Barcelona

From View

Jump to: navigation, search

[ Chip Multi Processor Watch ]

Contents

Purpose and Target Market

General purpose Server/Workstation/Desktop Processor.

Architecture

4 homogeneous next generation Opteron x86_64 cores.

Interconnect and Topologies

The chip has a crossbar switch between the shared L2 cache and the memory controllers and HyperTransport links.

Memory Structure and Hierarchy

Shared memory? Distributed memory?

  • 128 kB private L1 cache, split in 64 kB data and 64 kB instruction cache.
  • 512 kB private L2 cache
  • 2 MB shared L3 cache

I/O and Peripherals

Each chip has the following peripherals:

  • 2 64-bit DDR2 memory controllers (in stead of one 128 bits memory controller in K8)
  • 4 16-bit HyperTransport 3.0 lanes (3 for fully interconnected 4-way SMP and 1 to I/O) that can be split in in 8 8-bit lanes for fully interconnected 8-way SMP plus I/O access to each chip. HyperTransport speed will be initially 2.0 GT/s and in future may operate a 5.2 GT/s.


Physical Properties

Process Technology

Whose fab? which year? Fast or slow process? multi-Vt process?

11 metal layers

Die Size

Relative size of logic vs arrays? relative sizes of cores and others?

Die size: 291 mm^2

  • L3: ~20% of die area


Power Dissipation

Average or peak Watts? Joules per SPEC?


Usage Model

Program Model

Common programing languages for x86:

  • x86 Assembler
  • C/C++
  • Java
  • C#

...

Common parallel programming models for x86:

  • Pthreads
  • OpenMP
  • MPI
  • MapReduce

...

Software Development Environment

System design tool stack? Availability of layers in this tool stack?


Recent News and Publications

Personal tools