1 / 42

N ew CPU, new arch , KVM and commercial cloud

N ew CPU, new arch , KVM and commercial cloud. Michele Michelotto. The HEP WN for CPU farm. Two socket Rack mountable: 1U, 2U, dual twin, blade Multicore About 2GB per logical cpu x86-64 Intel or AMD. AMD roadmap. Steamroller moved to 2014?. Intel roadmap. AMD. Intel. HP Moonshot.

chalondra
Télécharger la présentation

N ew CPU, new arch , KVM and commercial cloud

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. New CPU, new arch, KVM and commercial cloud Michele Michelotto

  2. The HEP WN for CPU farm • Twosocket • Rack mountable: 1U, 2U, dual twin, blade • Multicore • About 2GB per logical cpu • x86-64 • Intel or AMD

  3. AMD roadmap • Steamroller moved to 2014?

  4. Intel roadmap

  5. AMD

  6. Intel

  7. HP Moonshot

  8. Intel S12x0 Centerton

  9. Intel

  10. The dual proc Xeon E5 26xx

  11. After E5 • E5-2600 version V2 • July September 2013 • Ivy Bridge core with 22nm Tri-gate fabrication process • Reduce TDP • Up to 10 ( or 12? ) cores with 25MB ( or 30? ) L3 cache • 15% - 20% clock update • DD3 memory at 1866 MHz

  12. Interlagos

  13. AMD 6272 2x16core 64GB at 2.1( up to 2.6) GHz

  14. Piledriver core • Substitute the “Bulldozer” core • Smarter prefetching • A perceptron branch predictor that supplements the primary BPU • Larger L1 TLB • Schedulers that free up tokens more quickly • Faster FP and integer dividers and SYSCALL/RET (kernel/System call instructions) • Faster Store-to-Load forwarding

  15. Abu Dhabi Family

  16. Configuration Software • Operating System: SL release 5.7 (Boron) • Compiler: gcc version 4.1.2 20080704 (Red Hat 4.1.2-51) • HEP-SPEC06 based on SPEC CPU 1.2 (32bit) • HEP-SPEC06 64 bit (default config + remove “–m32”) • 2GB per core unless explicitly stated

  17. AMD 6378 2x16core 64GB at 2.4GHz vs 6272 2.1 GHz – 32bit

  18. AMD 6378 2x16core 64GB at 2.4GHz vs 6272 2.1 GHz – 64bit

  19. Better architecture

  20. Intel vs AMD

  21. Intel Xeon E5

  22. CPU market Outlook • The Clock race has stopped since about several years • Essentially all improvements in throughput come from increase in the number of cores • I dumped all the results from SPEC CPU • I take Int Rate Baseline as a Proxy of HS06 • Keeping only X86-64 processor, dual socket

  23. Clock vs Time since Aug 2006

  24. Core count since Aug 2006

  25. SIRate2006 since Aug 2006

  26. Intel

  27. Power limitation for CPU processors

  28. Clock limitations From: The future of computing performance: Game Over or Next Level Ch4

  29. Quad core ARM Cortex A9 • Ultra compact size with full metal enclosure • Quad core ARM Cortex-A9 MPCore • 10/100Mbps Ethernet with RJ-45 LAN Jack • 2 x High speed USB2.0 Host port • Android 4.x & Ubuntu 12.10 • 89$ or for 233$ the full kit • 13.14 HepSpec06 • ARMv8/64 bit in 2014

  30. Thanks to Peter Elmer ACAT

  31. Elastic Cloud • European Project e-Fiscal • Sergio Andreozzi (EGI) • KashifIqbal (ICHEC, NUI Galway, Ireland) • HS06 is the main benchmark in EGI • How we compare it with Amazon Elastic Cloud (EC2) • Create Virtual Machines environment on SL • Map to EC2 type of computing node

  32. Xeon E5 2600 vs EC2

  33. Opteron 6272 vs EC2

  34. HS06 for Medium (SBridge) • SPEC score < with the > no. of VMs • Virtualisation + Multi-tenancy effect on performance ~ 3.28% to 58.48% • More realistic figure ~ 11.53 to 58.48 HS06 Score HPC/HTC vs. Cloud Benchmarking – eFiscal Workshop @ EGI TF 2012, Prague

  35. HS06 for Large (SBridge) • Virtualisation + MT effect on performance ~ 9.49% to 57.47% • Note the minimal effect of > no. of VMs HS06 Score HPC/HTC vs. Cloud Benchmarking – eFiscal Workshop @ EGI TF 2012, Prague

  36. HS06 for Xlarge (SBridge) • Virtualisation + MT effect on performance ~ 8.14% to 55.84% • Note the minimal effect of > no. of VMs HS06 Score HPC/HTC vs. Cloud Benchmarking – eFiscal Workshop @ EGI TF 2012, Prague

  37. HS06 for Medium (Opteron) • Virtualisation + MT effect on performance ~ 3.77% to 47.89% HS06 Score HPC/HTC vs. Cloud Benchmarking – eFiscal Workshop @ EGI TF 2012, Prague

  38. HS06 for Large (Opteron) • Virtualisation + MT effect on performance ~ 9.04% to 48.88% HS06 Score HPC/HTC vs. Cloud Benchmarking – eFiscal Workshop @ EGI TF 2012, Prague

  39. AMD 6272 2x16core 64GB at 2.1 GHz in 64bit mode

  40. New Intel Naming starting from E5 After Several generation of Xeon 5n xx 51xx (Woodcrest /Core 2c 65nm) 53xx (Clovertown / Core 4c 65nm) 54xx (Harpertown / Penryn 4c 45nm) 55xx (Gainestown / Nehalem 4c/8t 45nm) 56xx (aka Gulftown / Nehalem 6c/12t 45) Now Xeon E5 26xx “Sandy Bridge” EP 8c/16t ( @32 nm )

More Related