Upload
others
View
13
Download
0
Embed Size (px)
Citation preview
Where Innovation Begins
Hundreds of optimized system-on-chip solutions
The collaborative ARM business model enables innovative and diverse
ecosystems across markets
275+
Silicon
Partners
85% of
storage
50% of
DTV
25% of
Networking
100s of
Processors
900+
Ecosystem
Partners
30B+
ARM-based
Chips
25B+
Application
Downloads
1B+ Applications
95% of
Mobile
90% of
Tablets
50% of
Set Top
Box
CONFIDENTIAL 3
ARM Strength in Partnership
Most vibrant ecosystem
True multi-vendor
sourcing strategy
Leverage Software
Investment across markets
Joint Investment for success
CPU/GPU – ARM
Application Expertise – Partners
Mill
ion $
(H1x2) 1996 1999 2002 2005 2008 2011
0
50
100
150
200
250
300
CONFIDENTIAL 4
Server SoC’s – One Size Doesn’t Fit All
4
Sca
les w
ith b
raw
ny c
ore
s
Highly parallel
applications
benefit from many
cores / threads
Parallel
applications benefit
most from more
individual servers
with ‘sufficient’ I/O
Workloads that benefit from
frequency and high end cores
(more cache, memory,
bandwidth etc…). Often single
threaded performance.
CONFIDENTIAL 5
ARMv8 Architecture
Next Generation Fabric
Server SoC’s – One Size Doesn’t Fit All
5
Highly parallel
applications
benefit from many
cores / threads
Parallel applications
benefit most from
more individual
servers with
‘sufficient’ I/O
Workloads that benefit from
frequency and high end cores
(more cache, memory,
bandwidth etc…). Often single
threaded performance.
Cortex-A15
CCI-400 Cortex-A9
Sca
les w
ith b
raw
ny c
ore
s
CONFIDENTIAL 6
Architecture: Progression Details
ARMv7-A ARMv7-A – extensions ARMv8-A
32-bit Architecture
32-bit VA & PA
Supports SMP 1-4 cores
TrustZone – Security
ARM & Thumb1/2 ISA
NEON & VFP
L1 D-Cache Coherency
32-bit Architecture
32-bit VA & PA
Supports SMP 1-4 cores
TrustZone – Security
ARM & Thumb1/2 ISA
NEON & VFP
L1 D-Cache Coherency
Full HW Virtualization
40-bit PA
Hypervisor State
2 levels PTT
64-bit Architecture
31x64-bit registers (double)
32-bit instructions
>40-bit PA
New exception model
32-bit Architecture
32-bit VA & PA
Supports SMP 1-4 cores
TrustZone – Security
ARM & Thumb1/2 ISA
NEON & VFP
L1 D-Cache Coherency
Full HW Virtualization
40-bit PA
Hypervisor State
2 levels PTT
2004 2010 Next
Cortex-A9
Cortex-A15
CONFIDENTIAL 7
Why ARM Servers and Why Now?
Contributing Factor Trend in Data Center Arena
Economic & environmental
motivation to switch from
traditional approaches
Modest barrier to entry for new
solutions – prevalence of open
source in scale out deployments
Opportunity for innovation
Diversification of server
workloads mean one size
does not fit all
CONFIDENTIAL 8
Opportunity: Re-architect Efficient Servers
Highly efficient CPU
architecture
Integration &
Innovation
Volume Server Power Breakdown
CPU Power ~1/3
Everything else ~2/3
ARM Roadmap &
Business Model
Enables a New
Class of Energy
Efficient Servers
CONFIDENTIAL 9
Time
Pe
rfo
rma
nce
/Jo
ule
/$
O/S
Multiple Apps
CPU
Linux O/S
Single Application
Large Core CPU
Virtualization
Linux O/S
Multiple Small CPU’s
Integrated Fabric
Heterogeneous Processing
SOC’s
H/W Accelerators
Server Evolution and Drivers
Cloud
Power Efficiency
Distributed
Computing
Servers Will Mimic Mobile SOC Evolution
CONFIDENTIAL 10
ARM Server Market Will Grow Over Time
Identified Target Workloads
Search
Social Media
Web Serving
Offline Analytics Business Logic
Scalable Cloud
Distributed Database IT Infrastructure
Traditional Database
Compute
Time
Cortex-A9
Cortex-A15
ARMv8
Web Tier servers are a $14-$16 billion dollar market.
The Web Tier of Internet data centers account for $6-$8
billion
Online Gaming
CONFIDENTIAL 11
RAN Evolution
Trend: Cloud RAN (C-RAN)
•Benefit: Centralised macro BST with fibre
link to intelligent antennas
•Proposition: power efficient server,
signalling and control macro BST platforms
Trend: large & small cells in
heterogeneous networks
•Benefit: Increased QoE with better edge-
of-cell capacity
•Proposition: scaleable C-progammable
power efficient HW & SW BST platforms
Trend: RAN IT functionality
•Benefit: Reducing backhaul traffic &
improving QoE through distributed caching
and open-API cloud apps
•Proposition: energy efficient server
platforms
Network evolution challenges
• LTE topology: IP backhaul & flatter net
• Multiple RATs with intelligent antennas
• Higher frequencies
• Better QoS through local services
• Power consumption