparisons show that they're definately different, however Scheme-3 is more refined - I can set the volume louder and music still plays clear.

Next week I should receive my new soundcard and will be able to do some direct comparisons between PCI and USB-to-SPDIF -> DAC playback.

Well, you have a quad core, I had a dual xeon. Do more tests, i'ts interesting :)

Hi Marcin,

It indeed is based on dual cores allright, but up to 4 cores are explicitly used when necessary. But, this (4 cores) is used during preprocessing stages only.

For sound quality purposes there are two reasons two cores are explicitly (!) used only :

1. There are not more than 2 processes (threads) during playback, so more cores are not helpful in the first place, but :
2. The other processes running in your system - spread over all cores in theory - cannot intelligently be dedicated to individual cores (because they are unknown for their activities).

Ad 2.
I think you can imagine that something can be done like "1 core for me (XXHighEnd) and the other core for the rest of the processes", and this is really all. With more cores than two this comes down to "1 core for me (XXHighEnd) and the other cores (!) for the rest of the processes", both with the notice that XXHighEnd (and its associates) itself can be spread over two cores anyway. While there will be additional efficiency in saying "1 core for my 1st process, 1 other core for my 2nd process, and the two remaining cores for the other processes in the system", this is (not done and) hardly helpful. This is because the priorities are involved too, and all works fine together as it is. One exception : the highest priority processes of the OS itself, but a. they can't be influenced (I wish they could) and b. they can't be dedicated to cores. So, what's really left is a better spread of the playback processes, but since there are two only - there's nothing left to do about that. :)

I hope it is clear a little !
Peter

So, is sound quality at all changed using tri or quad core processors?

Not that I know of, can reason out, or heard from. :nea:

But keep in mind the convenience factor with the preprocessing !

I switched to an i7 processor a few weeks ago (don't even ask what I had to do to get it working in the totally fanless and silent Zalman case), and I really appreciate the shorter pre-processing time. This is especially true playing hi-rez Flac downloads.

Another advantage (of a faster processor, not extra cores) is being able to reduce the buffer size in KS Adaptive mode. I'm now at 32 samples with Q1=1...

Slightly off topic: I managed to spend a day at the Munich High End show last week. I said a quick hi to Daniel Weiss and also had a quick chat with the guys from RME. Apparently, the PCIe version of the RME AES-32 card has a latency of 24 samples vs. 32 samples for the PCI version (which I have). I'm tempted to buy the PCIe version...

Mani.

Mani,

Before you throw away some money ... please note that it only may be so that it helps;
It completely depends on the driver, and I think you too have seen how crazy the cpu behaviour can become at smaller buffers. This is pure driver response ! So, at a too low driver buffer setting, all your cpu may be eaten by the driver. To me this makes no sense but happens anyway. And at audio being being idle already ! So, in my case this prevents me from going lower than 128 samples driver buffer size, but which doesn't withold me from using 22 samples audio engine latency.
I can use my 32 samples buffer setting anyway, but won't be able to sleep good looking at the cpu behaviour. So I just don't use it.

Important : As you will know from the Fireface, there is a safety buffer involved as well. IIRC this is 116 samples and it can't be shut off. I don't think this will be much different(ly setup) for their PCI(e) products. So, if the same as with the Fireface, the buffer will be 22+116 = 138 at least.

Best,
Peter

Peter,

What would be the best way to use 2 SSD's in a dedicated PC (not in raid mode)
I was thinking of using 1x 30GB for OS and 60GB for gallery (for Gerards build)
Where should I put xx, on OS drive or on the Gallery drive.
And what about the intermediate files, can I use the advantage of 2nd SSD for that too, I guess not (its about writing speed, spinning disc would be better, I think)
Although I currently use a spinpoint hdd as intermediate drive, I think this solves the issues with "Start engine3 during converting" glitches at low latency settings. (This is why you made this right ?)

Your thoughts please, about the best layout......



PS: Nice build Mani, I7 860 should do 1M SuperPi calculation below 10 sec !
     (mine does 13.790 sec on a Q9550 @ 3.40Ghz) :prankster:

Hi there Roy,

At this moment, I don't think I have the definite answers. It may sound strange to you, but I really haven't.

First about the glitches during startup;

In 99% of cases this works better when you do have *ticked* "Start Engine3 during conversion". Yes, very controversely, but without that it ends up in an unrecoverable mess (the sound). For my system that is (and Scheme-3 !). But, usually the same happens when first the album had to be examined for the Normalization Volume. So, there's a 20 seconds of the whole system being very busy, and when right after that playback starts, well, it just can't cope.

The big problem is that always the SSD is involved, while it has been proven (for my system, and my Vertex one) that it can't write all the data at once. I think I wrote about that in the "Windows 7" topic (or maybe it was somewhere else or ... maybe I didn't post the topic at the time). So, this is about, say, 6 tracks being ok to write to the SSD in 3 secs, but then all stalls and the last two tracks may take 30 seconds. I guess it is during this time that you better not try to play with low latency.

If the above is correct, it means that XX shouldn't be on a SSD. So, the intermeadiate files used for the Volume Normalization, are only used for that. I could change it so that all go over that denoted volume, but it is of no use because it would come down to the same as installing XX on such a volume. The whole point is : no matter what, that volume will keep on spinning when the intermediate files go there.

I am fairly sure this is a W7 problem (the writing to the SSD).

The solution to all is "keeping everything under control" (by now I'm sure I didn't post that "analysis" post I just talked about) but this is complicated. Thus, I already went as far as waiting for 10 seconds (just doing nothing) after 4 tracks had been written to the SSD, then do a next two (two cores) wait 10 seconds, a next two etc., and the overall time went shorter of it. But I couldn't get consistent results.

