2. 2006 2007 2008 2009 2010 2011 2012
40
35
30
25
20
15
10
5
NVIDIA
Acceleration
31 petaFLOPS
Total
Acceleration
37 petaFLOPS
GPU‑Accelerated
computinghas become an important catalyst
in the advancement of science and
technology—enabling tremendous
breakthroughs by simply enabling us
to do more, faster.
The need to solve complex
computational problems is becoming
increasingly commonplace. And
GPU-powered accelerated computing
is meeting this need by delivering
an order of magnitude more
performance, more efficiently.
Accelerator
Performance in
Supercomputing
Performance delivered by
GPUs in supercomputers
has increased from 0 to
19% in five years and is
growing exponentially.
Source: Top500.org.
3. Developing with GPUs is now pervasive in computing:
430,000,000+CUDA GPUs
have been shipped
35,000+papers have been published
on CUDA
8,000+institutions have registered
CUDA developers
580+universities are teaching
GPU programming
200+major applications are now
GPU accelerated
50+of the world’s fastest
supercomputers are
powered by CUDA GPUs
The benefits of GPU‑accelerated
computing are more widely recognized
every day. This is leading to more
industry leaders and members of the
research community to adopt both
NVIDIA GPUs and CUDA
®
,
the world’s most pervasive
parallel‑computing platform
and programming model.
That’s more than
two shipped
every second!
4. Driven by the broad use of NVIDIA
visual computing technology, the GPU
Technology Conference (GTC) has
become the world’s most important
event for GPU developers.
GTCis where art
meets sciencemeets
engineering
meets business.
5. Air Force
Research Lab
Barcelona
Supercomputing
Center
BGI
Carnegie Mellon
Chinese Academy
of Sciences
CERN
ETRI Korea
Expressions
College for Digital
Arts
Fermi National
Accelerator Lab
Georgia Tech
Harvard
IIT Italy
JPL
Johns Hopkins
Julich
Supercomputing
Center
KAUST
Los Alamos
National Lab
MIT Lincoln Lab
NASA
NATO
Naval Air Warfare
Center
NOAA
Oak Ridge
National Lab
Russian Academy
of Sciences
Sandia National
Lab
Stanford
Swiss
Supercomputing
Center
Tokyo Tech
Audi
Amazon
Apple
Bank of America
BMW
Boeing
Carl Zeiss
Chevron
Chrysler
Deutsche Telecom
ESPN
E*Trade
Exxon Mobile
Fiat
Ford
Google
Goldman Sachs
Gulfstream
Aerospace
Harley Davidson
Honda Research
iRobot
LEGO
NASDAQ
Netflix
Nike
Pixar
Space-X
Tesla Motors
Walt Disney
Academic and Research Industry
Every year, GTC provides an important
venue for exploring the game-
changing impact GPUs deliver to
science, technology, and industry. And
the 2013 event was one of the most
impressive ever.
The four‑day conference featured
talks by industry and NVIDIA experts
on GPU-accelerated computing—from
problem solving in everything from
medicine to product design and big
data, to hands-on sessions on how
to take advantage of this disruptive
technology.
More than 3,000Attendees from over
50countries
425Conference
Sessions
150Research
Posters
Here’s a sampling of
organizations that
attended GTC 2013:
6. GPUs are already
finding their
way into systems
and applications that
were undreamed of a
decade ago. Soon, mobile,
desktop, and supercomputer
technologies will intersect in
powerful and surprising ways.
Accelerating
The Future
7. GTC 2013 also featured the unveiling of several
technologies that will impact the future of
computing. This included the introduction of the
NVIDIA GRID
™
VCA,
the industry’s first Visual Computing Appliance.
The GRID VCA enables businesses to deliver ultra-
fast GPU performance to any Windows, Linux, or
Mac client device on their network while providing
the same rich graphic user experience as a
dedicated desktop PC or workstation.
GRID VCA provides a
workstation-quality experience
on any PC, Mac, or Linux device
and runs applications for power
users such as those from
Adobe, Autodesk, and Dassault
Systems. It is a turnkey
appliance that is easy to install
and manage.
8. At GTC 2013, NVIDIA unveiled its
GPU architecture roadmap,
demonstrating that NVIDIA continues to focus on
dramatically improving performance while also improving
energy efficiency. Today, its Kepler™
architecture is at the
heart of the fastest, most energy-efficient accelerators,
including those that power the world’s fastest
supercomputer for open science. The next-generation
Maxwell architecture will deliver unified virtual memory,
increasing the performance and shortening development
time for application developers. The Volta architecture
will follow, which is designed to solve one of the biggest
challenges of computing today—memory bandwidth.
