文档介绍:FROMOPENCLDEVELOPER’SPERSPECTIVEHAIBOXIESEP5,2013NextgenerationAPU,forOpenCLperformanceWE’REINTHE“ARCHITECTURALINTEGRATION”STAGEPhysicalOptimizedArchitecturalSystemIntegrationPlatformsIntegrationIntegrationIntegrateCPU&puteGPUSpaceforCPUC++supportcontextswitchinsiliconandGPUGPUusesUnifiedMemoryUsermodepageablesystemGPUgraphicsControllerschedulingmemoryviaCPUpre-emptionpointersBi-monFullycoherentPowerMgmtManufacturingmemorybetweenbetweenCPUQualityofServiceTechnologyCPU&GPUandGPU3|GETREADYFORNEXTGENERATIONAPU|SEPTEMBER12,2013|PLIANTFEATURESOptimizedPlatformsSupportOpenCLC++directionsandMicrosoft’ingC++++,requiringnoOSkerneltransitionsUsermodeschedulingorservices,minimizingsoftwaredriveroverheadBi-DirectionalPowerEnables“powersloshing”whereCPUandGPUareabletodynamicallylowerorMgmtbetweenCPUraisetheirpowerandperformance,|GETREADYFORNEXTGENERATIONAPU|SEPTEMBER12,2013|,,,datahadtobecopiedbetweenthetwospacesorpage-,NOGPUmemorysizelimitation!AllowsfordatatobecachedbyboththeCPUandtheGPU,,mandbetweenCPU&,|GETREADYFORNEXTGENERATIONAPU|SEPTEMBER12,2013|FULLHSAFEATURESSystemIntegrationGPUtaskscanbecontextswitched,makingtheGPUamulti-,puteswitchinteroperation.