White Papers

13 Deep Learning Inferencing with Mipsology using Xilinx ALVEO™ on Dell EMC Infrastructure
Table I summarizes the accuracy achieved by Zebra based on int8 computations (obtained from
FP32 training and using Zebra quantization) compared to the accuracy obtained by a GPU/CPU
platform based on FP32 computations. The results were obtained with the same networks and
the same application, without modifications.
Table I. Summary of accuracy of results obtained in the evaluation
Conclusions
The purpose of the evaluation of PowerEdge R740/R740xd Dell servers hosting a Xilinx
ALVEO™ U200 boards configured with Mipsology Zebra was two-fold:
• Estimate the throughput of the setup in accelerating a set of images from
ImageNet without impact on the accuracy of the computation.
• Establish the ease-of-use of Zebra deployment.
Both objectives have been met successfully. On the one hand, the setup produces an
acceleration platform for neural network inference. On the other hand, Zebra does not require
changes to the neural network, does not need re-training of the NN, and it operates within the
popular frameworks without any extra work. Zebra conceals the FPGAs, paving the path for
data scientists and engineers to use their knowledge, talent, and expertise without spending
efforts necessary with alternative products.
References
➢ Mipsology - https://mipsology.com/
➢ Dell PowerEdge R740/R740xd - https://www.dell.com/en-us/work/shop/povw/poweredge-
r740
➢ Dell EMC Accelerator site: https://www.dellemc.com/en-us/servers/server-accelerators.htm