AMD~
DELIVERING THE FUTURE OF HIGH-PERFORMANCE COMPUTING MARK PAPERMASTER Aug 5, 2019
AMD HISTORY OF INNOVATION
:i ·~ ■ .. n. ;==> s 4 mi \'~~~,,
~; . W XBOXONE ~ ·-
First to break the First to break teraflops World 's First 4K game First 16-core World's first 7nm
historic 1GHz barrier performance barrier first APU consoles HEDT processor datacenter GPU
2000 2003 2006 2009 2011 2013 2017 2017 2017 2018 2018 2019
I I
mw .-- .. -=~~ ,. \,. r-. ~, / ~
I 'I!/: ,o ~Y Z i::il'
I ~!'fi': r -;.. ,. .
World 's first x86-64 First to break Inside every major First 32-Core First 32-core First 7nm bit architecture 1GHz GPU barrier gaming console x86 single socket HEDT processor chiplet design
server CPU
GPU AND CPU PERFORMANCE TRENDS
Vl c._
0 _, LL
l9 C 0 ·.;; ·c; ~
c._
Cl/ 01) C
vi
100000GF
10000GF
1000GF
100GF
GPU Single Precision Floating Point Operations Per Second Trend
8 ... ~·····o <i::l.•····· ·····
0 0 g:. ..... . 9 .. •·····ee······ '(£P--~s .. •
eJct•to•o Q t:\Jt_~'( 1.:JS _,,T
'J.'J. L
3 I FUTURE OF COMPUTING COi
100X
0 .:;
"' a:: Cl/ u C
"' E 0 't: lOX Cl/ Q. .... :::, 0.
..c 01) :::,
e ..c I-
Specint® _rate2006 2P Server Performance Trend Over Time
(mc5'···6 .. (9
(?.l,··OO w~ ((!Er' •••
0
.... .. y· ••• •·· o
0·· ~~·~·····
...
AMD~
PROCESS TECHNOLOGY DELIVERED SIGNIFICANT GAINS
Technology Energy Efficiency and Density Across Process Nodes
p,.'( 3 '(£.r>,.RS '(£.r>,.RS ot\'\S\~ t~~'{ £_\ltR'l 3.6
1-"' tf f ,c,~ ~ . .. ,~\\G'f ~
1-"' t,,
~ ..... •····················~
0 ··············~
.... •················1············
0 Energy/cycle ,!, Density Expon . (Density)
4 I FUTUREOFCOMPUTINGCOI AMD~
- - --·--- TO CONTINUED PERFORMANCE IMPROVEMENT
6 J FUTURE OF COMPUTING COi AMD~
MOORE'S LAW KEEPS SLOWING
qonm nSnm
7 I FUTURE OF COMPUTING COi
14nm ... .. 10/7nm ... ~~----_... ······ ~ ••••••
•••••• ••••••
•••••• ••••••
AMD~
WHILE COSTS CONTINUE TO INCREASE Cost Per Yielded mm2 for a 250mm2 Die
6.00
~ 5.00 ~ Cl LU Cl 4.00 ...J LU
i V)
3.00 0 u Cl LU N ::J <!
