opencl & the future of desktop high performance computing in cad
DESCRIPTION
Modern desktop computers have more compute capabilities than ever before. Most of these systems include both a central processing unit (CPU) and a graphics processing unit (GPU), each consisting of multiple computing cores providing tremendous processing power. To date, harnessing the total processing power of a desktop workstation, fully utilizing both the CPU and GPU, has proven difficult for software developers. CPUs and GPUs have few similarities in both design and programming models. OpenCL is the tool that bridges the gap for software developers and enables them to fully tap into the power of both processors with a single software programming interface. This presentation will examine the details of CPUs and GPUs, explore their differences and similarities, and highlight the computing power they can provide. We will also take a look OpenCL, what it is, what it does, and how this new computing interface will change the way software developers create software and help end users fully realize the compute power contained within today’s modern desktop computers.TRANSCRIPT
OpenCL™ & the Future of Desktop High Performance
Computing in CAD
This webinar will be available afterwards at
designworldonline.com & email
Q&A at the end of the presentation
Hashtag for this webinar: #DWwebinar
Before We Start
Moderator
Joe Gorse Design World
Presenter
Allen Bourgoyne AMD
OpenCL™ & the Future
of Desktop High
Performance
Computing in CAD Allen Bourgoyne, Director ISV Alliance Team
AMD Professional Graphics
Agenda • Introduction
• Processors everywhere
• History of GPU Computing
• What is OpenCL™
How is OpenCL™ changing CAD?
• Conclusion
Introduction
Who am I? • Allen Bourgoyne
o Director, ISV Alliances AMD Professional Graphics
What is OpenCL™? • Open Compute Language, OpenCL™, is the first open,
royalty free standard for cross platform programming for
personal computers, servers, workstations, hand-held
devices, supporting a variety of CPUs, GPUs, and DSPs.
Processors Everywhere
Modern Computers Have Lots of Processors
• Central Processing Unit (CPU)
• Graphics Processing Unit (GPU)
• Others o Network Controllers
o Device controllers (disks, DVD-ROMs, etc)
o Smart Batteries
o Most of these are not generally available for application software use
What Processors Can I Use?
• CPU: traditionally runs operating system, user programs,
system functions o General purpose design: run lots of things
o Some parallel processing capability
• CPUs have multiple processing cores (2, 4, 8, up to 12)
• Can use multiple CPUs to increase parallel processing capability
o Processing power & memory generally available to user programs
• Low-level compute functionality generally available to programs
• Programs can directly access memory, operating system can provide for
“virtual memory”
What Processors Can I Use?
• GPU: traditionally runs graphics programs o Highly focused design: Run graphics programs
o Highly parallel processing designs
• Modern GPUs can have over 1000 processing units!
o Low-level compute functionality not generally available to programs (until very
recently)
o Processing power & memory generally available to graphics programs
• Physical memory only, no virtual memory available
What Processors Can I Use?
• Others: o Not generally available for application software
o Can be used (not always for good!)
• Recent hacks have used “smart” battery controllers to ruin batteries (over
charge, not charge) and install malware5
o Historically, if there is a processor in the system that has an interface that can be
exploited, it will be eventually
• Focus today: CPUs & GPUs
What is HPC? • High Performance Computing
Historically refers to computing that involves extremely large amounts of computer
processing
Examples:
Simulating nuclear explosions
Global weather
Seismic data processing (looking for Oil & Gas)
Breaking codes/ciphers
Computing Pi to ridiculous numbers of decimal points!!
HPC Has Been Around for a Long Time
• Original mainframe computers designed to solve complex math
problems
• CAD: crash simulation, structural analysis, etc
• Historically done on compute servers o Lots of CPUs with lots of cores
o Jobs submitted in batches
o Long turn around times, some times takes days or weeks!
• Desktops/laptops getting more powerful o HPC workloads showing up here!
