9-6-1c FLAC messing with core appointment scheme?

[CoenP]

Peter,

Here's one for you:

I noticed the processor not being used in a way I was used to. I allways have the taskbar ready to get some insight into performance or SQ problems.

What is happening I illustrate with the attatched pics.

Pic 1 is a picture of XX playing a WAV file with settings as in my signature after a fresh reboot.

Pic 2 is a screenpicture of XX playing a FLAC file directly after the WAV with same settings.
Is this the cause of the difference I have been hearing between FLAC and WAV???

Pic 3 is a screenpicture of XX playing the same wav as in pic 1 after the FLAC of pic 2 (same settings).
The right -wav- part of the processorchart is obviously different from Pic 1.   



Two more things:
- same situation in unattended
- same situation when XXEngine3 is stopped before the "pic 3"

And pic 1 and pic 3 situations sound different!

regards, Coen

[PeterSt]

Nice ...

But as far as I'm concerned you are heavily overdriving your system. Do the same after a reboot and the same most probably will happen. But copying the first "pic" somewhere along the line will be the most difficult. So, apples and oranges.

And a way underperforming PC ... (small notebook probably)

Edit : well, it's in your sig ...

Peter

[CoenP]

Peter,

Thanks,

A 'non clipped' pic 1 is how Scheme 3 would look normally, isn't it?

Is there any way to relieve the poor basterd form its cpu burden?

regards, Coen

[PeterSt]

Hi Coen,

It is the first picture which will not be the normal situation. But, right after a reboot all is different. That's why I think you won't be able to copy that situation anymore, unless doing the same after a reboot again.
So, that second picture and further, things start to get spread over the cores, while the OS should have done that in the first place. But, it depends on what we see here (see below as well).

The "clipping" by itself is allowed, as long as it doesn't take too long (it doesn't here). But what worries me more is that this looks like to come from the USB transfer "blocks", while this shouldn't reach any 100% at all. So, you are close to holding back the USB transfer here.
In my picture you see XXHE in core 1 and the driver in core 8 (also 32/705.6). Where the USB transfer is ? I never saw it.
(this is Attended and Not Minimized, SFS 430, Buffer in Control panel at 2ms)

You may set Q1 to a lower amount to see it change (assumed I am right of course).
If you have set your SFS to 1 or so, it is that what I'm seeing. But if so, setting it higher and all will die.

I am sorry, but this won't be suitable for good SQ means. If you think it is okay, then it is okay of course.

Peter

[CoenP]

Thanks for the reply.

This whole 'cpu maxing out' thing since the nos usb was installed didn't make me feel confortable about getting the goods from it.

I will update the mobo drivers and expand memory to 8gb now i am running with win7 x64. Furthermore I can throttle down the AP and reduce q1 (now at 34 Mb) for an easier load.

It should be possible to run nos usb with an atom satisfactory, since Mani does it and finds little sound related objections (though he does have some practical issues). Actually the low powerconsumption enables him to use a linear psu (probably with some dc-dc conepveraion down the stream).

In this thread:http://www.phasure.com/index.php?topic=1673.msg19195#msg19195

My Atom PC is based on the 'CAPS' computer: http://www.computeraudiophile.com/content/Computer-Audiophile-Pocket-Server-CAPS-v20

However, I would NOT recommend this PC for use with XX. The maximum SFS that I can set is around 150 (I have it set to 100). Even with this SFS setting, I cannot get gapless playback - the system simply isn't powerful enough. Now for me, sacfricing gapless playback for a smoother, less edgy sound is worth the price. But there's no guarantee that this will be the case for anyone else.

My recommendation would be to wait until ultra-low wattage/TDP designs become as powerful as today's i5/i7 designs - it surely can't be too far away. The other alternative would be to look into a linear supply for a current i5/i7, but I think this would be too crazy... even for us Phasure posters!

Apart from not coping with gapless playback, my Atom works quite nicely with XX, especially in Unattended with OS minimised. It's quite a lot slower than my i7 when playing with the XX GUI, but not painfully so.

Mani.

No idea of his cpu load and xx profile for this setup.

Regards, Coen

[CoenP]

Ok,

All drivers are updated and the 8 gigs are in and i can push up to an sfs of 144. This increases the time between the peaks, but has a negative side effect of sudden drop outs and skipped (smal) parts of the music. Dropping q1 does not seem to have an effect on the load.

