White Papers

8 Deep Learning Inferencing with Mipsology using Xilinx ALVEO™ on Dell EMC Infrastructure
Figure 6. Mipsology’ s Zebra stack
Zebra Is Easy to Use
Deploying Zebra is a “plug play” process. Plug an ALVEO™ Board into a PC running Linux,
issue one single Linux command to configure it, and you are ready to go. See figure 6.
There is no R&D cost in using Zebra. No extra work is required to make Zebra compute a neural
network nor is any proprietary tool needed to understand how to migrate the neural network.
Zebra Uses 8-bit or 16-bit Fixed-point Math and Optimizes Calculation Precision
Zebra executes inference calculations using 8-bit or 16-bit fixed-point integers. Mipsology has
devised a quantization algorithm that converts floating points training parameters into fixed
points parameters in minutes maintaining similar accuracy achieved in training. The quantization
in Zebra does not require users specific effort nor does it require to retrain the network for lower
precision.
Zebra Scales in Size and Power
Zebra runs one or many FPGA boards of any size, scaling from supporting massive computing
in datacenters to limited computing at the edge or in embedded applications.
Zebra Supports Multiple Users and Multiple Neural Networks on the Same FPGA Board
The unique architecture of Zebra supports multiple users and multiple networks, simultaneously,
on the same ALVEO™ board. The resource sharing is selected when using the board,
maximizing the investment in hardware to avoid under-utilized computing resources. Each
processing session owns a “full stack”, making call through its framework, and executing its own
neural network with its own weights.
Zebra Does Not Require Knowledge of FPGA Technology
Zebra removes the burden to learn the FPGA technology. It does not require knowledge of any
hardware design language, new design tools, or to understand hardware-level details. Delivered
with pre-compiled FPGA binary files, it removes the need to learn FPGA coding and
compilation.
Zebra Boasts the Lowest Cost-of-Ownership
The collective characteristics of Zebra make it the accelerator with the lowest cost-of-ownership
(COO).
The ease-of-use, scalability, inherent lower power consumption and the ability to support large
number of neural networks make the Zebra stack running on Xilinx FPGA a pretty good
proposition.