2. 2 | “CARRIZO” | ISSCC 2015 |
CAUTIONARY STATEMENT
The following presentation contains forward-looking statements concerning Advanced Micro Devices, Inc. (“AMD” or the “Company”) including, among other things: timing, availability,
features and functionality of the AMD “Carrizo” APU, AMD “Carrizo –L” SoC; AMD’s ability to improve its energy efficiency technologies for its future products; AMD’s goal to improve
energy efficiency for AMD mobile platforms by at least 25 times by 2020; and AMD’s goal to outpace historical energy efficiency trends, which are made pursuant to the safe harbor
provisions of the Private Securities Litigation Reform Act. Forward-looking statements are commonly identified by words such as "would," "may," "expects," "believes," "plans,"
"intends," "projects," and other terms with similar meaning. Investors are cautioned that the forward-looking statements in this press release are based on current beliefs, assumptions
and expectations, speak only as of the date of this press release and involve risks and uncertainties that could cause actual results to differ materially from current expectations.
Investors are cautioned that the forward-looking statements in this presentation are based on current beliefs, assumptions and expectations, speak only as of the date of this
presentation and involve risks and uncertainties that could cause actual results to differ materially from current expectations. Risks include that Intel Corporation's pricing, marketing
and rebating programs, product bundling, standard setting, new product introductions or other activities may negatively impact AMD’s plans; that AMD will require additional funding
and may be unable to raise sufficient capital on favorable terms, or at all; that customers stop buying AMD’s products or materially reduce their operations or demand for AMD’s
products; that AMD may be unable to develop, launch and ramp new products and technologies in the volumes that are required by the market at mature yields on a timely basis; that
AMD’s third-party foundry suppliers will be unable to transition AMD’s products to advanced manufacturing process technologies in a timely and effective way or to manufacture
AMD’s products on a timely basis in sufficient quantities and using competitive process technologies; that AMD will be unable to obtain sufficient manufacturing capacity or
components to meet demand for its products or will not fully utilize its projected manufacturing capacity needs at GLOBALFOUNDRIES, Inc. (GF) microprocessor manufacturing
facilities; that AMD’s requirements for wafers will be less than the fixed number of wafers that it agreed to purchase from GF or GF encounters problems that significantly reduce the
number of functional die it receives from each wafer; that AMD is unable to successfully implement its long-term business strategy; that the completion and impact of the 2014
Restructuring Plan and AMD’s transformation initiatives could adversely affect AMD; that AMD inaccurately estimates the quantity or type of products that its customers will want in
the future or will ultimately end up purchasing, resulting in excess or obsolete inventory; that AMD is unable to manage the risks related to the use of its third-party distributors and
add-in-board (AIB) partners or offer the appropriate incentives to focus them on the sale of AMD’s products; that AMD may be unable to maintain the level of investment in research
and development that is required to remain competitive; that there may be unexpected variations in market growth and demand for AMD’s products and technologies in light of the
product mix that it may have available at any particular time; that global business and economic conditions will not improve or will worsen; that PC market conditions will not improve
or will worsen; that PC market conditions will not improve or will worsen; that demand for computers will be lower than currently expected; and the effect of political or economic
instability, domestically or internationally, on AMD’s sales or supply chain. Investors are urged to review in detail the risks and uncertainties in the Company's Securities and Exchange
Commission filings, including but not limited to the Quarterly Report on Form 10-Q for the quarter ended September 27, 2014.
3. 3 | “CARRIZO” | ISSCC 2015 |
Single, scalable infrastructure shared with “Carrizo-L”
New “Excavator” core optimized for low power
notebook/convertible form factors
Next Generation AMD Radeon™ Graphics Core Next
architecture with support for Mantle, DirectX® 12, and
Dual Graphics
Single-chip integration of the APU and the Southbridge
onto a single die
Significant performance and battery life improvements.
