Introducing Next Generation Low Power Microarchitecture: Silvermont
Dadi Perlmutter Executive Vice President General Manager, Intel Architecture Group Chief Product Officer
Risk Factors
•Today’s presentations contain forward-looking statements. All statements made that are not historical facts are subject to a number of risks and uncertainties, and actual results may differ materially. Please refer to our most recent earnings release, Form 10-Q and 10-K filing available for more information on the risk factors that could cause actual results to differ.
•If we use any non-GAAP financial measures during the presentations, you will find on our website, intc.com, the required reconciliation to the most directly comparable GAAP financial measure.
Rev. 4/16/13 Legal Disclaimers
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance .
Intel, Intel Atom and the Intel logo are trademarks of Intel Corporation in the United States and other countries.
1 Based on the geometric mean of a variety of power and performance measurements across various benchmarks. Benchmarks included in this geomean are measurements on browsing benchmarks and workloads including SunSpider* and page load tests on Internet Explorer*, FireFox*, & Chrome*; Dhrystone*; EEMBC* workloads including CoreMark*; Android* workloads including CaffineMark*, AnTutu*, Linpack* and Quadrant* as well as measured estimates on SPECint* rate_base2000 & SPECfp* rate_base2000; on Silvermont preproduction systems compared to Atom processor Z2580. Individual results will vary. SPEC* CPU2000* is a retired benchmark. * Other names and brands may be claimed as the property of others.
2 Based on a geometric mean of the projected power and performance of SPECint* rate_base2000 on Silvermont compared to expected configurations of main ARM*-based mobile competitors using descriptions of the architectures; assumes similar configurations. Numbers may be subject to change once verified with the actual parts. Individual results will vary. SPEC* CPU2000* is a retired benchmark; results are estimates. * Other names and brands may be claimed as the property of others
3 Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance. Introducing Silvermont
Next Gen 22nm SoC Microarchitecture Process
~3X Higher ~5X Lower Performance1 Power1
1 Based on the geometric mean of a variety of power and performance measurements across various benchmarks. Benchmarks included in this geomean are measurements on browsing benchmarks and workloads including SunSpider* and page load tests on Internet Explorer*, FireFox*, & Chrome*; Dhrystone*; EEMBC* workloads including CoreMark*; Android* workloads including CaffineMark*, AnTutu*, Linpack* and Quadrant* as well as measured estimates on SPECint* rate_base2000 & SPECfp* rate_base2000; on Silvermont preproduction systems compared to Atom processor Z2580. Individual results will vary. SPEC* CPU2000* is a retired benchmark. * Other names and brands may be claimed as the property of others.
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Silvermont Best In Class Performance • Out of Order Execution engine • New multi-core and system fabric architecture – Capable of delivering 8 cores • Higher single thread performance • New IA instructions and core technologies
Best In Class Power & Power Efficiency • Wider dynamic power operating range • Enhanced power management • Fast standby entry/exit
Great Scalability Across Multiple Market Segments
Architecture & Microarchitecture Definitions
Example Analogy Engine Type Architecture Intel® 64 IA-32 Instruction Set Definition
Engine Schematic Microarchitecture Ivy Bridge Haswell Saltwell Silvermont HW Implementation for ISA
Crypto Audio Type of Vehicle SoC (System on a Chip) Storage CPU Haswell Video Cloverview Single die with integrated IPs Graphics Display The Constantly Evolving Intel Architecture Performance Power Efficiency Media Manageability Security
SSE Extensions Thermal Monitoring Multimedia Instructions Intel® Virtualization Intel® Anti-Theft AVX Instructions Technologies Intel® Quick Sync Video Technology Technology Idle States Intel® vPro Technology Intel® Trusted Execution AES-NI Intel® HD Graphics Intel® Smart Idle Intel® CIRA Technology Technology Smart Caches Intel® HD Audio Intel® SpeedStep® Intel® AMT Intel® Secure Key Turbo Boost Technology Intel® Clear Video HD Technology Intel® Remote PC Assist Intel® Run Secure Intel® Hyper-Threading Intel® Flexible Display Intel® Demand Based Technolgy Intel® Remote Wake Execute Disable Bit Intel® Matrix Storage Switching Intel® Insider™ ...... Intel® Solutions For All Segments
