This content has been marked as final. Show 2 replies
When I use good old env variable (that supposed to do absolutely nothing in SDK 2.2, right ?) GPU_MAX_HEAP_SIZE and set it to 512 GPUinfo-like code starts to report 512MB available instead of 256 (Win7x64+Cat 10.10+SDK 2.2).
But app performance still unacceptable. Still ~10 times longer kernel calls.
Should I degrade to prev catalyst ? x64 platforms not supported with HD4xxx GPUs ? Or what? Such poor performance makes GPU completely unusable in this software configuration.
Do you face this issue with openCL samples also or with your app only?