31 Oct
2016
31 Oct
'16
9:56 p.m.
On Mon, October 31, 2016 2:18 pm, termtech wrote:
Recall: My Delta1010-LT card was not OK in my new
PC...
...
This time however, I noticed something: When using the
card in an 'effect'
situation, where audio input is passed through to an audio output, either
directly or with some effect, horrible noise would appear after 15
minutes and then... slowly drift into working again for 15 minutes,
and then slowly drift into noise again for 15 minutes, then
working again for 15 minutes and so on.
This was a HUGE clue. This slow drifting in and out of noise is familiar.
It is instantly recognizable to me (as should you) as a synchronization
problem, not a PCI voltage or bus problem.
This is with a single 1010LT card, correct?
I think that clock sync problems should not be possible with a properly
configured single card.
Can you show the clock configuration for your card some way? Maybe screen
capture of envy24control? I think that there should be options for
internal clock, S/PDIF recovered clock, or word clock input for the
output. If you are using external S/PDIF connected devices you will
probably need to verify that there is one device set to use internal
clock, and all other devices synchronized to the received S/PDIF clock.
So... on this replacement PC there is a setting in the
BIOS for the number
of CPU cores to activate. The cheaper original new PC had NO such
setting.
Physical cores, or virtual cores? Virtual cores would usually be called
simultaneous multi-threading (SMT, generic term) or hyper-threading (Intel
trademark). Enabling hyper-threading can increase latency, disabling and
measuring change in worst case latency is usually one of the things to
check when tuning a machine for low latency.
I had even researched these CPUs and had read that it
is possible that
the CPU Clock signals can be independent for each core, thus causing
some synchronization problems.
There are multiple things that can accurately be described as a "clock" in
modern CPU's, some of which change with core speed if using power savings
modes which modify the CPU frequency. Competently designed modern
software would not use those timers and would use a core frequency
invariant timer.
At this point I'm not sure if this a clock signal
sync problem or a
timer/counter problem. I had also read that 'local' timers/counters
can be a problem with multi-core CPUs - that time is not quite the
same in each core.
That is correct, you can research how that is handled in modern CPU's when
software needs accurate timekeeping. It is basically a solved problem,
although using inappropriate timers could be a problem if using software
which was developed many years ago.
IIRC Someone told me that a quad core is two chips
each with two cores ??
That tended to be true in 2005, not so much these days. Most processors
with up to 12 cores, sometimes more, are single die now. Those which are
not do not tend to show up in desktop computers.
Can anyone shed some more light on these multi-core
problems?
Comments welcome, please!
I personally have not experienced any problems with multi-core processors.
When you enable all four cores, what is in the contents of /proc/cpuinfo?
First impression is that this is a problem with the card synchronization
settings, but I have no idea how changing the CPU configuration could
affect that.
What software is running? Is this system using jackd, or direct ALSA
access, or something else?
--
Chris Caudle