SDC Server Sao Jose
- 1. Roberto Brandão AMD Brasil AMD Technology Update - Servers and Stream Processors -
- 2. Agenda The Green IT challenge Stream Processors Project Shanghai AMD Opteron on the work CPU Roadmap
- 3. AMD computing products Notebooks 1 and 2 cores Platform certification Servers 2 and 4 cores Up to 8 processors FireStream accelerators Desktops Up to 4 cores Native multi-core Infrastructure Chipsets Graphics cards
- 4. The Green IT Challenge
- 5. The Green IT Challenge
- 6. Performance por watt US datacenters consume more power than five 1000 megawatt nuclear power plants – at a cost of almost $3 billion This is 150% more than the consumption in 2001
- 7. AMD CoolCore™ Technology Turns off Blocks of Logic When Not in Use Coarse Control (Core) Ex, FPU (hottest part of die) Memory Controller Reads (turn off write logic) Writes (turn off read logic) Fine Control (Core) Incrementally Smaller Sections
- 8. AMD Opteron
- 9. 8-core system comparison AMD FSB FSB Hyper Transport CS MEMORY MEMORY MEMORY Twice the number of memory controllers Moving data affected by FSB Bottleneck
- 10. HPC Performance Advantage 100% = Xeon E5472 13% 19% The results stated above reflect results published on www.spec.org and www.amd.com as of June 23, 2008. The STREAM, FLUENT, STAR-CD, and LS-DYNA comparisons above are based on results measured in AMD’s performance lab. For the latest results visit http://www.spec.org/ and http://www.amd.com . 17% Average Performance Advantage for Opteron 2356
- 11. The Clear Leader for HPC Efficiency Based on measurements at AMD performance labs as of June 23, 2008. FLUENT results based on FLUENT 12 beta release from ANSYS, Inc. Performance on the final software release may vary. LS-DYNA results based on LS-DYNA mpp970 release from LSTC. STAR-CD results based on STAR-CD v4.02 release from CD-adapco. Higher performance and lower power consumption – the CLEAR leader Benchmark Rating (higher is better) Energy Usage (lower is better) Rating per Energy Usage (higher is better) FLUENT (CFD) sedan_4m, 8 cores 133% 86.7% 154% LS-DYNA (FEA) 3 Vehicle Collision, 8 cores 114% 99.6% 115% STAR-CD (CFD) A_Class_6m, 8 cores 125% 93.9% 133% Relative to Xeon E5472 processor-based server
- 12. Introducing “Shanghai” The new price/performance-per-watt leader for x86 servers. 1 Immersion lithography contributes to major power and performance benefits 2 Still the industry’s most manageable and consistent server platform 3 New innovations result in superior virtualization and power management capabilities 4 Unprecedented levels of price/performance for businesses 5
- 13. AMD Technology Leadership Roadmap 180nm 130nm 90nm 65nm 45nm 32nm 2000 2002 2004 2006 2008 2010 Cu Interconnect Low-K dielectric Strained-Si Multi-Strain Transistor 130nm 2003 SOI 2003 AMD Opteron™ 2005 Dual-Core AMD Opteron™ 2007 Quad-Core AMD Opteron™
- 14. Previous-generation innovations that continue Independent Dynamic Core Technology AMD CoolCore™ Technology Dual Dynamic Power Management™ Low-Power DDR2 Memory … and these power-efficiency features AMD Wide Floating- Point Accelerator AMD Memory Optimizer Technology Dual Dynamic Power Management™ AMD Balanced Smart Cache These performance-enhancing features… Low-Power DDR2 Memory Dual Dynamic Power Management™ ~ 150% 100% Dual-Core* Quad-Core* *Stream bandwidth comparison of Second-Generation & Third-Generation AMD Opteron™ processor
