On 10/6/2014 8:28 AM, Mark Tinberg wrote: > I don't think that the shared memory integrated cards work that way, the video output has dedicated frame buffer memory so that every screen refresh doesn't go over the main memory bus, what goes over the main bus are all of the graphical assets and commands used to compose the final image, which often need to be shared with the main CPU anyway. >yes, it all goes over the same memory bus, but that bus is so insanely >fast now, that its only using like 1-2% of the total bandwidth to >maintain a medium-high resolution laptop screen, such as 1920x1080 at >60Hz refresh and 24 bit/pixel. The Intel integrated graphics adapters >are actually integrated into the CPU chip itself, along with the memory >controller. Yeah, I'm aware that they are part of the CPU die but I was splitting some hair about the output being buffered on the video output port (HDMI/VGA) so that if the image doesn't change then the video output port doesn't even hit main memory, it can just resend it's buffer. Maybe I'm remembering how it worked when the GPU was integrated with the motherboard chipset and memory controller, which had a couple of megabytes of video ram, before those features were pulled into the CPU itself. It looks like also on the higher specced Haswell chips that there is a new 128MB eDRAM L4 cache shared between the GPU and CPU execution units which also greatly reduces the memory bandwidth needs — Mark Tinberg mark.tinberg at wisc.edu