White Papers
9 Real Time Streaming Analytics with Megh Computing on Dell EMC PowerEdge Servers
Figure 7. Dell EMC PowerEdge R740/R740xd.
The Dell EMC PowerEdge R740/R740xd is Dell EMC’s latest two-socket, 2U rack server designed to run
complex workloads using highly scalable memory, I/O capacity, and network options. The R740/R740xd
features the Intel Xeon processor scalable family, up to 24 DIMMs, PCI Express (PCIe) 3.0 enabled
expansion slots, and a choice of network interface technologies to cover NIC and rNDC. In addition to
the R740’s capabilities, the R740xd adds unparalleled storage capacity options, making it well-suited for
data intensive applications that require greater storage, while not sacrificing I/O performance.
Results
[*1]
The performance of VAS is summarized in the table below for two configurations:
1. CPU Only: where the complete pipeline is implemented in software. As expected, the
performance varies based on the Xeon processor used. The throughput for Resnet50 based
inference is 30 fps for a dual socket Xeon Silver server and 150 fps for a dual socket Xeon
Platinum processor.
2. CPU+FPGA: where the real time analytics pipeline is implemented in the FPGA. We used two
Intel PACs : the first one to ingest the data and do the video decoding and the second one to do
the deep learning inferencing.
8 Channels
CPU
CPU + FPGA
CPU
2x Xeon E5 8280 Platinum
Xeon E5 4214 Silver
FPGA
2x Arria 10
Power
2x 400W =
800W
250W
(150W + 2x 50W)