[ODE] CULLIDE vs OPCODE

[Sam Hale]

At 11:42 PM 8/19/2004, you wrote:
>I recently ran a number of Vtune profiles on some fairly massive rigid 
>body simulations.  I found it interesting that most of the CPU is spent 
>dealing effectively with data management issues, not so much actual 
>floating point physics type stuff.  Once "N" becomes very large, just 
>shuffling all that data around and organizing it into appropriate data 
>structures quickly sucks the bulk of the CPU.  I wonder how this reflects 
>on the topic of GPU or dual-CPU systems.

Ever considdered dedicated hardware to shuffle data around, ie- ASIC for a 
binary sort?(j/j)

Efficient data structures are a topic of great study and debate, no doubt 
about it. And some things just can't be done in parallel as quickly. 
Communication between CPUs is a big issue. Sending the data to the second 
CPU might take longer than the processing it's subjected to once there. 
Shared dual access RAM helps, but I haven't seen any in desktops yet.

All depends on what it is you're doing, I guess.

-Sam H