parallel processing with playstation3 lawrence kalisz

Parallel Processing with PlayStation3

Lawrence Kalisz

Topics

Cell Processor1. History2. Architecture

Parallel Programming1. Install Linux2. Examples

PS3 Cluster1. Applications2. Examples

PS3 Cell Processor: History

Created by Sony, Toshiba, and IBM (STI)

400 Engineers

½ Billion Dollars

PS3 Cell Processor: Architecture

PS3 Cell Processor: Architecture

PS3 Cell Processor: Architecture

Power Processing Element (PPE)

Synergistic Processing Element (SPE)

Element Interconnection Bus (EIB)

Memory System

Network Card & Graphics Card

Power Processor Element

PPE handles operating system and control tasks

● 64-bit Power Architecture with VMX● In-order, 2-way hardware simultaneous

multi-threading (SMT)● 32KB L1 cache (I & D) and 512KB L2

Synergistic Processing Element

Specialized high performance coreThree main components

1. SPU: Supplemental Processing Units 2. LS: local store memory3. MFC: memory flow control manages data in and

out of SPECan only access (load & store) data in the

SPE local store7 SPEs used for rendering, 1 SPE reserved

for image compression

SPE: Data IN and OUT Steps

SPU needs data1. SPU initiates MFC request for data2. MFC requests data from memory3. Data is copied to local store4. SPU can access data from local storeSPU operates on data then copies data

from local store back to memory in a similar process

SPE: Data IN and OUT Steps

Element Interconnect Bus

• Physically overlaps all processor elements• Central arbiter supports up to 3 concurrent transfers per ring

• 2 stage, dual round robin arbiter• Each port supports concurrent 16B in and 16B out data path

• Ring topology is transparent to element data interface• Each EIB Bus data port supports 25.6GBytes/sec each way

PS3 Cell: Parallel Programming

PS3 Cell: Parallel Programming

Current working Linux distros:1. Fedora Core 52. YellowDog 5.03. Gentoo PowerPC 64 edition4. Debian

OpenMPI (for use with cluster)

IBM’s CELL SDK

PS3 Cell: Parallel Programming

Cell performance ~10x better than GPU for media and other applications that can take advantage of its SIMD capability◦PPE performance is comparable to a traditional

GPU performance◦SPE performance mostly the same as, or better

than, a GPU with SIMD◦Performance scales with number of SPEs

PS3 Cell: Parallel Programming

Programming becomes exercise in partitioning, mapping (layout),routing (communication) and scheduling

PS3 Cell: Parallel Programming

AI Backgammon player

PS3 Cell: Parallel Programming

AI Backgammon player1M board evaluations in ~3 seconds (6 SPEs)Data parallel implementation, linear speedup

PS3 Cell: Parallel Programming

SPU programs are designed and written to work together but are compiled independently

Separate compiler and toolchain (ppu-gcc and spu-gcc)

Produces small ELF image for each program that can be embedded in PPU program

PS3 Cell: Parallel Programming

BLUE-STEEL

PS3 Cell: Parallel Programming

BLUE-STEELFull ray tracer running on each SPEData parallel implementation://www.youtube.com/watch?v=C3ARXUSKXAM&fe

ature=player_detailpage

PS3 Cell: Parallel Programming

BLUE-STEELA Solution to the rendering equation

◦ Triangle Rasterization– Fast – possible in real time on a single core– Inaccurate or tedious for global effects such as shadows,reflection, refraction, or global illumination

Ray Tracing– Slow – unless done on multiple cores– Accurate and natural shadows, reflection, and refraction

PS3 Cell: Parallel Programming

BLUE-STEELBuild a fast ray tracer from the ground up to take

advantage of multiple cores.– 6 accessible cores for rendering

PS3 Cell: Parallel Programming

Ray Tracing Shoot a ray through each pixel on the screen Check for intersections with each object in the scene Keep the closest intersection

PS3 Cell: Parallel Programming

Ray Tracing Shade each point according to the material of the object,

as well as the lights in the scene Cast rays for shadows, reflection, and refraction

PS3 Cell: Parallel Programming

BLUE-STEEL

PS3: Cluster Applications

Air Force

Folding@home

PS3 Gravity Grid

LACAL Student Cluster

References

http://groups.csail.mit.edu/cag/ps3/http://

impact.asu.edu/cse520fa07/lec19-PS3-cell-tutorial.pdf

http://www.youtube.com/watch?v=VxaLmS7XPiI

http://en.wikipedia.org/wiki/PlayStation_3_cluster

http://www.netlib.org/utk/people/JackDongarra/PAPERS/scop3.pdf

Any Questions ?