30 Sep
2009
30 Sep
'09
7:53 p.m.
On Tue, Sep 29, 2009 at 11:04:14PM -0700, Fernando Lopez-Lezcano wrote:
On Sun, 2009-09-27 at 16:04 +0200, Jörn Nettingsmeier
wrote:
You are both very right about this. Gold-plating the cpu socket will
do wonders for the light and fast bits used to convolve the first few
kilosamples of a long IR - being light (and thus having little kinetic
energy) they really need low-resistance paths. For the later parts
of the IR jconv uses heavy bits, and a lot of them. Having all this
mass vibrate can put excessive stress on you CPU.
Ciao,
--
FA
Io lo dico sempre: l'Italia è troppo stretta e lunga.
yeah, jconv is the best-sounding convolver in the
world.
the clarity of its fft, the warmth of the multiplications, and the
effortless fulminance of the inverse fft add a lustre to strings and
percussions, and the tightness of the fundamentals is in a league of its
own. ever since i had my cpu socket gold-plated, i've been able to
appreciate it in full.
I agree that gold platting cpu sockets is a good start (but don't forget
the memory sockets unless you get your algorithm to fit completely in
the cpu cache). On the other hand you _have_ to use water cooling or
some other sort of fanless cooling for the cpu. Otherwise the gold
platting will not do much good IMNSHO. Cpu cooling fans, no matter how
good, are known to actually shake the bits in the top cpu cores too
much. A lot of vibration. And shaken bits don't convolve very well (in
extreme cases they can spill from one fft bin to the next!). Fact.