When different neural networks run on the DPUCZDX8G, the I/O bandwidth requirement will vary depending on which neural network is currently being executed. Even the I/O bandwidth requirement of different layers in a single neural network will be different. The I/O bandwidth requirements for some neural networks, averaged by layer, have been captured with one DPUCZDX8G core running at full speed. The peak and average I/O bandwidth requirements of three different neural networks are shown in the table below. The data provided is for two commonly used DPUCZDX8G architectures (B1152 and B4096).
Note: When multiple DPUCZDX8G cores run in parallel, each core might not be able to achieve optimum performance due to I/O bandwidth limitations.
|Peak (MB/s)||Average (MB/s)||Peak (MB/s)||Average (MB/s)|
|SSD ADAS VEHICLE||1516||684||5724||2049|
If one DPUCZDX8G core needs to run at full speed, the peak I/O bandwidth requirement shall be met. These bandwidth requirements are primarily associated access to data though the AXI master interfaces (DPU0_M_AXI_DATA0 and DPU0_M_AXI_DATA1).