First processor in the world with full HSA 1.0 support
AMD Secure Processor, leveraging ARM® TrustZone
technology for Enterprise-class security
NEW PERFORMANCE MOBILE APU – “CARRIZO”
NEXT GENERATION PERFORMANCE APUs
WITH FULL HSA CAPABILITY
4. 4 | “CARRIZO” | ISSCC 2015 |
INTRINSIC ENERGY EFFICIENCY WITH ACCELERATED PROCESSORS (APU)
75.0
80.0
85.0
90.0
95.0
100.0
75.0
80.0
85.0
90.0
95.0
100.0
75.0
80.0
85.0
90.0
95.0
100.0
GPU-centric Balanced CPU-centric
CPU <-> GPU Power Trading
GPU Power CPU Power
Combines CPUs, GPU and multi-media
accelerators on a single die
APUs are optimized for greater efficiency
– Efficient, fine-grained power management between
CPU and GPU
– CPU <-> GPU communication power dramatically
reduced relative to separate chips
– Shared memory interface helps save power
Temperature Distribution Plots
GPU
Dual Core x86
Module
Dual Core x86
Module
L2 CacheL2 Cache
UNB
GMC
DDR3 Controller
Multimedia
5. 5 | “CARRIZO” | ISSCC 2015 |
AMD APU ENERGY EFFICIENCY WITH HSA
“CARRIZO” IS THE FIRST FULLY HSA COMPLIANT SOC
COMPUTE CAPACITY TREND IN PCs
Source: AMD Internal data
0
100
200
300
400
500
600
2010 2011 2012 2013 2014GFLOPS
Product Year
CPU GFLOPs GPU GFLOPs
PCMark®7
PCMark®8 v2
FUTURE
Example Improvement from GPU-Compute
1st APU
WHAT DOES THIS MEAN FOR POWER?
Many workloads execute more
efficiently using GPU compute
resources rather than CPU only
‒ E.g. video indexing, natural human interfaces,
pattern recognition
For the same power, much better
performance: lower energy per
operation greater efficiency
INTEGRATED GPU TREND
CPU TREND
6. 6 | “CARRIZO” | ISSCC 2015 |
POWER OPTIMIZED CPU “EXCAVATOR” WITH HIGH DENSITY LIBRARY DESIGN
“Steamroller” Library Implementation “Excavator” High-density Library Design
L2
XV
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
0.0W 5.0W 10.0W 15.0W 20.0W 25.0W
NormalizedFrequency
Power Per Core Pair
SteamRoller_x86-64 CPU High performance library
Excavator_x86-64 CPU High density library
HP Library HD Library
Floating Point
Scheduler
38%
FMAC
35%
I-Cache Control
35%
High Performance vs. High Density Cell Example
Prior generation
CPU-centric
tapered metal
stack
General
purpose
GPU-oriented
stack enables
greater density
Achieves 23% area reduction, and lower
power in the same 28nm technology node
L2
L2
SR
7. 7 | “CARRIZO” | ISSCC 2015 |
“CARRIZO” LOW POWER OPTIMIZED GRAPHICS
0.0W 5.0W 10.0W 15.0W 20.0W 25.0W 30.0W
NormalizedFrequency
Power (8 GCN core GFX engine)
High density power optimized High perf legacy design
18% leakage reduction and timing with faster RVT
devices enables 10% higher frequency at same power
level, or up to 20% lower power at same frequency
34.9%
23.7%
8.2%
32.7%
0.3%
0.01
0.1
1
10
0.6 0.7 0.8 0.9 1 1.1 1.2
NormalizedIoff
Normalized Ion
high-density device selection high-performance device selection
28nm universal curve
48.8%
43.2%
7.9%
0.1%
0.3%
Leakage reduced with heavy
use of high Vt devices
Speed set by
faster devices
28nm device suite
8. 8 | “CARRIZO” | ISSCC 2015 |
VOLTAGE ADAPTIVE OPERATION
Voltage adaptive feature
applied to both CPU and
GPU in “Carrizo” results in
19% and 10% power savings
respectively
Vdd above threshold
Vdd drops below thresholdResponse Time
< 1 nano-second
Traditional margined frequency
Operating frequency
temporarily decreases
~ 5%
Frequency
Benefit
Vdd Supply
Delivering low noise voltage to high performance CPUs, GPUs and
APUs has always been a challenge for the industry
The variations that happen are typically about 10% of the nominal
value – that means at least 20% power is wasted covering these
