Xenon (CPU): Difference between revisions

From Xenon Wiki
Jump to navigation Jump to search
imported>Spamreplay
imported>Spamreplay
Line 42: Line 42:
*[http://www.ibm.com/developerworks/power/library/pa-fpfxbox/ Just like being there: Papers from the Fall Processor Forum 2005: Application-customized CPU design] at IBM (Explore the customized IBM PowerPC processor designed for the Microsoft XBox 360)
*[http://www.ibm.com/developerworks/power/library/pa-fpfxbox/ Just like being there: Papers from the Fall Processor Forum 2005: Application-customized CPU design] at IBM (Explore the customized IBM PowerPC processor designed for the Microsoft XBox 360)


*]http://forums.xbox-scene.com/index.php?showtopic=231928 Xenon hardware overview] by Pete Isensee, Development Lead, Xbox Advanced Technology Group, written some time before 23 June 2004
*[http://forums.xbox-scene.com/index.php?showtopic=231928 Xenon hardware overview] by Pete Isensee, Development Lead, Xbox Advanced Technology Group, written some time before 23 June 2004
*[http://arstechnica.com/old/content/2005/06/xbox360-2.ars Ars Technica explains the Xenon CPU]
*[http://arstechnica.com/old/content/2005/06/xbox360-2.ars Ars Technica explains the Xenon CPU]



Revision as of 13:25, 8 August 2010

Xenon is the CPU that used in the Xbox 360. The processor, internally codenamed Waternoose by IBM and XCPU by Microsoft. "The basic design is a 64-bit PowerPC architecture, with the complete PowerPC ISA available."

It is different from the Intel Pentium III in the original Xbox: the Xenon CPU is composed of three symmetrical cores on a single die, aach core has two symmetric hardware threads (SMT), for a total of six hardware threads available to games. Each individual core also includes 32 KiB of L1 instruction cache and 32 KiB of L1 data cache.

Specifications

  • The package is a 2-2-2 FC-PBGA, measuring 31mm by 31mm
  • 168 mm2 die size; one cpu core is about 28mm2
  • Support logic for Array and Logic Built in Selftest
  • 165 Million transistors
  • Three symmetrical cores running at 3.2 GHz each = 9.6Ghz throughput.
    • 32 KiB of L1 instruction cache and 32 KiB of L1 data cache per core
    • 64 bit CPU cores based on the IBM 970 "G5" design
    • Each core is two way SMT-capable and clocked at 3.2Ghz
    • 2× (128×128 bit) register files for each core
    • Two hardware threads per core; six hardware threads total
    • VMX-128 vector unit per core; three total
    • 128 VMX-128 registers per hardware thread
  • 1 MB L2 cache
    • (lockable by the GPU) running at half-speed (1.6 GHz) with a 256-bit bus
    • 51.2 gigabytes per second of L2 memory bandwidth (256 bit × 1600 MHz)
  • 21.6 GB/s front side bus (aggregated 10.8 GB/s upstream and downstream)
  • 115 GFLOPS theoretical peak performance
  • CPU Integer Perf (VPR): 1089 (363 each)
  • Dot product performance: 9.6 billion per second
  • IBM eFuse technology
  • IBM's PowerPC instruction set architecture
  • Restricted to In-order code execution
  • Big endian architecture

Linux Support

  • We have full SMP support, so we can use all three cores.
  • However, we currently need to disable the secondary threads because of a yet-to-be analyzed stability issue.
  • The CPU is quite slow on general purpose code. Due to the non out-of-order execution core, it heavily relies on the compiler to do proper optimizations. GCC currently doesn't know how to do this, resulting in running but very inefficient code.
  • The Cell people worked on PPU-GCC, from which the Xenon will benefit as well, as the PPC cores is quite similiar to the Cell's PPU. All PPU-GCC work have been merged on mainline GCC 4.4

External Links