AMD OpteronTM Processor for MP Server Systems Fred Weber VP & CTO, Computation Products Group AMD Agenda * AMD OpteronTM processor overview * Contrast to existing MP system topology * Glueless MP system topology Microprocessor Forum 2002 2 AMD OpteronTM Processor Technology Overview - - - - - - - Support for AMD's 64-bit technology 12-stage int, 17-stage fp pipelines Enhanced TLB structures TLB flush filter Enhanced branch prediction Large L2 cache (up to 1MB) ECC protection DRAM MCT SRQ * Processor Core Overview CPU * Memory Controller Overview XBAR HT Dual-channel DDR memory PC2700, PC2100, or PC1600 DDR memory support Registered or Unbuffered DIMMs ECC and Chip Kill High bandwidth (up to 5.3GB/s) HT - - - - - HT * HyperTransportTM Technology Overview - - - - One, two, or three links 2, 4, 8, 16, or 32-bits full duplex Up to 6.4 GB/s bandwidth per link 19.2 GB/s aggregate bandwidth HT = HyperTransportTM technology Microprocessor Forum 2002 3 AMD OpteronTM Processor Glueless MP System Overview MCT CPU SRQ DRAM SRQ DRAM MCT CPU non-Coherent cHT HT cHT cHT I/O XBAR cHT HT I/O XBAR I/O HT HyperTransportTM Link I/O cHT cHT CPU HT = HyperTransportTM technology SRQ XBAR cHT XBAR MCT DRAM Microprocessor Forum 2002 CPU SRQ I/O cHT HT Coherent HyperTransport TM MCT DRAM 4 Existing MP System Topology AMD CPU CPU Intel CPU 2.1 GB/s each 4.2 GB/s total CPU CPU CPU 3.2 GB/s total Gfx System Logic DRAM System Logic I/O I/O I/O System Logic 1.6 GB/s each 4.8 GB/s total DRAM 6.4 GB/s total 528 MB/s total System Logic 2.1 GB/s total 132 MB/s total I/O I/O Microprocessor Forum 2002 5 Local vs. Remote Memory Access 1-hop 0-hop 2-hops P0 P1 P0 P1 P0 P1 P2 P3 P2 P3 P2 P3 * 0 Hop: Local Memory Access * 1 Hop: Remote 1 Memory Access * 2 Hop: Remote 2 Memory Access * Diameter: maximum hop count between any pair of nodes * Average distance: average hop count between nodes Microprocessor Forum 2002 6 Local vs. Crossfire Memory Bandwidth Xfire BW Local BW P0 P1 P0 P1 P2 P3 P2 P3 * Local memory access bandwidth - Each processor reads data from its own local memory Microprocessor Forum 2002 * Xfire memory access bandwidth - All processors read data from memory at all nodes 7 Single Processor Population I/O I/O DDR DDR Mem Mem AMD AMDOpteronTM OpteronTM AMD OpteronTM AMD OpteronTM (Proc (Proc0) 0) I/O I/O System Parameters: * 8 DIMMs (up to 16 GB using 256Mb DRAM) * 2 HyperTransportTM links available for I/O * Processor-to-Memory Read Bandwidth = 5.3 GB/s * I/O Bandwidth = 6.4 GB/s (per link) Microprocessor Forum 2002 8 SPEC(R) CPU 2000 * System Configurations - AMD OpteronTM processor operating at 2.0GHz - Registered PC2700 DDR memory * SPECint(R) 2000: - Estimated base score = 1202 * SPECfp(R) 2000: - Estimated base score = 1170 Microprocessor Forum 2002 9 Dual Processor System Topology I/O I/O DDR DDR Mem Mem Bisection plane AMD AMDOpteronTM OpteronTM AMD OpteronTM AMD OpteronTM (Proc (Proc0) 0) I/O I/O AMD Opteron AMD Opteron AMD OpteronTM AMD OpteronTM (Proc (Proc1) 1) I/O I/O DDR DDR Mem Mem I/O I/O System Parameters: * 16 DIMMs (up to 32 GB using 256Mb DRAM) * 4 HyperTransportTM links available for I/O * Bisection-bandwidth = 6.4GB/s * Diameter = 1, Avg distance=0.5 * Local Memory Read Bandwidth = 10.67 GB/s - Local Bandwidth/processor = 5.3 GB/s * Xfire Memory Read Bandwidth = 7.06 GB/s - Xfire Bandwidth/processor = 3.53 GB/s Microprocessor Forum 2002 10 Quad Processor System Topology I/O I/O Bisection plane I/O I/O DDR DDR Mem Mem AMD OpteronTM AMDOpteronTM OpteronTM AMD AMD OpteronTM (Proc 0) (Proc 0) AMD Opteron AMD Opteron AMD OpteronTM AMD OpteronTM (Proc 1) (Proc 1) DDR DDR Mem Mem DDR DDR Mem Mem AMD Opteron AMD Opteron AMD OpteronTM AMD OpteronTM (Proc (Proc2) 2) AMD Opteron AMD Opteron AMD OpteronTM AMD OpteronTM (Proc (Proc3) 3) DDR DDR Mem Mem I/O I/O I/O I/O System Parameters: * 32 DIMMs (up to 64 GB using 256Mb DRAM) * 4 HyperTransportTM links available for I/O * Bisection-bandwidth = 12.8GB/s * Diameter = 2, Avg distance = 1 * Local Memory Read Bandwidth = 15.59 GB/s - Local Bandwidth/processor = 3.9 GB/s * Xfire Memory Read Bandwidth = 11.23 GB/s - Xfire Bandwidth/processor = 2.8 GB/s Microprocessor Forum 2002 11 MP System Scalability Memory Bandwidth Memory Bandwidth Scalability 18 16 14 GB/s 12 10 Local B/W Xfire B/W 8 6 4 2 0 1P 2P 4P Number of Processors in System Microprocessor Forum 2002 12 Summary * The AMD OpteronTM processor is designed to provide industry leading performance for enterprise class servers - 32-bit performance leadership substantiated by delivering on AMD's promise of nearly doubling x86-based SPEC(R) CPU performance from a year ago - Simultaneous 32 and 64-bit performance * AMD Opteron "plumbing" is designed to provide exceptional MP scalability - Performance advantage grows versus competitive platforms - Memory capacity and bandwidth scales - I/O capacity and bandwidth increases Microprocessor Forum 2002 13 Trademark Attribution AMD, the AMD Arrow Logo, AMD Opteron and combinations thereof are trademarks of Advanced Micro Devices, Inc. HyperTransport is a licensed trademark of the HyperTransport Consortium. Other product names used in this presentation are for identification purposes only and may be trademarks of their respective companies. SPEC, SPECint, and SPECfp are registered trademarks of the Standard Performance Evaluation Corporation (SPEC). Microprocessor Forum 2002 14