voltage variations (power goes as the square of voltage)
AMD’s unique voltage adaptation feature recovers much of that
wasted power by operating at the average voltage and quickly
reducing frequency for the brief periods when the voltage reduces
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
1 1.15 1.2 1.3 1.33 1.4 1.5 1.65 1.7
PercentPowerSavings
Normalized Clock Frequency
Adaptive Clocking Power Savings
with Voltage Adaptive feature vs. without
CPU Power Savings GPU Power Savings
9. 9 | “CARRIZO” | ISSCC 2015 |
AVFS TO OPTIMIZE PERFORMANCE PER WATT
“Excavator” core
incorporates
10 AVFS modules
containing
~500 frequency
sensing paths
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
0.0W 5.0W 10.0W 15.0W 20.0W 25.0W
NormalizedFrequency
Power Per Core Pair
SteamRoller_x86-64 CPU High performance library
Excavator_x86-64 CPU High density library
Excavator High Density library AVFS
AVFS provides additional
power reduction at a given
performance level over the
High Density library gains
Reliably extract the true silicon speed capability of CPU
‒ Includes effects of part-to-part processing, temperature and power delivery
‒ Add both a voltage and frequency sensor to existing power and temperature sensors
Enables accurate setting of the optimal operating point for a given power or
performance level across process, voltage and temperature ranges
‒ Improved energy efficiency across the entire voltage/temperature operating range
10. 10 | “CARRIZO” | ISSCC 2015 |
RUN-TIME ACCESS TO LOW POWER STANDBY STATE
The S0i3 state achieves the same power level as the legacy
S3 state, traditionally known as “standby” which is very
time-consuming to enter and exit because it requires
operating system intervention.
‒ This state has almost all of the APU silicon power-gated and all
relevant I/O devices in their low-power states, driving platform
power to extremely low levels.
By enabling access to this state on the fly, under the control
of power management, the APU can achieve standby
equivalent power levels transparently at sub-second time
frames
Lower average power consumption for typical use conditions.
Boost Active Sustained Idle S0i3
APUPower
Power vs. State
< 50mW
GFX
IO
ACP FCH
NB
PLL
Core
Core
Core
Core
GFX
IO
ACP FCH
NB
PLL
Core
Core
Core
Core
GFX
IO
ACP FCH
NB
PLL
Core
Core
Core
Core
I/O
< 1.5W
> 10 W
11. 11 | “CARRIZO” | ISSCC 2015 |
DRAMATIC IMPROVEMENTS IN ENERGY EFFICIENCY
TRENDS IN THE ENERGY EFFICIENCY OF COMPUTATION1
IMPROVE
ENERGY
EFFICIENCY
for AMD mobile platforms
1
10
100
1000
10000
100000
1000000
10000000
1985 1990 1995 2000 2005 2010 2015 2020
EfficiencyRelativeto1985(1985=1.0)
Post 1946 historical trend 2000 to 2009 historical energy efficiency
Efficiency growth
starts to fall off
after 2000
AMD goal: outpace the
historical trend by 70%+
2
Energy use drops While performance increases = Increased efficiency
by at least
25TIMES
GOAL BY 2020
12. 12 | “CARRIZO” | ISSCC 2015 |
“CARRIZO” TAPS INTO A DEEP PIPELINE OF IMPROVEMENTS
AMD has been building a pipeline of
power focused IP for many years
Watch for more leading edge innovations
enabling accelerated power gains in the
future on all product lines
2013201220112010
Dynamic power tracking and
management with DVFS
Thermal aware
power
management
Platform aware dynamic
thermal management
Fine grained power
gating
Dynamic CPU
GPU power sharing
2014
Dynamic thermal
tracking for short
term boost
Finer grained
voltage planes
Voltage-adaptive
frequency
scaling
ACPI driven workload
specific optimization
Intelligent boost and
