I have an intel 8259U CPU with an iris plus 655 iGPU which has 128 MB of on-die cache (eDRAM) that can be shared between CPU & iGPU.
Now as far as understand, the main overhead of copying EFB to RAM in addition to GPU is the transfer overhead on the PCIx bus. But since CPU + iGPU are in the same chip , sharing the memory, is this overhead effectively negated in the case of hardware like mine? How about the XFB copies?
I'm running on Vulkan, with driver support for 1.2.148 of Vulkan API.
Couldn't tell much difference in the game I tested (NFS: Underground) but am curious what the devs think on this.
Now as far as understand, the main overhead of copying EFB to RAM in addition to GPU is the transfer overhead on the PCIx bus. But since CPU + iGPU are in the same chip , sharing the memory, is this overhead effectively negated in the case of hardware like mine? How about the XFB copies?
I'm running on Vulkan, with driver support for 1.2.148 of Vulkan API.
Couldn't tell much difference in the game I tested (NFS: Underground) but am curious what the devs think on this.