Intel® Atom™
Datacenter and HPC Ultrabook™ Tablets Phones Intelligent Systems
From TERAFLOPS to MILLIWATTS Intel® Solutions For All Segments
Intel® Atom™
Datacenter and HPC Ultrabook™ Tablets Phones Intelligent Systems
From TERAFLOPS to MILLIWATTS Enabling a Full Spectrum of Mobile Products
Ultrabooks™ Convertible Tablet Smartphone Mobility Spectrum
Best Performance and Outstanding Battery Life Best Visuals
Best Battery Life Thinnest and Lightest Form Factors Intel® Atom™ As Well as Datacenter and Intelligent Systems Solutions
Server Network Storage Embedded IVI
Highest Performance Best Performance/Watt Highest Density Designs Intel® Atom™ Lowest Power Silvermont Microarchitecture in Next Generation Intel Products
From DATACENTER to DEVICES
Avoton Rangeley Baytrail Merrifield TBA
Port of Choice
Enabling Exceptional Experiences
* Other names and brands may be claimed as the property of others. Committed Leadership Roadmap
45nm 32nm 22nm 14nm 14nm
Silvermont; The First in a Family of Yearly Refreshes Intel’s IDM Advantage
PROCESS TECHNOLOGY INTEL ARCHITECTURE PRODUCT DESIGN
Common Common Tools Goals
SOFTWARE MANUFACTURING PACKAGING Silvermont Technical Overview Belli Kuttanna Intel Fellow Intel Architecture Group Key Messages
• Next Generation Low Power Microarchitecture • Manufactured in custom Intel® 22nm SOC process technology • Present in a wide range of low power products from Phones to Servers • ~3x peak performance1 improvement or the same performance at ~5x lower power1 over the current generation Atom™ core • Leading performance and performance per watt efficiency2 • First in a family of cores that will be refreshed every year
1 Based on the geometric mean of a variety of power and performance measurements across various benchmarks. Benchmarks included in this geomean are measurements on browsing benchmarks and workloads including SunSpider* and page load tests on Internet Explorer*, FireFox*, & Chrome*; Dhrystone*; EEMBC* workloads including CoreMark*; Android* workloads including CaffineMark*, AnTutu*, Linpack* and Quadrant* as well as measured estimates on SPECint* rate_base2000 & SPECfp* rate_base2000; on Silvermont preproduction systems compared to Atom processor Z2580. Individual results will vary. SPEC* CPU2000* is a retired benchmark. * Other names and brands may be claimed as the property of others. 2 Based on a geometric mean of the projected power and performance of SPECint* rate_base2000 on Silvermont compared to expected configurations of main ARM*-based mobile competitors using descriptions of the architectures; assumes similar configurations. Numbers may be subject to change once verified with the actual parts. Individual results will vary. SPEC* CPU2000* is a retired benchmark; results are estimates. * Other names and brands may be claimed as the property of others. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Silvermont Highlights Full Advantage of Intel® Better Better Power 22nm SoC Process Performance Efficiency Technology
• Out of Order Execution engine • Wider dynamic operating range • 3-D Tri-gate transistors tuned for SoC products • New multi-core and system fabric • Enhanced active and idle power architecture management • Architecture and design co- • New IA instructions extensions optimized with the process (Core™ Westmere level)
• New security and virtualization technologies ~3X the Performance Or ~5X Lower Power1
1 Based on the geometric mean of a variety of power and performance measurements across various benchmarks. Benchmarks included in this geomean are measurements on browsing benchmarks and workloads including SunSpider* and page load tests on Internet Explorer*, FireFox*, & Chrome*; Dhrystone*; EEMBC* workloads including CoreMark*; Android* workloads including CaffineMark*, AnTutu*, Linpack* and Quadrant* as well as measured estimates on SPECint* rate_base2000 & SPECfp* rate_base2000; on Silvermont preproduction systems compared to Atom processor Z2580. Individual results will vary. SPEC* CPU2000* is a retired benchmark. * Other names and brands may be claimed as the property of others. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Agenda Agenda Silvermont Micro-Architecture BENEFITS FEATURES Out-of-Order Execution Pipeline High Performance Without Macro operation execution pipeline Sacrificing Power Efficiency Improved instruction latencies and throughput Smart pipeline resource management
