WHITE PAPER
Tackling workloads in real time with PCIe Gen 4.0
mrcy.com
GPUS ALREADY NEED MORE BANDWIDTH
Nowhere is the need for PCIe Gen 4.0 more apparent
than in systems implementing Graphics Processing Units
(GPUs). A GPU is a chip with thousands of processing
cores that work in parallel, executing identical math
operations on segmented portions of a data stream.
Originally designed to render complex images quickly, the
massive parallelism in GPUs makes them extremely efficient
at processing very large datasets across a range of real-
time applications. (The label 'General Purpose GPU' (GPGPU)
is sometimes used when they are applied to non-graphics
functions, but the distinction is no longer really important.)
Successive generations of GPUs have achieved
exponential performance growth, with steady
increases in clock speeds and tremendous increases
in the number of processing cores. More recently,
performance was boosted by faster access to data.
About Mercury
Mercury Systems (Nasdaq: MRCY) is a leading technology company serving the aerospace and defense
industry, positioned at the intersection of high tech and defense. Headquartered in Andover, MA, we
deliver solutions that power a broad range of aerospace and defense programs, optimized for mission
success in some of the most challenging and demanding environments. We envision, create and deliver
innovative technology solutions purpose-built to meet our customers' most-pressing high-tech needs.
mrcy.com
The Mercury Systems logo and the following are trademarks or registered trademarks of Mercury Systems, Inc.: Mercury Systems, Innovation That
Matters, and BuiltSECURE. Other marks used herein may be trademarks or registered trademarks of their respective holders. Mercury believes
this information is accurate as of its publication date and is not responsible for any inadvertent errors. The information contained herein is subject
to change without notice.
© 2021 Mercury Systems, Inc. 8052.00E-0421-wp-PCIe_Gen4 White Paper
MADE IN USA
Learn how we can improve your
application efficiency.
Contact our server experts today:
servers@mrcy.com
Corporate Headquarters
50 Minuteman Road
Andover, MA 01810 USA
+1 978.967.1401 tel
+1 866.627.6951 tel
+1 978.256.3599 fax
International Headquarters
Mercury International
Avenue Eugène-Lance, 38
PO Box 584
CH-1212 Grand-Lancy 1
Geneva, Switzerland
+41 22 884 51 00 tel
GPUs typically deal with huge data sets, which can't fit fully
into system memory. Continual data movement from I/O
sources to memory used to require direction, or control,
from a general-purpose CPU. The GPU operated as an
adjunct processor in the system architecture, with data
movement often a gating factor for application throughput.
Now GPUs can autonomously access data from the network
using Remote Direct Memory Access (RDMA). Components
with this capability are sometimes referred to as 'data
center GPUs'. RDMA over Converged Ethernet (RoCE) is
a standard protocol already used in some systems.
For these types of systems, any PCIe Gen 3.0 interconnect
between processors, or between processors and storage,
is a serious performance bottleneck, making Gen 4.0 a
necessity. Even PCIe Gen 4.0 will have limitations when 400
GigE networks become a reality. Designs must be ready to
evolve with new generations of ever-faster interconnects.