AMD's Mr. Greg Poole, Director, Worldwide Commercial Field Applications Engineering, AMD, gave a speech at COMPUTEX TAIPEI Industry Forum 2012 on Today’s Cloud Computing Technology Landscape.
2. THE CHANGING FACE OF THE DATA CENTER
x86 Server CPUs CAGR
18
Cloud
16 New business models, growth of +15%
tablets/smartphones
14
12 Virtualization
(Millions)
The quest for greater data center +13%
10 efficiency
8
HPC
6 Market-rate growth +7.3%
4
Traditional IT
2 Shrinking as cloud and -1.2%
- virtualization grow
2012 2013 2014 2015
Source: IDC Server Forecast
2 | AMD’s Private Cloud | June 2012 |
3. AMD’S PRODUCT CAPABILITIES WILL ENABLE MARKET TRENDS
Implications for
our products
Value pricing
Performance to get the job done
Consumerization Versatile solutions
Integration of graphics and video IP
Low-power solutions
Cloud Efficient I/O
Datacenter solutions
Products with interoperability
Convergence Secure computing
3 | AMD’s Private Cloud | June 2012 |
4. THE CLOUD DATA CENTER IS AMD’S FOCUS SERVER MARKET
CAPTURING THE INFLECTION POINT
AMD has strong momentum with the latest
AMD Opteron™ products
─ Focused on execution for our customers
Today, we are making a strategic investment
to double down on the server market addressing
the Cloud data center
The Cloud is the fastest growing segment in the
server market A BETTER OPTION FOR
THE HYPER-EFFICIENT,
Customers focusing on low-power, energy VIRTUALIZED,
efficient data centers CLOUD-READY WORLD
Sources: Customer interviews, AMD Internal forecasts
4 | AMD’s Private Cloud | June 2012 |
5. TODAY’S AMD OPTERON™ PROCESSORS
Greater Performance Greater Efficiency
World’s first 16-core x86 First and only processor
processor1 with TDP Power Capping
for maximize power
First processor with up to budgets and floor space5
1GHz boost over base
frequency2 using AMD Turbo C6 power state enables
CORE Technology ultra low power by gating
power to idle cores
20% faster memory access3
World’s first truly As low as 5.3 W/core6,
Up to 84% more throughput4 modular x86 reduced processor power
on HPC workloads processor core at idle by up to 46%7
design
See Appendix B for footnotes.
5 | AMD’s Private Cloud | June 2012 |
6. AMD OPTERON™ PLATFORMS
A WIDE RANGE OF PLATFORM CHOICES TO MEET BOTH STANDARDIZED AND
CUSTOMIZED ENVIRONMENTS
AMD Opteron™
Performance-per-
6000 Series Platform
watt and
Expandability for Standard Platforms
Traditional Rack/Tower/Blade
2P/4P
AMD Opteron™
Highly Energy 4000 Series Platform
Efficient and Custom, purpose-driven Twins/
Cost-optimized Container/”Skinless” Scale Out
for 1P/2P
Low cost SMB servers
AMD Opteron™
Price-optimized 3000 Series Platform
cost-effective Custom, purpose-driven low
infrastructure for power systems
1P servers
Low cost, dedicated hosting and
small business servers
6 | AMD’s Private Cloud | June 2012 |
7. LESSONS FROM AMD'S PRIVATE CLOUD
• Challenges:
• Software was non-standardized within AMD
• Resources not flexible and accessible
• Compute power in isolated pockets
• Solution:
• The AMD IT organization transforms the AMD
infrastructure into a private cloud
• Impact:
• Grew base compute resource capacity over 20%
“ We’ve never seen increases in productivity, • Achieved sustained utilization rate greater than 90%
collaboration and efficiency like we’ve seen with the • Saved over $6 million through in-place upgrades
cloud. Projects that used to take weeks can take days.
We believe that the AMD cloud represents a significant
business advantage.”