Now, that something goes wrong in the first place is one, but what it is that can't cope is another, and just yesterday I ran into something that may change a lot of thouhgts and things. As a matter of fact, I am just sitting down here for an attempt for KS Special Mode X or something. Yesterday I could fake something I had in mind (for it), and I had no glitches whatsoever with 44 samples at 352.8 while before this was impossible at all (with 352.8 ) and I had to use Adaptive Mode or live with the glitches. In that faked situation startup was fluent, but far more important (because it tells me more) at bringing up XXHighEnd during playback, it was just normally useable ! aha ...

Let me add to this, the sound became completely different again. I imagined it to be more aggressive (and I am not sure that was for the better). This was one setting though, and the others (controlled by Q1) ... I will make today. The theory (whatever that is :)) must be good though, just because it *exactly* did what I expected ...

Ok Roy, this seems to be offtopic, but it isn't. Because if this works, we're free from the glitches at startup (if not a big mess in the first place) and your proposed choices will be more up to you, instead of that it's dependend on "techniques" we don't understand (because they are OS bugs IMO).

In the mean time, I think I would use a spinning disk for XX (fast copy of intermediate files) and an SSD for the Galleries (fast retrieval), and ASSUME the "better sound from an SSD" just comes from the SSD being in there. Although I have never proven it (or tried to do that), it seems sufficient to have the SSD just in there. But since it's there anyway, I'd put the OS on it.
The fact that the spinning disk will keep on spinning, well, bad luck. I also never proved *that* caused harm, but others may have more experience with this. Next, I see no reason why this should be another dedicated spinning disk; I'd put XX on one of the data disks. But keep in mind ! this disk should not spin down in a minute; it must survive the longest track length ! (or otherwise you will be out of sound and other anomalies will happen).

So much for my thoughts ...
Peter

I'm certainly no expert, but...

I have two SSDs, both 30GB Vertex Turbos, with the OS (W7 x64) on one and XX on the other. Furthermore, I use the 'Copy to XX-drive'. All music and artcover sits on a NAS drive (yes, I know Peter doesn't like these) in my basement, connected via a 1Gb LAN cable. I went for the fastest NAS I could find - a Synology DS1010+.

The important point is that I have absolutely no issues whatsoever! No glitches on playback, very little waiting time on loading (even with full 24/176.4 wav albums), and certainly no limit on how many tracks I can load (up to the limit imposed by the size of the SSD).

However, there is one strange occurance. A few weeks ago I connected my DAC directly to the power amp and started using the XX volume control, which BTW is vastly superior to my pre-amp (a top Pass Labs model). Now, whenever I change the volume and then press play, all loading slows down to a snail's pace. But all I have to do is to press stop, and press play again, and everything is back to normal.

On a final note, I have to admit that I have a wierd way of loading tracks. I can't set the 'Music Root' folder to the correct shared root folder in the NAS drive, so rather than using the Library, I simply use the 'Explorer' button. I tend to first clear the playlist and then load a whole album. Probably not very convenient for most people, but I don't mind. Maybe this way of loading changes things also...?

Mani.

EDIT: I know that SSDs are considered to have a shorter lifespan than spinning drives, especially if you're writing to them a lot. In which case, it's probably not a good idea to put XX there and copy files to it. However, I don't mind - if/when the drive fails, I'll just get another!

Mani, can you somehow elaborate on this ? To me it looks like you're forgetting to tell something. I mean, "I change" the volume is an acitivity, and without further context this tells me that after playback and press Stop, change the volume and press Play causes all to suddenly be very slow. But mind the "press Stop" because this is not what you told. Most probably though your story indeed begins in a "no playback" situation, which would make changing the volume an as dead thing as leaving it alone.
Using -0dBFS vs. -6dBFS (etc.) *would* make a difference though.

All 'n all, can you try to be as clear as possible on this one ? it looks important to me ... (or I do something very wrong, or the OS can't take something).

Thanks,
Peter

OK Peter, I've just checked this. This is what happens:

- Start XX (volume at -15 on start up, from previous session)
- Clear the playlist (10 tracks)
- Open Explorer
- Load an album (12 tracks, all 16/44.1)
- Change volume to -12
- Press Play
- Nothing happens... Here's a correction on what I said before - I DON'T think things are happening at snail's pace, things just aren't happening at all
- Press Stop
- Press Play again... all tracks get pre-processed, i.e. tracks get highlighted in blue, one at a time (in 1-2 seconds)
- XX goes into Unattended mode and plays music

This is totally consistent and repeatable with my setup.

HTH.

Mani.

Ah, that's different. Indeed I recall something like that could happen, but I never dedicated it to changing the volume. If I check it here now, all is right. Also I didn't notice it anymore for weeks, so I guess I have solved it.

Thanks.
Peter

This has been solved ... and IIRC longer ago than the latest version. Maybe you don't use the latest version (0.9y8-c) ? Or otherwise it will be in 0.9z.

Btw ... with "shared root folder", I take it that you mean a "share name" ? like \\St01\drive_c (st01)\WavTest\ in my case.
If so, notice that you can assign a drive letter to that ... See below at the mouse pointer, which says "Make network Connection". But I guess you knew that.

Peter

Sorry Peter, I don't think I ever thanked you for the advice... I think I'll stick with my PCI AES-32 in that case - it has low enough latency to allow me to digitise 2-channel analogue at 24bit 176.4K/192K rates.

This will free up some funds to buy the two (I'm serious!) NOS1 DACs that I want from you. Does that put me absolutely top of the list? :)

Mani.