2008 2010 2012 2014
Tesla
CUDA
Fermi
Full Double Precision
Kepler
Dynamic Parallelism
Maxwell
Unified Virtual Memory
Volta
Stacked DRAM
32
16
8
4
2
1
0.5
DoublePrecisionGigaFLOPSper Watt
The next-generation Maxwell
architecture will deliver unified
virtual memory, followed by
the Volta architecture that’s
designed to achieve 1 TB/sec of
memory bandwidth, equivalent
to transferring the entire
contents of a Blue-ray DVD in
1/50th
of a second.
9. 2011 2012 2013 2014 2015
100
10
1
RelativePerformance
Tegra 2
First Dual A9
Tegra 3
First Quad A9
First Power‑Saver Core
Tegra 4
First LTE SDR Modem
Computational Camera
Logan
Kepler-Based GPU
CUDA
OpenGL 4.3
Parker
Denver-Based CPU
Maxwell-Based GPU
FinFET
The NVIDIA Tegra
®
Mobile
processor roadmap
was also unveiled at the conference. It demonstrated that
NVIDIA’s investment in computing is everywhere—not
just in PCs and datacenters, but also in cars, phones,
tablets, gaming portables, and anything with a display.
For example, Tegra 4 leverages the CPU, GPU, and ISP to
deliver advanced computational photography features like
real-time HDR and intelligent object tracking. Tegra 4 will
be followed by Project Logan, which pairs ARM®
-based
mobile processor cores with our Kepler GPUs. This will be
followed by Project Parker, which will join the new 64‑bit,
ARM-compatible CPU cores with our next-generation
Maxwell GPU architecture.
Tegra 4 will be followed
by Logan, bringing
technologies currently found
in high‑performance PCs
and workstations to mobile
devices.
With Parker, a server‑class
CPU will be combined with
our Maxwell GPU’s unified
virtual memory and advanced
performance per watt.
10. To address the industry’s requirement for developing
applications for low-power architectures, NVIDIA also
introduced the Kaylaplatform
—the ARM development platform for mobile computing
and HPC applications. It’s designed to deliver the highest
performance and efficiency for the widest range of next-
generation ARM-based OpenGL and CUDA applications
by combining a Tegra quad-core ARM processor with a
Kepler-based GPU. This gives developers a great way to take
advantage of the next-gen Tegra SoCs based on the Logan
architecture.
Real-time ray tracing,
FFT‑based ocean simulation,
and smoke-particle simulation
on the Kayla platform.
11. Accelerating
Industry
>> safer and smarter automobiles
>> medical applications, including 4D heart
ultrasound that can potentially save lives
>> cinema and special effects
>> geospatial intelligence made possible by video
and image processing
>> digital product design across various
industries
>> big data analytics applied to every day uses
Taking center
stage at the
conference were
breakthroughs in
various fields—from
science to a wide range
of industries—spurred on
by accelerated computing.
GTC 2013 showcased a number
of industry-changing discoveries
made possible by GPUs.
These included:
12. Imaging capabilities in broadcast media
have progressed by leaps and bounds over
the last few decades, especially in sports
broadcasting. This provides today’s sports
fans with a far more intimate and entertaining
experience in the games they love.
The ESPN Emerging Technology Group
is behind many of these technology
breakthroughs. Based on an idea from a prior
GTC, ESPN developed a software architecture
for sports broadcasts using NVIDIA GPUs.
Today ESPN is using GPUs to convert 4K
video input for spectacular “lossless” zoom to
720p for sports broadcasts. Virtual cameras,
real-time overlay effects, and other features
deliver a closer, enhanced experience of
sports broadcasts to viewers.
Harley-Davidson has been
designing and manufacturing
motorcycles for more than a
century. Today, they still produce
vehicles that are coveted by
one of the most loyal customer
bases in the world.
In recent years, the design
process has evolved to include
digital visualization tools during
the conceptual phase, between
styling and engineering. GPU-
accelerated industrial design
integrates art with modeling and
simulation, ultimately reducing
time for product development,
improving styling intent,
allowing greater conceptual
exploration, and delivering
higher-quality designs earlier.