Save for these interruptions of sorts, i think it sounds excellent. Too bad  it doesn't work properly .

With quad ap things get a lot easier for the pc. A much larger interval between peaks (still there) and a smooth ride through the tracks. Definitely a different kind of sound, all seems a little smoothed over, however very nice by itself.

The "clipping" by itself is allowed, as long as it doesn't take too long (it doesn't here). But what worries me more is that this looks like to come from the USB transfer "blocks", while this shouldn't reach any 100% at all. So, you are close to holding back the USB transfer here.
In my picture you see XXHE in core 1 and the driver in core 8 (also 32/705.6). Where the USB transfer is ? I never saw it.
(this is Attended and Not Minimized, SFS 430, Buffer in Control panel at 2ms)

Any suggestion what could be the matter on the usb side? With my pci soundcard i also had a rythmic peak but only in the 'left' core.
Minimised or not, it does not make a difference of the cpu load.

You may set Q1 to a lower amount to see it change (assumed I am right of course).
If you have set your SFS to 1 or so, it is that what I'm seeing. But if so, setting it higher and all will die.

Q1 below 15 with quad-ap makes a difference in 'cpu baseload' from  average 7% to 3%. Sfs in pics above is 34...

I will update my signature (nos usb on 12ms and sfs of 144).

Regards, Coen

[CoenP]

New signature see below.

To avoud the clicks and drop outs @ 16x ap i run the player now at normal priority (was real time). A also learned to exit the nos driverapplication, this causes only more clicks and dropouts.

Unfortunately sfs is maximised by the music stopping out for a few seconds. An sfs of 140 works flawlessly. I experimented up to 180, which worked allright and even sounded better, but i get regular drop outs of 3 seconds.

I also changed the bios to enable hyperthreading before conducting the optimisations. Now there is a nice peak in core 1 and 2-3 spikes per interval in the other three cores. Base load reduced to 5% av. The sound gained some colour and freedom.

So it works yet (still?) suboptimal and i am not complaiming about the sq. 

Regards, Coen

[PeterSt]

Hi Coen,

Well, it seems that you sure "learned" what causes what. And as you will have concluded (I think), it now just is the cpu which is not fast enough to cover for it within the time it is given. But :

Notice that there may be a small contradiction, which is about the hyperthreading being active now, which cuts the core speed in half. So, if that doesn't slow down the process (and it seems to me that it only speeded up), you really must think about the disk subsystem being too slow. Could be USB connected, slow Ethernet, otherwise ?

Regards,
Peter

[CoenP]

Hi Peter cs,

Let me start with the best wishes for 2012!

Some month ago I did a speedtest on the hdds and those matched the expectations of data transferrates. The os and music discs are connected with sata2, nothing else than the nos1 on the usb side of things.
I did select 'ide' mode on my mobo that also has an old style pata connector.
Some xx versions ago I did some testing on ide vs 'ahci', but found little difference in performance and worse sound for ahci.

Another thing is the volume adjustment that -no surprises here- leads to lower processorload when multiples of 6db are selected. It also seems that mono recording exibit less cpu load.

I don't feel like putting much more time in this setup and have now some sandy bridge on order (i5-2500/ z68) that should do all xx tricks with two fingers in the nose . And on top of that with comparable power consumption.

Regards, Coen

[CoenP]

Hi all,

Last weekend I finally assembled my new pc. Amazingly everything worked from the start. What surprised me the most was that win 7-64-ult booted with the new mobo with new processor. I had the wrong setup-boot usb that didn't boot as intended, yet everything on the 'old' w7 install worked fine.

Well, almost everything, since the processorload pattern discussed in this thread is still there. That is the music is loaded with (now very brief) peaks on all cores. I did not expect this to happen since it was believed to be a result of the poor atom performance. This setup is about 10X as potent...

Sq wise the sound is very similar to my old setup, yet a tad bit less relaxed and full. I have to play with the hardware and bios settings and xx setup to get sq on par with what i am used to. Some mobo burn-in may also play a role.

I am very happy with much  improved response of xx. The pc is not completely silent yet. Processor temperature stays very low, i am considering a completely passive setup.