2.00
~ c::: 0 1.00 z
~2nm 14/1 finm
II\ 1-r"'\r- I\ r""II\ I_.,.......,- r""l"""'Jlr-r"" I\ r"'\r- ,-_,...,.I\ 1,...,.1\ Ill- I\ I 1,1 r"'\r"'\,...,.r"'\I r-1\ II l\""T'"I-
8 I FUTURE OF COMPUTING COi SOURCE: AMD AMD~
- .... ,.. ■ ■ ■ ~-,..~ ■ ■ .... - ■
ADVANCES WILL CONTINUE
9 J FUTURE OF COMPUTING COi AMD~
AMD "Zen 2" CPU-based system scored an estimated 15% higher than previous generation AMD “Zen” based system using estimated SPECint®_base2006 results. SPEC and SPECintare registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org. RZ3-34
CORE DESIGN AND IP
e
A r- n / I I I ,-.. I I r- r""'- I R. I ,-. ""T"" r""'- I I ,-..-r- I '"' R. I ,-. r""'- r- r""'- ,-.. I '"' ,-.. I ,
10 I FUTURE OF COMPUTING COi AMD~
MAINSTREAM 2P SERVER PERFORMANCE
AMD~
c?YC 7nm AMD EPYc'M
QJ C
QJ (' Vl ro co QJ +-' ro a:: ~
0 ········ 0 0 N +-' C
u LU (' a.. V')
12 I FUTURE OF COMPUTING COi AMD~
EXTREME MODULARITY
Chiplets andMCMs
Passive lnterposers and 30 Memory
13 I FUTURE OF COMPUTING rn1
Today
.......................... ,,. .. , ... """"····· ................................. "····"""'
Silicon Modularity from Advanced
30Stacking
System and Component Modularity
Future
AMD~
COMPUTING PLATFORMS
14 J FUTURE OF COMPUTING COi AMD~
HETEROGENEOUS PLATFORMS FPGAs/ Accelerators High Speed Interconnects
. \ \, """' \', " '\\\\\\\\\\\W
~~ \\\\\\\\\\
ex, FABRIC
AMD~
OPTIMIZING SYSTEM PERFORMANCE WITH HETEROGENEOUS COMPUTING
- Semi-Custom Socs & FPGAs
- GPGPU
- General Purpose Cores
APPLICATIONS
16 I FUTURE OF COMPUTING COi CHART FOR ILLUSTRATIVF PURPOSFS AMD~
MACHINE LEARNING EXPANDS ITS REACH
Today
17 I FUTURE OF COMPUTING COi
Future
Future Al algorithms will require unique configurations
Inference for Low-Power Environments
AMD~
FUTURE PROLIFERATION OF HETEROGENEITY
18 I FUTURE OF COMPUTING COi
Today
"'"""""'"· --. - ~ ,;ma ~~ ~ ·
"11 1111 11111111
.. ,.i .... ~. ·1r ...
' ,;ffJ' -~ I I
Future
System level
Memory level ·~ ·- -=- - DOR
~ . • NVDIMM ~ . . " - • Fabric~,ed Memory
AMD~
I 11-1 I '"'r-'"'r--'"'ft II A ft 1-r- --ft 11'"'1 1--lft •-
RESEARCH APPLICATIONS
SPACE EXPLORATION
19 I FUTURE OF COMPUTING COi
CLIMATE CHANGE
· ...... ·- .... ·-.· , ... . .. ~
CHEMICAL SCIENCES
!'7
·.1······
ENERGY SOLUTIONS
MACHINE LEARNING REAL TIME SIMULATION
AMD~
1exaFLOP
1petaFLOP
THE DOE, AMD & CRAY ANNOUNCE FRONTIER
FRONTIER .. -◊
I A ·-"""'I,...,.,_ I\ ,. __ ""T"" """'-' A ,r-"""'r-1 II _,,-""T""r-1\ II A r- r-"'\11\ r-1 -"""'-
21 I FUTURE OF COMPUTING COi AMD~
SOFTWARE CHALLENGES IN EMERGING SYSTEMS
SOFTWARE OPPORTUNITIES IN EMERGING SYSTEMS
FUTURE HIGH-PERFORMANCE COMPUTING REQUIRES INNOVATION IN ALL DIMENSIONS
SILICON
Efficient, power optimized designs
with innovative architecture
24 I FUTURE OF COMPUTING COi
SYSTEM
Interconnects, memory and topologies balanced
to address the key performance bottlenecks
SOFTWARE
Self-optimizing software to
accelerate system capabilities
SECURITY
Secure Virtual Machines
Components that are Robust to Side Channel and
Denial of Service Attacks
Scalable Security from loT to Cloud
Trusted Supply Chain
AMDl1
"Any opinions, findings, conclusions or recommendations
expressed in this material are those of the author(s) and do not
necessarily reflect the views of the Networking and Information
Technology Research and Development Program."
The Networking and Information Technology Research and Development
(NITRD) Program
Mailing Address: NCO/NITRD, 2415 Eisenhower Avenue, Alexandria, VA 22314
Physical Address: 490 L'Enfant Plaza SW, Suite 8001, Washington, DC 20024, USA Tel: 202-459-9674,
Fax: 202-459-9673, Email: [email protected], Website: https://www.nitrd.gov