- 15. Innovations new to this generation L3 Cache Index Disable 3 RAS Enhanced RVI AMD Smart Fetch 45nm with Immersion Lithography AMD Memory Optimizer Technology AMD Balanced Smart Cache HyperTransport ™ Technology 3.0 2 Performance DDR2-800 Enhanced Pre-fetching 2x Core Probe Bandwidth 1 Up to 21% CPU Power Savings Greater Frequency 1 8MB: 2x total cache 1 Faster World Switch 1 Superior Data Integrity Up to 17.6GB/s Compared to Quad-Core AMD Opteron™ (“Barcelona” vs. “Shanghai”) Available Q2’09 on all new product introduction Planned availability in 2009 for select operating systems (to be determined) 45nm with Immersion Lithography Power Efficiency ~35% Lower Idle Power 1 Virtualization Core 3 Core 4 Core 1 Core 2 L1 L2 Integrated Memory Controller L3 L2 L2 L2 L1 L1 L1 Core 3 Core 4 Core 1 Core 2 L1 L2 Integrated Memory Controller L3 L2 L2 L2 L1 L1 L1 HT3 CPU CPU Core 3 Core 4 Core 1 Core 2 L1 L2 Integrated Memory Controller L3 L2 L2 L2 L2 L1 L1 L1 L1 VM1 VM2 Virtual Memory 1 Virtual Memory 2 Physical Memory Performance 90nm 65nm 45nm Power
- 16. Record-setting Performance 4P Java Performance AMD Opteron™ Processor Model 8384 “ Shanghai” (2.7 GHz) Intel Xeon Processor Model X7460 “ Dunnington” (2,66 GHz) $2149 $2729 26% better performance for 2P servers with 29% lower processor price 24% better performance for 4P servers with 21% lower processor price * All results based on SPECjbb®2005. See backup slide 23 for system configuration and details. Pricing reflects 1kU tray pricing published on http://www.amd.com and http://www.intel.com as of October 16, 2008 and/or suggested “Shanghai” 1kU price at launch. 24 cores 16 cores Intel = *NEW* X5482 (3.2GHz / 1600 FSB) 2P SPECfp
- 18. Acceleration Concepts Reduction in Latency Add-in Chipset “ Socket Filler” Package Integration Core Integration Outside the system Closer to the cores Chipset PCI-E A c c e l e r a t o r CPU Accelerator AMD Opteron Socket AMD Opteron Socket Opteron Socket PCIe Accelerator HTX Accelerator Accelerator
- 19. Enhancing GPUs for Computation RV770 Architecture AMD FireStream SDK Open systems approach to drive broad customer adoption AMD Multicore CPUs AMD Runtime Compilers Libraries 3 rd Party Tools Brook+ Rapidmind ACML/Cobra Compute Abstraction Layer (CAL) AMD Stream Processors 750 MHz 800 GDDR5 3.6 Gbps 1.2 teraFLOPS 2GB 160W Compute Power Core Clock Speed Stream Processors Memory Type Memory Peak Board Power
- 21. AMD in 7 of the Top10
- 22. AMD native quad-core FSB HyperTransport AMD64 Cache L3 PCI-e PCI-e Chipset Chipset Memory controller Memory controller Hyper Transport Interface Core 1 Core 2 Core 1 Cache L2 CROSSBAR Core 3 Core 4 Cache L2 Cache L2 Core 2 Cache L2 Core 3 Cache L2 Core 4 Cache L2
- 23. Evolving to enable Fusion 2008 GPU – Crossbar 2006 GPU – Ring Bus
- 24. OpenCL Compliance Giving Programmers a Fully Open Cross-Platform Interface The OpenCL Working Group consists of major industry players like AMD, Apple, IBM, Intel and Nvidia The Working Group is on track to ratify and publish the specification later this year OpenCL compliant implementation will begin rolling out next year AMD is participating in the Khronos OpenCL Working Group to help define an industry standard for GPGPU programming
- 25. Ecosystem Model for ATI Stream Computing Open Standards: DirectX, OpenCL Gaming Science Productivity Entertainment CAL, Brook+ A complete solution based on: Open Standards SDK Usage Scenarios
- 26. AMD + Tyan + SDC
- 28. Trademark Attribution AMD, the AMD Arrow logo and combinations thereof are trademarks of Advanced Micro Devices, Inc. in the United States and/or other jurisdictions. Other names used in this presentation are for identification purposes only and may be trademarks of their respective owners. ©2008 Advanced Micro Devices, Inc. All rights reserved.