Tír Na HiFi

Aleg wrote:

sbgk wrote: I've tried different sizes of streaming and decode buffers

large streaming buffer gives delayed start
small streaming sounded better
large decode sounded better
was able to have 2 tracks in memory and could play them repeatedly without lms running, don't know if more tracks could be loaded like this

the squeezelite architecture takes the highest quality ie wav and gives it the same treatment as the lowest, so not much hope of better sq without a bit of work, am going to try and get pa working on mqn to see if it's worth doing any more with squeezelite.

I also tried a few.

What I think I'm most pleased with is smallest stream buffer possible and a very large output buffer.

Am now listening to 260:8000 (255:8000 did work earlier but not now).
I feel this got some more clarity and additional depth. Esp. increasing the output buffer from 300 to 8000.
Am not completely sure though, as my hearing is not at its best at the moment (have a slight cold).

Cheers

Aleg

Thanks for tip Aleg. Itried your settings with -a 3 and got lovely SQ. I am sure though that these are system specific - I am using WaveIO with Thesycon driver 2.15

the squeezelite architecture takes the highest quality ie wav and gives it the same treatment as the lowest

All I can say is that it is scary if that is true and can't wait for further progress.

Cheers

Can some of ye try a 0 b 260 4000
could be just good because I'm too lazy to but in another 8gb ram

thanks Aleg good find

and 4core affinity is best if you have em

cvrle59 wrote:What does "pa" stand for here?

Portaudio - an open source library that provides a portable, operating-system independent interface to the sound system of your computer. It is designed to allow portability of audio applications between Windows, Mac OSX and Linux. Audacity is probably the best-known application using portaudio.

Personally I think portaudio is a somewhat strange choice for the tweaks/optimisations we are discussing in this thread, as they are very much non-portable - not only are they windows-specific and processor architecture specific, but at least theoretically affected by lots of low-level system details.

Julf wrote:

cvrle59 wrote:What does "pa" stand for here?

Portaudio - an open source library that provides a portable, operating-system independent interface to the sound system of your computer. It is designed to allow portability of audio applications between Windows, Mac OSX and Linux. Audacity is probably the best-known application using portaudio.

Personally I think portaudio is a somewhat strange choice for the tweaks/optimisations we are discussing in this thread, as they are very much non-portable - not only are they windows-specific and processor architecture specific, but at least theoretically affected by lots of low-level system details.

What does Adrian/Triode think ?

sbgk wrote:What does Adrian/Triode think ?

No idea - I guess you should ask the organ grinder and not the monkey.

In this case the organ grinders would not be just Adrian, but also Ross Bencina, the portaudio V19 lead, and maybe Stephane Letz (ASIO implementer) and/or Dmitry Kostjuchenko and David Viens (WASAPI).

Julf wrote:

cvrle59 wrote:What does "pa" stand for here?

...
Personally I think portaudio is a somewhat strange choice for the tweaks/optimisations we are discussing in this thread, as they are very much non-portable - not only are they windows-specific and processor architecture specific, but at least theoretically affected by lots of low-level system details.

Who cares if that is what it takes to make a difference that is worthwhile?
It only shows that uniformity over all platforms will make use of the smallest common denominator and is probably the worst solution for all, being suboptimal for all platforms involved.
It is just more easy for a developer, but that's what I said: it has been made by a bits-are-bits man and that has been shown to be a grave mistake.

uploaded portaudio_x64.dll v18 and squeezelite_win.exe v20 - optimisation setting tune and profile guided optimisation of squeezelite.

Sounds quite different to previous versions, hopefully the flat sound has been removed, wonderful thing pgo. Haven't managed to do pgo with portaudio yet.

Don't know if this will help Gordon in developments but taggart and I have been doing a few tests; we both use the WaveIO USB/SPDIF converter with the Thesycon v2.15 drivers and running Windows Server 2012 R2 with MMCSS disabled. The Thesycon driver allow one to set buffer latency. The 'Minimum' setting is 1ms and the next one up is 'Low' 2ms. Both with TotalBufferCount of 2.

Although taggart runs Squeezelite on an Intel NUC with Intel Core i5-4250U (Haswell) CPU and 8GB RAM and I use an Intel D510 mobo with Atom 1.6GHz dual-core CPU and 4GB RAM (much less powerful), we both noted that 1ms latency strained the CPU and 2 ms Latency made the SQ better and fuller (and really good). We used the same Squeezelite settings for the test of -a 5 -b 280:8000.

We looked at CPU activity in Task Manager and although squeezelite-win.exe was around 1% CPU it was Windows System Interrupts CPU % that caused interest. With the Atom on 1ms latency it was 10%, reducing to 5% on 2ms latency; on the i5 it was 6% and 1% respectively. So as you would expect, the Atom was straining to cope with System Interrupt activity a lot more than the i5, but both processors sounded significantly worse on 1ms latency.

Whilst I couldn't swear to it, I don't recall the same findings with MQn on 1ms latency which sounded the best on my Atom.

So Squeezelite code seemingly is causing high System Interrupt CPU activity and that is a negative impact on SQ.

Gordon may know this, understand why and may be cannot do anything about it, but I thought it worth raising for him/others to comment or try measuring for themselves on their systems.

using -a 1 b: 1000:200000 -q with default amanero driver settings getting 0.6 % cpu system interrupts.

Still like the large decode buffer.

sbgk wrote:using -a 1 b: 1000:200000 -q with default amanero driver settings getting 0.6 % cpu system interrupts.

Still like the large decode buffer.

can you go under 1000 streaming seems to help as much as the low latency