Re: Multi-core and memory

As higher-performance display adapters contain a reasonable amount of
local high-performance memory, how will multi-core CPU/GPU solutions
deal with memory bandwidth constraints? Understanding that contem****ary
GPUs utilize caching, how is the best balance for performance achieved -
with large local caches in the multi-core part for rendering activities,
or is increasing main memory throughput the solution? As visual realism
increases, we can anticipate the model complexity and associated files
for rendering will become ever more complex and memory intensive. Will
a local cache provide enough capacity for these intensive data sets?




-- 
earth