First time here? Check out the FAQ!

Would a Pacarana with extended memory be able to handle more cross-filter instances ?

0 votes
Generally i find the crossfilter processing demands to be probably the only performance problem i ever had with Pacarana (this and the strict dichotomy between the two prototypes short and long).

I can't really tell why but comparing this with an external convolution suite in max i've been working with the last couple of years (with a modest notebook) the difference in instances and their lengths i can use is quite striking..

Do you see this changing in the future, i mean could it be a code thing?


p.s i didn't want this to be a separate thread somehow i thought i was answering in the previous one..
asked Sep 22, 2016 in Using Kyma by greg-grigoropoulos (Practitioner) (810 points)
edited Sep 23, 2016 by greg-grigoropoulos

1 Answer

+1 vote
Some unique attributes of CrossFilter require additional computational resources — for example, the option to capture a new Response while the Sound is playing, the low latency, the option to use different parts of the captured Response while the Sound is playing.  For the most part, the CrossFilter is constrained by computational requirements, so adding more RAM would not allow you to schedule more CrossFilters than you can already play in real time.

We will add some "degenerate" cases of the CrossFilter requiring less computation to our wishlist; a CrossFilter with a precomputed, unchanging Response file and long initial latency would require less real-time computation and would allow for longer responses and more CFs in parallel (assuming you did not want to take advantage of the unique, dynamic attributes of the CF).
answered Sep 22, 2016 by ssc (Savant) (127,140 points)
That would be great!  I find there are circumstances where having more instances is beneficial...For example at some point for my thesis i had programmed a matrix of convolutions which i used for spatializing monophonic events (each speaker had 4 responses to move between)..I ended up using 32 responses for a typical 8channel system, something i wasn't able to do here. Anyway thanks for clearing things up!

Since you mentioned it though... Is there some sort of smoothing happening when scanning through the response buffer? If i were to put a number of responses on after the other and select them realtime would it as if interpolating between different cf's with the  same input?