Power and Performance Efficient Branch Processing Accurate branch predictors Improvements Fast recovery pipeline
Low Latency, high bandwidth caches Faster and More Efficient Out of order memory transactions Access to Memory Multiple advanced hardware prefetchers Balanced core and memory subsystems
Building upon Intel’s Strengths and Expertise In Defining Microarchitectures Multi-Core Support
• “Module” building block-based Module A Module B Core Core Core Core … expansion from 1 to 8 cores L2 Cache L2 Cache • “Module” contains: IDI IDI
Memory – Two cores System Agent DRAM Controller – Tightly coupled second-level cache (up to 1MB): very low latency, high bandwidth – Dedicated point-to-point interface (IDI) to SOC Fabric – Independent read, write channels – Higher bandwidth, Lower Latency, OOO transaction support • Per-core frequency and power management support Balanced Core and Memory Subsystem Design Evolving Architecture: New Instructions and Technologies Performance Virtualization Security
VM 1 VM 2
VMM
Intel® Core™2 64b ISA + Core™ Intel® Core™ Westmere AES-NI, New Westmere SSE4.1, SSE4.2, Intel® Secure Key Instructions POPCNT
Intel® VT-x2: Intel® OS Guard New Real Time Instruction Tracing Extended Page Tables Support for McAfee® DeepSAFE Technologies TSC Deadline Timer Virtual Processor IDs Unrestricted Guest Fully Compatible with the Breadth of IA Software Installed Base Agenda Process/Design/Architecture Co-optimization
22nm SoC process 3D Transistors that offer unprecedented improvement in performance and power Metal stack co-optimized with key SOC IPs for density and performance Multiple 22nm process versions optimized for specific products + Architecture for Low Power Completely redefined for efficiency and scalability
+ Design for Low Power Custom arrays and libraries optimized for power and performance Power-Aware design and automation = Wide Dynamic Range of Operation + High Efficiency Unique Intel Leadership in Process, Design and Architecture Burst Mode Improvements
• Prior Atom™ cores – Opportunistically exposed additional Power P-states based on available thermal Sharing headroom Between Cores • Silvermont enhancements – Burst frequency managed in hardware based on Thermal, Electrical and Power Delivery constraints Core0 Core1 GFX – Power sharing between CPU cores and SOC IPs (e.g. Graphics) is supported – Burst operating points can be dynamically adjusted
Maximize Performance within Platform Capabilities Burst Mode Improvements
• Prior Atom™ cores – Opportunistically exposed additional Power P-states based on available thermal Sharing headroom Between Cores • Silvermont enhancements – Burst frequency managed in hardware based on Thermal, Electrical and Power Delivery constraints Core0 Core1 GFX – Power sharing between CPU cores and SOC IPs (e.g. Graphics) is supported – Burst operating points can be dynamically adjusted
Maximize Performance within Platform Capabilities Burst Mode Improvements
• Prior Atom™ cores – Opportunistically exposed additional Power Power Sharing P-states based on available thermal Sharing Between Cores headroom Between Cores and GFX • Silvermont enhancements – Burst frequency managed in hardware based on Thermal, Electrical and Power Delivery constraints Core0 Core1 GFX Core0 Core1 GFX – Power sharing between CPU cores and SOC IPs (e.g. Graphics) is supported – Burst operating points can be dynamically adjusted
Maximize Performance within Platform Capabilities Burst Mode Improvements
• Prior Atom™ cores – Opportunistically exposed additional Power Power Sharing Dynamic P-states based on available thermal Sharing Between Cores Burst headroom Between Cores and GFX • Silvermont enhancements – Burst frequency managed in hardware based on Thermal, Electrical and Power Delivery constraints Core0 Core1 GFX Core0 Core1 GFX Core0 Core1 GFX – Power sharing between CPU cores and SOC IPs (e.g. Graphics) is supported – Burst operating points can be dynamically adjusted
Maximize Performance within Platform Capabilities Power C-States
C0 C1/C2 C6
Core Clock OFF OFF L1 Caches
Core Voltage off Core State Core
Wakeup Time active Power C-States
C0 C1/C2 C6 Improved C6
Core Clock OFF OFF Modes with Cache State L1 Caches Retention
Core Voltage off Core State Core Fast Standby Wakeup Time active Mode Entry/Exit Fine Grained Power Gating
L2 Cache partial retention off flush 3D Transistors PLL OFF OFF Enable Low Wakeup Time Leakage and Low
ModuleState active active Voltage Operation Power Silvermont’s Wide Dynamic Range of Operation