Performance aware energy
optimization
Finer grained power tracking,
increased voltage granularity
2015
Video encode
acceleration
Video decode
acceleration
Audio
acceleration
Power efficient APU architecture
integrating GPU+CPU and accelerators
In market In product In development
2016
Environment and
reliability aware
boost
Advanced
bandwidth
compression
Inter-frame
power gating
Per-IP
adaptive
voltage
Per part
adaptive
voltage
Workload
aware energy
optimization
Integrated voltage
regulation
OpenCL GPU compute moving to
full HSA enabled programming
"Puma" "Tigris" "Danube"
"Llano"
"Trinity"
"Richland"
"Kaveri"
"Carrizo"
2008 2009 2010 2011 2012 2013 2014 2015
EnergyEfficiency
Typical-Use Energy Efficiency
13. 13 | “CARRIZO” | ISSCC 2015 |
High density design library resulting in 29% more transistors
than “Kaveri” in approximately the same die area
‒ 3.1 billion transistors
Excavator cores: 5% more IPC at 40% less power and 23%
less area
H.265 support and > 3.5x transcode performance of “Kaveri”
Device selection and implementation tuning enable the
eight AMD Radeon™ cores to reduce power 20% from
“Kaveri”
Double digit increases in performance and battery life
NEW PERFORMANCE MOBILE APU – “CARRIZO”
ISSCC 2015 DISCLOSURES
A 28NM X86 APU OPTIMIZED FOR POWER
AND AREA EFFICIENCY – SESSION 4.8*
*A 28nm x86 APU Optimized for Power and Area Efficiency , presented by Kathryn Wilcox. Session 4.8, Solid-State Circuits Conference Digest of
Technical Papers (ISSCC). 2015 ISSCC Conference, February, 2015.
14. 14 | “CARRIZO” | ISSCC 2015 |
ENDNOTES
1 Koomey, Jonathan G., Stephen Berard, Marla Sanchez, and Henry Wong. 2011. "Implications of Historical Trends in The Electrical Efficiency of Computing." IEEE Annals of the History of Computing. vol. 33, no. 3. July-September. pp. 46-54.
[http://doi.ieeecomputersociety.org/10.1109/MAHC.2010.28]
2 Typical-use Energy Efficiency as defined by taking the ratio of compute capability as measured by common performance measures such as SpecIntRate, PassMark and PCMark®, divided by typical energy use as defined by ETEC (Typical Energy
Consumption for notebook computers) as specified in Energy Star Program Requirements Rev 6.0 10/2013
15. 15 | “CARRIZO” | ISSCC 2015 |
ATTRIBUTIONS
DISCLAIMER
The information contained herein is for informational purposes only, and is subject to change without notice. While every precaution has been taken in the preparation of this
document, it may contain technical inaccuracies, omissions and typographical errors, and AMD is under no obligation to update or otherwise correct this information. Advanced
Micro Devices, Inc. makes no representations or warranties with respect to the accuracy or completeness of the contents of this document, and assumes no liability of any kind,
including the implied warranties of noninfringement, merchantability or fitness for particular purposes, with respect to the operation or use of AMD hardware, software or other
products described herein. No license, including implied or arising by estoppel, to any intellectual property rights is granted by this document. Terms and limitations applicable to the
purchase or use of AMD’s products are as set forth in a signed agreement between the parties or in AMD's Standard Terms and Conditions of Sale.
AMD, the AMD Arrow logo, Radeon and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification
purposes only and may be trademarks of their respective companies. DirectX is a registered trademark of Microsoft Corporation in the U.S. and other jurisdictions.