- Bob Luong, Director for Systems Engineering, AMD
7 | AMD’s Private Cloud | June 2012 | Link to AMD Cloud Case Study
8. CHALLENGES OF DEVELOPMENT COMPLEXITY (CIRCA 2009)
Semiconductor design requires massive compute power!
Global development teams
Multiple disciplines (Thermal, HW, SW, BIOS, Logical, Physical, Implementation, Verification…)
Many servers (over 15,000 servers used for development)
Multiple Operating Systems
Heterogeneous server environment
Various storage and management solutions
Numerous applications
Significant network bandwidth required
Growing infrastructure
Upgrading HW and SW was a near constant activity
8 | AMD’s Private Cloud | June 2012 |
9. OTHER CHALLENGES
Managing and moving data
– Huge datasets required for design, validation and simulation
– Global installations not connected
– Mirror images of data created (management and synchronization problem)
– Establishing usable project data could take weeks
Testing Semiconductor Designs
– Requires access to thousands of cores for vector generation and validation
– Limited “excess” compute capacity to avoid over provisioning at any given site
– No way to shift compute power to a given project
9 | AMD’s Private Cloud | June 2012 |
10. AMD COMPUTE ENVIRONMENT IN 2009
Massive compute capability which did not meet AMD’s need for
– Flexibility
– Manageability
– Accessibility
– Cost effectiveness
10 | AMD’s Private Cloud | June 2012 |
11. FOUNDATION FOR CLOUD ENVIRONMENTS
AMD OPTERON™ PROCESSORS
Performance: Meet the changing needs of the
demanding private cloud environments.
Dependability: Achieve predictable and consistent
throughput and deliver reliable system access for multiple
teams.
Security: Help protect system integrity and sensitive
data with strong security features.
Manageability: Administer multiple server platforms
across diverse departments, sites, and geographies.
Value: Optimized performance per dollar per watt.
11 | AMD’s Private Cloud | June 2012 |
12. AMD IT MIGRATES INFRASTRUCTURE TO THE CLOUD
Private cloud transformation started April 2009
Strategy: “Compute Anywhere”
– Centralized Datacenters
– Connectivity Tools
– Support for all projects, all global locations
Early Phase Benefits
– Platform upgrades Newer, Power efficient processors
– Leveraged existing socket infrastructure
– Doubled computing infrastructure
– Over US$6 million in savings
– Standardized software (OS, Workload management, EDA tools)
12 | AMD’s Private Cloud | June 2012 |
13. THE AMD CLOUD TAKES SHAPE
Continuous program execution during migration
Compute capacity increased by 20%
Today
– More than 115,000 AMD CPU cores
– More than 4 Petabytes of storage
1 Gigabyte = 7 minutes of HD-TV video or 1,073,741,824 bytes
4.7 Gigabytes = Capacity of a standard DVD-R
1024 Gigabytes = 1 Terabyte
1024 Terabytes = 1 Petabyte
– Dynamic compute allocation to meet project needs
– Enhanced manageability
– Wide access to resources
– Lower storage costs
http://mozy.com/blog/misc/how-much-is-a-petabyte/
– Increased productivity
13 | AMD’s Private Cloud | June 2012 |
14. LESSONS FROM AMD'S PRIVATE CLOUD
• Challenges:
• Software was non-standardized within AMD
• Resources not flexible and accessible
• Compute power in isolated pockets
• Solution:
• The AMD IT organization transforms the AMD
infrastructure into a private cloud
• Impact:
• Grew base compute resource capacity over 20%
“ We’ve never seen increases in productivity, • Achieved sustained utilization rate greater than 90%
collaboration and efficiency like we’ve seen with the • Saved over $6 million through in-place upgrades
cloud. Projects that used to take weeks can take days.
We believe that the AMD cloud represents a significant
business advantage.”
- Bob Luong, Director for Systems Engineering, AMD
14 | AMD’s Private Cloud | June 2012 | Link to AMD Cloud Case Study