Bringing Fans Closer
to the Game
ESPN
Melding Art and
Simulation in
Industrial Design
Harley-Davidson Motor Company
13. GTC 2013 featured presenters
from Audi, BMW, Chrysler, Fiat,
Honda, and Peugeot Citroen
sharing breakthroughs in how
GPUs are increasingly playing
a central role in auto safety and
infotainment systems, as well
as breakthroughs in design.
Researchers from Audi
revealed how GPUs are used
as part of an initiative to
make driving safer in urban
areas. Highly intelligent,
power‑efficient systems in
cars will soon suggest when to
leave for your daily commute,
whether or not to stop for
coffee, even where to find best
parking choices, all based
on real‑time information.
It’s all about communicating
with the driver in an intuitive,
non‑distracting way.
Navigating Chaotic
Roadways More Safely
Audi
Better Safety and
Infotainment in Cars
GPUs are increasingly
playing a central role in auto
safety and infotainment
systems. Companies
including Audi and
Lamborghini have already
adopted NVIDIA technology,
and it will soon power
models from BMW, Tesla
Motors, Mini, and Rolls
Royce, among others.
For instance, researchers
from Audi revealed how
they are processing big
data in real-time to make
driving safer in urban areas,
eliminate traffic bottlenecks,
and make parking more
efficient. Honda Research
is also working on future
technologies, such
as merging of digital
instrument clusters and
head-up displays. And
researchers at Carnegie
Mellon are using GPUs to
enable gesture recognition
and natural language
processing, enabling a
new generation of human
machine interfaces to
be developed for safer
in‑vehicle use.
With increasing demand for
advanced, power-efficient
computing, NVIDIA unveiled
the Jetson automotive
development platform at
GTC. With this car stereo
sized system, developers
can easily create and
test automotive, image
processing, and computer-
vision applications.
14. Shazam is a commercial
mobile phone-based music
identification service that
connects more than 300
million people in more
than 200 countries and 33
languages. It uses GPUs to
instantly search and identify
songs from its 27 million
track database more than 10
million times a day. This is
accomplished by assigning an
acoustic fingerprint to each
song sample, matching that
sample to their track library,
and returning the answer in
just a few seconds. Today, every
search is done using GPUs.
Because of the performance
and power efficiency of the
GPU, Shazam is able to scale
the operations at less than half
the cost.
Identifying Audio
Patterns
Shazam
Big Data Trend
This year’s conference highlighted a
growing trend of top enterprise and mobile
application companies like IBM and Groupon
using GPUs to accelerate consumer and
commercial big data applications. Industry
leaders such as Shazam, as well as
pioneering startup Cortexica, also use GPUs
to accelerate large-scale audio search, real-
time Twitter analysis, and image matching. In
each use case, GPUs dramatically accelerate
the processing of massive datasets with
complex algorithms, and make it possible
for these big data companies to scale their
infrastructure cost effectively to meet
growing demand.
15. Accelerating
Science >> intelligent object
recognition by robots
and cars
>> image processing
for geospatial
intelligence
>> 3D visualization,
better weather
prediction for
disaster prevention
>> affordable whole
genome sequencing
to predict genetic
defects and diseases
GTC has also been
the venue for the
latest breakthroughs in
science and research on a
variety of topics.
These included:
16. Stretch a strand of human DNA out to its full
length and it’s two meters long. Yet all that
material—and the information it carries—gets
balled up inside the nucleus of a single cell.
By unraveling the human genome, we can
unlock the mysteries around genetic causes
of disease and the environmental factors that
impact genetic behavior.
Using GPUs, Harvard Fellow Erez
Lieberman‑Aiden discovered that DNA comes
together in fractal globules (the same shape
as uncooked ramen noodles) and its folds
determine whether healthy or malignant cells
will be produced. The technique relies on
looking at the billions of snapshots generated
by modern DNA sequencing techniques and
comparing their 3D relationships. GPUs are
essential in analyzing the enormous amount
of data at the heart of the process that
enables researchers to map out a person’s
genome and predict diseases.
Better Human Genome
Mapping and Disease
Prediction
Baylor College of Medicine
Rice University
Faster, Affordable
Gene Sequencing
GPU-accelerated gene
sequencing is driving
down the cost of genomic
research significantly.
The cost to sequence an
entire human genome can
be reduced to $1000 very
soon.
“By democratizing genome
sequencing, we expect
to see an unprecedented
wave of innovation in life
sciences.”
— Alan Williams,
Life Technologies
17. Imagine the impact on public
safety if we could pinpoint a
significant natural disaster such
as the landfall of Hurricane Sandy
five days in advance. Our ability
to do that may be closer than you
realize.