Comp - large
Silvermont
Core Power Core
Comp - small
Performance3
Wide Dynamic Range of Silvermont Makes It More Efficient Than Asymmetric Cores
3 Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Silvermont’s Wide Dynamic Range of Operation
Higher power Comp - large → Lower efficiency
Silvermont
High Burst frequencies at lower power Higher efficiency Long switching times → longer residencies
→ Lost performance & → higher performance Core Power Core Lower efficiency
Comp - small Fast, Seamless Complex Transitions switching Low frequencies + 3 algorithms Lower voltage Performance → Higher efficiency
Wide Dynamic Range of Silvermont Makes It More Efficient Than Asymmetric Cores
3 Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Agenda Tablet: Silvermont vs. Saltwell
Core Power and Performance1 4.0 3.5 Perf Perf Lower Perf Perf Lower 3.0 Peak-to-Peak Iso Power Power Peak-to-Peak Iso Power Power iso Perf iso Perf 2.5 2.0 1.5 1.0 0.5 0.0 STW SLM STW SLM STW SLM STW SLM STW SLM STW SLM 1C1T 1C1T 1C1T 1C1T 1C1T 1C1T 2C4T 4C4T 2C4T 4C4T 2C4T 4C4T peak to peak iso-power iso-perf peak to peak iso-power iso-perf Single Threaded Multi-Threaded
1 Based on the geometric mean of a variety of power and performance measurements across various benchmarks. Benchmarks included in this geomean are measurements on browsing benchmarks and workloads including SunSpider* and page load tests on Internet Explorer*, FireFox*, & Chrome*; Dhrystone*; EEMBC* workloads including CoreMark*; Android* workloads including CaffineMark*, AnTutu*, Linpack* and Quadrant* as well as measured estimates on SPECint* rate_base2000 & SPECfp* rate_base2000; on Silvermont preproduction systems compared to Atom processor Z2580. Individual results will vary. SPEC* CPU2000* is a retired benchmark. * Other names and brands may be claimed as the property of others. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Not All Cores Are Created Equal
DC Competition QC Competition DC Silvermont
Phone Power Target Core Power Core
Performance3 Dual-Core Silvermont is more efficient than Dual-Core Competition2 Efficient dual-cores outperform2 inefficient quad-cores under power constraints
2 Based on a geometric mean of the projected power and performance of SPECint* rate_base2000 on Silvermont compared to expected configurations of main ARM*-based mobile competitors using descriptions of the architectures; assumes similar configurations. Numbers may be subject to change once verified with the actual parts. Individual results will vary. SPEC* CPU2000* is a retired benchmark; results are estimates. * Other names and brands may be claimed as the property of others. 3 Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Not All Cores Are Created Equal
Dual Core Silvermont vs. Quad Core Competition2 Dual Core Silvermont vs. Quad Core Competition2 Silvermont Performance Speedup at 1W Core Power X Less Core Power at 1W Comp Performance 3.1x 2.1x 2.8x
1.5x 1.4x 1.6x
Comp1 Comp2 Comp3 Comp1 Comp2 Comp3 1.6x The Performance 2.4x Lower Power
2 Based on a geometric mean of the projected power and performance of SPECint* rate_base2000 on Silvermont compared to expected configurations of main ARM*-based mobile competitors using descriptions of the architectures; assumes similar configurations. Numbers may be subject to change once verified with the actual parts. Individual results will vary. SPEC* CPU2000* is a retired benchmark; results are estimates. * Other names and brands may be claimed as the property of others. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance . Iso core-count comparisons Competitive View – Tablets Silvermont vs Competition2 Silvermont vs Competition2 Silvermont Performance Speedup at 1.5W Core Power X Less Core Power at Peak Comp Performance 2.3X 2.3X 5.8X 5.2X 1.7X 1.6X
3.6X 3.0X
Comp1 Comp2 Comp3 Comp4 Comp1 Comp2 Comp3 Comp4 2.0x The Performance 4.3x Lower Power
2 Based on a geometric mean of the projected power and performance of SPECint* rate_base2000 on Silvermont compared to expected configurations of main ARM*-based mobile competitors using descriptions of the architectures; assumes similar configurations. Numbers may be subject to change once verified with the actual parts. Individual results will vary. SPEC* CPU2000* is a retired benchmark; results are estimates. * Other names and brands may be claimed as the property of others. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to: http://www.intel.com/performance .