White Papers

6 Deep Learning Inferencing with Mipsology using Xilinx ALVEO™ on Dell EMC Infrastructure
Figure 4. Xilinx ALVEO™ U200 data center accelerator card Figure 5. U200 Block
Diagram.
Zebra Acceleration Stack from Mipsology
Mipsology, an innovative AI/DL startup, conceived a software stack called Zebra for ultra-fast
inference acceleration of CNN. Based on FPGAs, Zebra sits on top of the FPGAs and conceals
them to the user.
The Mipsology Zebra stack outshines its competitors with several advantages: exceptional
throughput, very low latency, very high efficiency, and remarkable ease-of-use.
The following lists the main Zebra’s capabilities:
Zebra Delivers High Performance, Low Latency and High Efficiency
Taking advantage of the rich set of FPGA resources and executing them at high frequency
(650MHz), Mipsology implemented Zebra to deliver high throughput on all CNN. Several
computational optimizations within Zebra push its efficiency to a level not achievable with
conventional hardware accelerators. The typical pruning required by other FPGA solutions is not
useful with Zebra to enhance performance, avoiding re-training of the NN.
The FPGA architecture also allows Zebra to deliver low latency for application requiring real
time response.
Further, Mipsology releases periodic software updates to improve performance. Ultimately, the
user can preserve the hardware investment and deliver more throughput by a simple software
upgrade.
Zebra Accelerates Inference Calculation of Any Neural Network
Zebra accelerates a large set of the layers commonly used in CNNs. No change is required to
the neural network or its training to run on Zebra. This gives Zebra the ability to supports a wide
range of CNNs, from the most popular neural networks found on the internet to custom
designed CNNs for commercial use. All the layers are accelerated within the FPGA to maximize