1970 1980 1990 2000 2010+
Brief History of GPU Computing
Ikonas (1978)
Pixel Machine (1989)
Pixel-Plane 5 (1992)
PixelFlow SIMD gfx cracks UNIX password encryption(1999)
OpenGL (1992)
SGI GL (1984)
ATI CTM (2006)
Nvidia CUDA (2007)
OpenCL (2009)
Using the GPU for Computing
• Using GPUs for computing has been around for a long
time! o As soon as they started showing up in computers, people started trying to use them
to help speed up compute tasks
• Why? o GPUs have some unique design characteristics that enable them to perform certain
mathematical functions extremely fast
Computer Graphics 101 • How to draw a triangle on your computer screen:
Y
X
Z
A(x,y,z)
B(x,y,z)
C(x,y,z)
Computer Graphics 101 – Math to Draw that
Triangle!
• Draw lines, fill triangle:
Determine slope for each line:
Slope = (Bx – Ax) / (By – Ay)
Any coordinate on line:
Cx = Ax + slope * (Cy – Ay)
• Project 3D points onto 2D display:
o For point (Ax, Ay, Az), projected point (Px, Py)
+ = Px
Py
Cx
Cz
Ax
Ay
Az
Sx 0 0 0 0 Sz
A(x,y,z)
B(x,y,z)
C(x,y,z)
Do this for vectors
AB and AC, we will have
the points to create lines
to fill the triangle
GPU Processing Power • As you can see, it takes a bit of math processing to draw
that triangle
• Problem: Stuff we want to draw has lots of triangles! o Interactive rates are 30 frames per second (fps)
• Example: 1 million triangle model @ 30 fps requires drawing 30 million
triangles per second!
o And that’s a small model!
• Solution: GPUs have to be able to process lots of triangles!
GPU Processing Power • Modern GPUs can process hundreds of millions of triangles
per second o That’s a lot of vector math: DOT products, matrix multiplies, etc.
• High degree of parallel processing enables GPUs to handle
this workload o Modern GPUs do thousands of operations in parallel in order to meet the demands
that graphics applications place on the hardware
GPU Processing Power • So what does this mean?
o GPUs have a lot of processing power!
• CPU vs GPU computing power:
• CPU: ~ 200 GFLOPS2
• GPU: > 3 TFLOPS1
So Why Should I Care if My GPU is used for Compute?
• Answer: Money & Performance!
• Not quite that simple, let’s take a look at how our
computer generally operates…
Typical Computer Utilization
GPU CPU
Compute Usage • In most cases, the CPU and GPU aren’t that busy when the
other guy is working: o GPU mostly idle while loading files, writing to disk, etc.
o CPU can be less busy when waiting for the GPU to complete graphics tasks
• These ebbs & flows of workloads create idle cycles, but also the opportunity to
move compute tasks to the available resource
o Try to make use of the idle time!
Compute Power • Remember the CPU vs GPU performance comparison?
ALU
ALU ALU
ALU
CACHE
MEMORY MEMORY
CPU
GPU
How do I get my $$$s Worth?
• Software developers are working on tapping into those
unused compute cycles and untapped compute power of
the GPU o End user software demands ever increasing:
• Modern automotive models can contain up to 50,000 parts with 10 to 20 GB of data. The number of
triangles can reach 40,000,000 polygons/model6
o OpenCL™ is a tool that will enable software developers to tap into the full power
available on the computer!
What is OpenCL™?
OpenCL™ • Industry standard programming language for parallel
computing
• Specification by Khronos
• Software using OpenCL runs on many enabled devices o Runs on CPUs, GPUs, ARM processors, Windows, Linux, Apple OS, Android Os’s
o Supported by major hardware & software vendors including AMD, Intel, Nvidia,
Apple, ARM
Who is Khronos? • Open consortium creating standards
o The Khronos Group is a not for profit industry consortium creating open standards
for the authoring and acceleration of parallel computing, graphics and dynamic
media on a wide variety of platforms and devices
o Commitment to royalty free standards
o Founded almost 10 years ago, over 100 members, any company welcome to join
o Standards include OpenGL®, OpenCL™, WebGL, WebVG™, OpenWF™
OpenCL™ Working Group Members
Developers Want Open Standards!
