Matthew Nish-Lapidus schrieb:

I'm not sure how to resolve this.. running in -rt mode used to help, but i'm running on os x and -rt mode doesn't exist any more.

Can somebody please enlighten me what's up with -rt on OS X? I once tried it (after logging in as root) and the pd-watchdog started as expected (haven't checked the priority settings). But audio drops were still there: Pd is _unusable_ on my G3 400MHz Pismo with 640 MB Ram, OS X 10.2.6, internal sound or Motu 828 (Max/MSP works fine on the very same machine) - Maybe there are less audio drops when I run Pd wih -rt but maybe that's just what my ears expected to hear. At least the watchdog does not kill Pd even if I have 120% CPU load reported by Pd. So it seems there's something wrong with -rt on OS X or it isn't working at all....

BTW, below is what I get on my machines. I tried to load as many [osc~ 200] objects as possible with only one connected to dac~ to check the sound. Shown are the number of osc~ objects where trouble starts.

Olaf

G3 400MHz (internal audio, 44.1kHz):

• pd 0.36 w/o -rt flag:

  • 1x osc~ - dropouts when moving windows
  • 140x osc~ - constant audio dropouts that get heavier when moving the mouse

• pd 0.36 with -rt flag:

  • 1x osc~ - dropouts when moving windows
  • 340x osc~ - constant audio dropouts that get heavier when moving the mouse

SGI Octane R12k 300MHz (internal analogue IO, 44.1kHz):

• pd 0.36 w/o -rt flag:

  • 220x osc~ - audio dropout when moving windows

• pd 0.36 with -rt flag:

  • 500x osc~ - no audio drops but watchdog kills Pd when moving windows (92% CPU usage reported by top)

Pentium 4, 1.8GHz (Motu 828, 44.1kHz, asio driver):

• pd 0.36 (no -rt flag, of course):

  • 248x osc~ - audio dropouts when moving windows (about 70% CPU)

• pd 0.36 set to realtime with task manager:

  • 248x osc~ - audio dropouts when moving windows (they sound a bit different and the gui is slower...)

derek holzer schrieb:

these are some pretty interesting "benchmarks", espc in regards to the performance of your SGI! it inspired me to do some late night shopping for old SGI machines here in holland. seems you can pick up a 200 MHz Octane, or a dual 150 MHz, for something like EUR 350-400.

I got my Octane for 500,- Euro. - I just needed a machine for sound installations that I wouldn't miss at home. Trying to find a cheap solution I discovered the low prices for 'old' SGIs. Since these machines have ADAT IO build in you even safe the costs for a Hammerfall or similar card!

can you use linux, or only IRIX? if linux is runnable, which distro?

I haven't tried that. There should be at least one Linux distro out there for SGI machines, but I don't knwo anything special. IRIX works so I saw no need to install Linux. And I guess IRIX is better optimized for these machines.

how is general audio performance on this machine? i was quite amazed at the cpu speed-to-pd performance ratio.

Me too! I didn't expect to find such a great difference on the platforms I tested. I'm currently toying around with low latencies on the Octane and 5 ms seem to be quite possible.

that you use pd 0.36 over 0.37 for reasons of the Portaudio... are other problems you have noticed regarding software incompatibilities?

Soeren Bovbjerg spent some wotk on getting Pd 0.36 to run on IRIX and he also provided working binaries. That's why I started with 0.36. Sound input was not working so I fixed it and recompiled the whole stuff with GCC to avoid the MIPSpro compiler one would need to buy otherwise. I have to admit that I haven't tried 0.37. I guess it won't even compile (or will it, Miller?). The Portaudio thing just scared me!

if you are buying an older SGI, like one of the first series O2s or Octanes, how is support for more modern PCI cards, like soundcards or firewire cards?

Well, you have to buy the few cards that are supported in these machines. They do exist (soundcards at least, not shure about firewire), but expect to pay quite a lot of money.

last question: are SGIs more efficient with RAM? it seems like all the machines i saw were pretty low in this dept [maybe that's why there are cheap!]

SGIs don't use a bus system where all devices have to share one physical connection. They instead use a 'crossbar switch' with 1.6Gbit/s throughput. That is, different parts can exchange data directly without blocking a bus system. I guess thats the secret for the speed. You just can't compare them with 'ordinary' PC architectures.

Regarding my 'benchmarks', a test on a PC running Linux would be nice. Attached is a patch using clone (http://www.akustische-kunst.org/puredata/clone/) to test your machine. Just increase the number of abstractions and hit the bang. Listen to the sound and remember the number of abstractions when the sound get's ugly. That's what I did on all my machines to compare the performance.

Olaf

#N canvas 435 76 667 576 12; #X floatatom 61 216 5 0 0 0 - - -; #X obj 241 513 dac~; #X obj 42 443 clone test-abs 2; #X obj 241 485 osc~ 200; #X text 319 485 reference sound to listen to; #X text 111 216 number of osc~ objects to load; #X obj 42 298 int; #X obj 71 252 bng 15 250 50 0 empty empty empty 0 -6 0 8 -262144 -1 -1; #X text 98 252 hit bang to load; #X msg 42 373 reload test-abs $1; #X connect 0 0 6 1; #X connect 3 0 1 0; #X connect 3 0 1 1; #X connect 6 0 9 0; #X connect 7 0 6 0; #X connect 9 0 2 0;

#N canvas 0 0 454 304 12; #X obj 59 91 osc~ 200;