 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| • |
Many system
memory graphics data structures
|
|
|
benefit from
being created as USWC
|
|
|
|
– |
large data
structures that are written once by the CPU and then
|
|
|
read only by the
GPU unnecessarily thrash cacheable memory
|
|
|
|
– |
Examples include
Push Buffers, Texture Maps, Vertex Arrays
|
|
|
|
– |
special cases
benefit from being in cacheable memory
|
|
|
| • |
Allocate USWC
buffers to permit the GPU to optimize
|
|
requests by
setting the No Snoop attribute
|
|
|
|
– |
chipset can avoid
a snoop cycle which reduces FSB traffic
|
|
|
|
– |
lowers latency by
not needing to wait for a snoop result
|
|
|
| • |
Allocate
contiguous USWC buffers if possible
|
|