• June 2011 developer
survey shows adoption of
OpenCL3
OpenCL™ & CAD • OpenCL™ is a powerful tool designed to unleash the power
of processors in a platform independent way
• Software developers are taking advantage of OpenCL™ to
provide significant increases in both computation and
graphics compute performance
OpenCL™ Based Solutions • Engineering analysis
o Very high compute requirements, GPUs can help offload some of the compute
workload
o Solutions shipping today from Dassault Systemes, OpenCASCADE
Performance Gains with OpenCL™
220
222
224
226
228
230
S4B Benchmark customer Data #1
Over 200% Speed-up with OpenCL™ VS. CPU only4
Abaqus/Standard 6.11
Job Name Operations per Iteration Solver Speed Up Overall Speedup using CPU plus GPU & OpenCL
s4B Benchmark dataset 10.3 TFLOPS 2.4X 2.3X
Customer Data #1 5.75 TFLOPS 2.4X 2.2X
% speed-up
with GPU &
OpenCL™
% S
pee
d-u
p
OpenCL™ & Design • OpenCL™ will impact the design phase as well
o Ability to accelerate high-quality, photo-realistic rendering in real time
• Interact with realistic models in realistic environments in real time!
o Provide physically accurate rendering: lights, reflections, etc. are rendered physically
where they will show up on the actual product
OpenCL™ & Design
OPTIS Theiea simulate reality for design review pipelines
Image courtesy of OPTIS
Chaos Group V-Ray render engine with GPU acceleration provides interactive rendering
Image courtesy of Chaos Group
OpenCL ™ & CAD • Just scratching the surface
o OpenCL only available for less than 2 years
• Already impacting analysis & design software
• Migration from other software verticals o Digital Content Creation (DCC)
• Realistic cloth simulations, physics, particle simulation, AI already supported
What’s Next? • Any computationally intensive task is a good candidate:
o CFD, FEA, particle simulation, etc.
• Most major ISVs already working with OpenCL™ o Many more software titles will implement OpenCL in 2012
Thank You
References 1. http://www.amd.com/us/products/desktop/graphics/amd-radeon-hd-6000/hd-6990/Pages/amd-radeon-hd-6990-
overview.aspx#3
2. http://www.maxxpi.net/pages/result-browser/top10---flops.php
3. Market data provided by Evans Data Corporation | June 2011
4. Testing conducted on a Dell T7400 w/ Intel Xeon E5405 CPU, 64GB RAM, Red Hat® OS v5.5, ATI FirePro™ V9800,
AMD Catalyst Pro 10.12, Dassault Systemes SIMULIA ® Abaqus/Standard 6.11
5. http://www.v3.co.uk/v3-uk/news/2099616/black-hat-charlie-miller-explains-apple-battery-hack
6. http://cadspeed.wordpress.com/2011/05/18/opencl-will-rock-the-cad-world-part-1-why-you-want-it/
Disclaimer The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical
errors.
The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and
roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing
manufacturers, software changes, BIOS flashes, firmware upgrades, or the like. AMD assumes no obligation to update or otherwise correct or revise
this information. However, AMD reserves the right to revise this information and to make changes from time to time to the content hereof without
obligation of AMD to notify any person of such revisions or changes.
AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY
INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION.
AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL
AMD BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY
INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
Questions?
Design World Joe Gorse Email: [email protected] Phone: 440.234.4531 ext. 111 Twitter: @DesignWorld_EE
AMD Allen Bourgoyne Email: [email protected] Phone: 512.602.4738
Thank You
This webinar will be available at designworldonline.com & email
Tweet with hashtag #DWwebinar
Connect with
Twitter: @DesignWorld
Facebook.com/engineeringexchange
LinkedIn: Design World Group
YouTube.com/designworldvideo
Discuss this on EngineeringExchange.com