Regards, Coen

[PeterSt]

That is the music is loaded with (now very brief) peaks on all cores.

A bit depending-on, I think this is normal.
Try and see whether the interval doubles when you halve the SFS ...

Peter

[CoenP]

A bit depending-on, I think this is normal.
Try and see whether the interval doubles when you halve the SFS ...

Peter

Thanks for the reply,

Well, the interval Halves when SFS is Halved.... (from 220 to 110) iow the frequency of the peaks doubles when halving the SFS...

I guess this is ok.

I reduced the NOS driver buffer to 4 ms and see more activity on the first core.

regards, Coen

[PeterSt]

All matters. For fun (only that) you can also look into the combinations with the Clock Resolution.

I recall you just bought a new PC. Why an i5 ?
(seems such a waste of 4 additional cores)

Regards,
Peter

[CoenP]

All matters. For fun (only that) you can also look into the combinations with the Clock Resolution.

I recall you just bought a new PC. Why an i5 ?
(seems such a waste of 4 additional cores)

Regards,
Peter

Well why not an i5? The performance per core does not improve with next up i7 while being much more expensive. The interesting i7s are way out of budget (actually the i5 was allready a stretch).

The ui runs smoothly and music plays with all settings (high sfs, low nos buffer). Sq apparently suffers from the usb mouse, but that is another matter to be solved.

Regards, Coen

[PeterSt]

Well why not an i5? The performance per core does not improve with next up i7

Correct. But the number of cores doubles.
And notice that now the performance per core halves. But it is not about that ...

Peter

[CoenP]

So what is it about then  !

With Intel ht you don't really get more cores, its just the same cores that are loaded somewhat differently by the software (below 100% load anyway).

What more is there to gain? Track peaks are 100% load and in inbetween zero (save for some pre and after peak activity).

Regards, Coen

[praphan]

Hi Coen,

You might probably know that XXHE is multi-threading sensitive application. Speed per core is not as important as more cores.

Intel disables HT on their chip and labels i5. This way they can apply premium pricing strategy by enabling HT on i7 chip. They are all coming from the same wafer.

So you have desktop i5 in which HT is not available. Please check this out :
http://ark.intel.com/products/series/53248

However, the mobile i5 has dual cores with HT to 4 threads.

I ran 12 threads and it is lightening fast to load FLAC tracks.

Hope this gives you more transparency.
Rgds,
Praphan

[PeterSt]

Coen, HT = Hyperthreading (each physical core can behave as two logical ones). An i5 can't do that.
For an i7 it must be switched on in the BIOS.

Further see Prapan's post.

[CoenP]

Yes, I am aware that HT is not possible with i5.

Is this multi core thing only about FLAC decoding or more about SQ?

I usually play WAV for optimum SQ.

regards, Coen

[praphan]

Coen,

Both.
Please refresh your memory with this post :
http://www.phasure.com/index.php?topic=1356.0
Scroll down to
But we have more cpu's - ehh, cores of them

Regards,
Praphan

[CoenP]

Thanks Praphan,

I am just getting the finger behind 'real' vs 'HT' cores. The virtue of HT is more efficient use of the cpu when multi threaded applications run. This effect is real but is in the order of 10% more instructions in a given time (in those specific mt applications) than without HT.

Sq wise this may be important. Critical listening to ht on/off on my atom setup favoured the standard two core (ht off) over the "4" core (ht on) situation though.

Increasing the count of real cores most certainly helps. I have been considering amd legacy 6 core and 8 core bulldozer for that matter but found bost cost and powerconsumption prohibitive. The same costlimitation goes for the new 6 core i7s (from 580 euro up).

The i5 system certainly feels responsive and handles all -demanding- configurations with ease (no glitches whatsoever).

Regards, Coen

[praphan]

Great Coen.

Happy listening 
Praphan

[CoenP]

This weekend I did a little speedtest.

Since i have an unlocked processor, I switched of EIST and underclocked the processor to 2.5GHz (coming from 3.3 with 3.7 turbo). This seriously impaired the handling of XX. Everything takes much more time, for a split second I even had the idea that te atom was back .

This confirms my idea that pure core horsepower remains important. I am convinced that the system could even be faster with a cpu that supports more threads, but that comes at a price premium that you have to be willing to pay.

Happy listening to ya all,

Regards, Coen