The National Oceanic and
Atmospheric Administration
(NOAA) presented the latest
research in high-resolution
weather models. Such
computationally intense models
were able to pinpoint the landfall
of major storms and hurricanes
such as Sandy. GPU computing
will be essential to the daily use
of highly accurate models for
operational weather modeling,
delivering better power and cost
efficiency in data centers.
Accurate Weather
Modeling and Prediction
NOAA Earth System Research Laboratory
18. The floors of the Adriatic and
Mediterranean seas are littered with tens
of thousands of mines, bombs, and other
munitions that were lost or abandoned
after World War I and II.
To locate and identify the dangerous
materials, NATO is using autonomous
underwater vehicles equipped with
synthetic aperture sonar (SAS) running
on GPUs. The SAS application runs up to
100X faster with GPUs, enabling real-time
object recognition and intelligent decision-
making capabilities within the vehicle’s
six-hour operational window. With GPU
acceleration, mine hunting is faster, more
affordable, more reliable, and safer.
Real-Time Mine
Hunting with
Unmanned
Submarines
NATO STO Centre for
Maritime Research and
Exploration
Istituto Italiano di
Tecnologia, Italy
Researchers have long believed
that human cognition is developed
through interacting with the
environment and other humans
using limbs and senses. And that
human-like manipulation plays
a vital role in the development of
the cognition.
At Plymouth University,
researchers are contributing
to the emergence of humanoid
robots by modeling biological
neural networks to better
understand both human cognition
and artificial intelligence.
These networks consist of
thousands of neurons connected
to each other through millions
of synapses. The systems
integrate visual processing,
linguistics, and other inputs
such as touch, temperature,
and position. This is made
possible with GPUs that perform
the millions of calculations to
activate the neural network every
50-100 milliseconds, allowing
researchers to teach the robot to
think like a human.
Robot
Cognition:
Thinking Like
Humans
Plymouth University
19. GPU Supercomputing
On the Rise
At GTC, the Swiss National
Supercomputing Center
announced it will deploy
NVIDIA GPUs to build
Europe’s fastest GPU
supercomputer. Piz Daint
will be used for scientific
discovery in weather
modeling, astrophysics,
material science, and life
science—the latest evidence
that GPU computing has
passed the tipping point.
GPU accelerators have evolved into
general-purpose processors ideally suited
to tackle massively parallel computing
problems. Today, more than 50 systems
on the Top500 list of supercomputers are
powered by GPUs.
At GTC, representatives from Oak Ridge
National Laboratory presented early
science on the Titan supercomputer—the
world’s fastest for open science—as well
as how researchers and scientists can
gain access to this powerful computing
resource. Titan delivers peak performance
of 27 petaflops, with 18,688 GPUs providing
90% of the computing power, and is open to
academia, government labs, and industry
from across the globe.
Powering
the Biggest
Breakthroughs
Oak Ridge National
Laboratory
20. GTC is where
researchers,
developers, and
technologists
from around
the globe meet
to learn how
others are solving
the toughest
computational
problems.
“It is the best conference for meeting
peoples—it even beats Supercomputing.”
— Guido Juckeland,
Sr. Systems Engineer, TU Dresden
“Unbelievable, to see in the same
place financial engineers, physicians,
astrophysicists, game creators… It is the
only event in the world where you can see
all those talented people!”
— Jonathan Lellouche,
Quantitative Analyst, MUREX
“GTC is action-packed and stimulating like no other
conference. NVIDIA has placed scientific content top
and center of GTC, while at the same time organizing an
exciting program with educational sessions, stunning
demos, and great networking opportunities. There will
be so many people there that I want to meet again or for
the first time. GTC is simply too good to pass [up].”
— Lorena Barba,
Assistant Professor, Boston University
“I was really impressed
with the breadth of the
subjects and the focus
on performance. Really
impressive.”
— Mikael Sorboen,
Head of Risk Systems,
BNP Paribas
“Every year I’ve made a lot of important
contacts for new directions for my
research.”
— Peter Lu,
Post-Doctoral Research Fellow,
Harvard University
“Though it was my first GTC, I was blown
away. I felt like an ant in New York City, but
in a good way. Interacting with strangers
across the spectrum and knowing that
there was so much to learn from them was
mind‑blowing, but an amazing feeling.”
— David Norman,
Engineer/Tool Developer, The Boeing Company