PIO on Current HW Outperforms DMA Over a Range of Payload Sizes In A Number of Different Applications (ETH Zurich)


A new technical paper titled "Rethinking Programmed I/O for Fast Devices, Cheap Cores, and Coherent Interconnects" was published by researchers at ETH Zurich. Abstract: "Conventional wisdom holds that an efficient interface between an OS running on a CPU and a high-bandwidth I/O device should be based on Direct Memory Access (DMA), descriptor rings, and interrupts: DMA offloads transfers fr... » read more

Accelerating Simulation Of PCIe Controllers For DMA Applications


For memory-intensive and high-performance computing, direct memory access (DMA) is indispensable. A typical DMA operation in PCI Express (PCIe) entails the transfer of data from the system memory to end point devices using a point-to-point PCIe bus to reduce latency and increase memory access throughput between the CPU and the device. Verification of DMA engines is concentrated on the data t... » read more

Machine Learning Inferencing At The Edge


Ian Bratt, fellow in Arm's machine learning group, talks about why machine learning inferencing at the edge is so difficult, what are the tradeoffs, how to optimize data movement, how to accelerate that movement, and how it differs from developing other types of processors. » read more