19 Mar
2005
19 Mar
'05
10:14 p.m.
On Sat, 2005-03-19 at 03:45, bjay(a)SAFe-mail.net wrote:
Hi, just spent a month getting a linux audio
workstation going and I've got a load
newbie questions I have'nt been able to solve, or find answers to; some technical,
some bigger picture :
Machine Specs
-------------
Uniwill 340s2 (Advent 5480DVD) laptop
384Mb RAM
PIII 800Mhz (8 speed steps)
SIS 7018 Integrated sound chip
Chipset mostly SIS I guess.
40Gb HD (Fujitsu, recent model)
(Yea I know it's a piece of junk but I could write a heafty tome about my money
situation at the moment)
Software
--------
Win2K NTFS 20Gb
NTFS 10Gb
Fedora Core 2 + CCRMA (everything from rpms), CCRMA Low latency kernal latest stable
(2.6) ext2 10Gb
Rosegarden4-1, built from source
Hokay,
1) I boot up the low latency kernal everything is fine until I run a midi app.
The machine has no physical ports but presents an input port and 5 output ports
in Qjackctl (SIS7018,trident driver). I think the first is raw, then Midi0-3.
The clearest repeatable falut I can give is when I run Qjackctl, with jack
either running or not running and try to connect to one of the SIS midi ports,
the machine freezes when I connect (Caps lock does nothing).
Probably an alsa driver problem for your particular card. Check to see
(just in case) if there is anything related to the hang in
/var/log/messages.
3) Should I be building all my own software for i686 ?
Most of the CCRMA
stuff is i386 I can't remember at which point the 'big change' in intel
architecture happend (was it 286 got protected mode or somthing). Anyone
give me a ball park performace improvment percentage for rebuilding the
music software ?
Probably 0% in most cases. While the binaries in Planet CCRMA are
compiled using the instruction set of the i386, the instruction
_ordering_ is optimized for the i686 architecture (same as the rest of
Fedora Core packages). In modern Intel processors the lack of the i686
instructions does not really impact performance. The ordering of
instructions, on the other hand, does have an impact on Intel
processors. AMD processors do their own instruction reordering so that
that does not matter either (in practical terms, I don't know if maybe
things would be 1% faster, which would not matter in the real world,
IMHO).
I do build some things i386/i686 but I have never been able to measure
any significant difference in speed (not that I have tried very hard).
This topic resurfaces every once in a while in the Fedora lists and the
answer from the gurus is always "give us feedback with performance
numbers and the improvement" (from compiling a given pacakge i686 _with
everything else the same_) and so far I have seen no numbers (that I
remember).
-- Fernando