Stereo Tool

Then the clicking gets less and less when turning more and more steps off. Then i can also lower the Buffer size to 1.4ms without clicks and also no clicks when i hit RESTART.

I've been meaning to ask Hans about this for a while, so this is as good an opportunity as any...since he's missing my question to him again...

The reason I asked about what processor he's using, since he can't replicate the issue, and you're having the issue on an AMD system, but not your Intel system, is that back in the day, using Intel compilers or various Intel libraries (such as Math Kernel Library or Performance Primitives), the processor dispatch function, the part that looks at the processor to determine which code to execute, ran a check for the presense of "GenuineIntel" in the processor identification string. If it found GenuineIntel, it would run the optimal codepath, otherwise it would take the slower codepath.

I've personally seen evidence of this kind of thing happening, which was again with a distributed computing project.

So, Hans, what processor do you have, and are you including any Intel libraries and/or using any Intel compilers, or are you using GCC, Visual Studio, or something other than Intel to compile?

Might be a wild goose chase here, but I asked for a reason

I suppose that's a single core system? In that case, if the ASIO thread priority is lower or equal to the processing thread, you'll probably get horrible results. What I don't get is why the ASIO thread priority isn't higher.

--> Will create a test version where I'm attempting to boost the ASIO thread priority. (Will put it IN THIS POST in a few minutes).

Edit: Here it is:
- ASIO thread priority now at TIME_CRITICAL.
- Two other threads now at HIGHEST (just below TIME_CRITICAL).
http://www.stereotool.com/download/radi ... itical.exe

This version could do anything - upto causing crashes of Windows etc - since I'm making changes to the priority of a thread that was created and is managed by a driver. Anyway, the ASIO stuff here is higher priority now than the processing, which should work better...

Hm, then again it looks like the ASIO thread doesn't get enough processing time to finish in time. The version I just built for Bojcha might work better... (or not)

Testing with the time critical version i get the same things as the previous versions so no differences for me.

Sorry!
- Intel Q9450 (quad core @ 2.66 GHz)
- Yes, Intel compiler + IPP
- Still, a higher CPU load should only affect the minimum possible buffer size, and should not cause clicking etc. That is assuming that the processing thread gets a LOWER priority than the ASIO thread, which is called far more often and hence has far less time to finish.

It's Dual Core AMD (X2) but i think that's not reason for "clicks". because..
BA Asio works on all PCs i found, My Intel c2d, my laptop intel c2d, AMD x2 3.3GGHz...
Maybe Brian is right..
BA Asio tests CPU for the first start, so maybe it choose what library it will use.

Where I was wanting to go with this was asking if you could compile in GCC and with the AMD Core Math Library instead of Intel compiler and IPP, due to lingering possibilities that AMD systems are being given a sub-optimal codepath to follow.

I don't know how much ACML costs. If they have an evaluation version, as long as you limit yourself to not making any money from the compilation under the eval version, you might could have Bojcha (or someone else) test it to see if it is better... From there, you might could branch out to Intel versions and AMD versions, after purchasing anything that's necessary...

At any rate, you're using software that is known to have the potential to cause sub-optimal performance on AMD systems, and could be why I've had significantly increasing processor loads from StereoTool for a while now...

The clicks i'm having or on a Intel Pentium Dual Core E6500 @ 2.93 GHz. BA is for example also optimized for intel processors. But maybe it's possible to create a AMD optimized version.

you posted just before I did this...

Could be that there's an underlying issue that would impact both processor types and it gets more pronounced on an AMD system. I'm not sure. I just know that this is a potential issue anytime you use Intel compilers (usually C++ or FORTRAN) and / or IPP and / or MKL.

Allegedly they were supposed to not include performance limiting techniques per a November 2009 settlement with AMD, but the US Federal Trade Commision still has an open suit against Intel in regards to various monopolistic practices.