Transcript Student Presentation 10

Chris Foster
Brian Moore
Scott Thibaudeau
Overview
•
•
•
•
I/O
EE – Emotion Engine!!
Graphics Synthesizer
Comparison
System Overview
RDRAM
Emotion
Engine
Graphics
Synthesizer
IO
Processor
Sound
Processor
Analog RGB
IEEE1394
USB (x2)
Digital RGB
Controller
SRAM
PCMCIA
Interface
OS
ROM
DVD
Digital L/R
Analog L/R
Controllers
• Two controller ports
– PS2 Dual Shock 2!
• Directional Pad
• Force Feedback
• 15 buttons
– All but Start/Select
are ANALOG!
Memory Cards/Ports
• 8 Mb storage capacity
• Transfer rate 250 times
that of PSX memory
cards
• Incorporates
“MagicGate”
DVD/CD-ROM
• Supports DVDs, Audio CDs, PSX CDs, PS2
DVDs
• 4.7 Proprietary DVD
• 4x DVD-ROM
• 24x CD-ROM
• DVD video playback
Built into hardware
Miscellaneous Ports
• IEEE1394 Standard – i.LinkTM firewire
• Expansion Unit for network interface
• One Type III PCMCIA card slot
– Flash Memory
– Small Hard Drives
• 2 USB Ports
Input/Output Processor
•
•
•
•
•
•
Handles all input and output
PSX Core Processor
33.8688MHz or 36.864MHz (Selectable)
2 MB Memory
32-bit Bus
Interfaces with i.LinkTM, USB, Controller
and Memory ports
The Emotion Engine Designed By Toshiba
Combination CPU and DSP Processor
Simulates Massive 3D worlds
The Emotion Engine
Geometry Calculations
Physics calculations
AI
Data Transfers
•MIPS IV CPU core
•Two Vector Processing Units
•Floating-point Coprocessor (FPU)
•Image Processing Unit (MPEG2 decoder)
•10-channel DMA controller
•Graphics Interface Unit (GIF)
•RDRAM interface and I/O interface
Geometry
+ Perspective Transformation 66 Million Polygons/sec
+ Lighting 38 Million Polygons/sec
+ Fog 36 Million Polygons/sec
Curved Surface Generation (Bezier) 16Million Polygons/sec
Image Processing Unit MPEG2 Macroblock Layer Decoder
Image Processing Performance 150Million Pixels/sec
DSP Features
Most Important DSP Feature Today: the MAC
Vector Calculations: Dot Product
Summing of Products
x0*x1 + y0*y1 + z0*z1
PS2 has 10 FMACs
Itanium Contains Only 4!
Each FMAC Computes One
32-bit Floating-point MAC Operation per Clock Cycle
DSP Processor Requirement 2
Memory Bandwidth
Memory Availability
Solution
Increase the Pipe while
Decreasing the Bucket
New Caching Method
Static Application
Dynamic Media App
Static Application
Dynamic Media App
High Degree of
Instruction-level Parallelism
High Degree of
Data Parallelism
PS2 has Moved to SIMD
(Single Instruction stream,
Multiple Data streams)
Caching on the PS2
DMAC Coordinates
Transfers on the 128
and 64-bit Data Paths
L1 Cache: 16K Ins.
32K Data
SPRAM: 16K
VU0 Cache: 32K
VU1 Cache: 32K
PIII has 3X the Amount
of Cache as the PS2
Busting Out All Over
VU Caches Too Small
to Hold Models
PS2 Code & Data
Constantly Streamed
Over Wide Internal Buses
Results in 10 Channel
DMAC Being Used
at Near Capacity
Two Independent Vector Processing Units
CPU + FPU: Basic Program Control and Housekeeping
CPU + FPU + VU0: Behavior Synthesis and Physics Calculations
VU1: Generates Display Lists
IPU: Image Decompression
FPU and VU0:
Coprocessors for CPU Core
GS and VU1:
Tied Together with GIF
Parallel Connection
Both Groups Sending Display Lists to GIF
Main
Memory
VPU0
SPRAM
CPU
VPU1
Rendering
Engine
Serial Connection
CPU/FPU/VU0 Acting as a Preprocessor for VU1
VPU0
SPRAM
CPU
VPU1
Rendering
Engine
CPU core 128bit RISC
(MIPS IV-subset)
Data Cache 8KB (2-way)
Scratch Pad RAM 16KB (Dual port)
Clock Frequency 300MHz
Integer Unit 64bit
(2-way Superscalar)
Main Memory 32MB
(Direct RDRAM 2ch@800MHz)
Memory bandwidth 3.2GB/sec
Multimedia extended instructions
107 instructions at 128bit width
DMA 10 channels
Integer General Purpose Register 32
at 128 bit width
Co-processor1 FPU
(FMAC x 1, FDIV x 1)
TLB 48 double entries
Co-processor2 VU0
(FMAC x 4, FDIV x 1)
Instruction Cache 16KB (2-way)
Floating Point Performance
6.2GFLOPS
Vector Processing Unit VU1
(FMAC x 5, FDIV x 2)
Gate Width: 0.18 micron
VDD Voltage: 1.8 V
Power Consumption: 15 Watts
Metal Layers: 4
Total Transistors: 10.5 Million
Die Size: 240 mm2
Package: 540pin PBGA
Sound Processor
Specifications At A Glance:
•Sound "SPU2+CPU"
•Number of Voices ADPCM: 48ch on SPU2 plus definable,
software programmable voices
(Adaptive Differential Pulse Code Modulation)
•Sound Memory 2MB
•Output Frequency Variable up to 48 KHz (DAT quality)
“Speech processing
is important, sir!”
Graphics Synthesizer
Specifications At A Glance:
ON DIE MAIN MEMORY!
•Clock Frequency 147.456MHz
•DRAM Bus bandwidth *Very High Throughput*
•Pixel Configuration RGB:Alpha:Z Buffer (24:8:32)
•Polygon Drawing Rate 75 Million Polygons Per Second
•Screen Resolution Variable from 256x224 to 1280x1024
“With Z-buffering, textures, lighting and
alpha blending (transparency), a
sustained rate of 20 Million polygons per
second can be drawn continuously.”
-Ars Technica
Redundancy for smoothness/error correction
4MB On Die
VDRAM
•VRAM is accessible from on die register files
•VRAM itself is on die
•VDRAM bus? Register to Main memory all on die?
•Allows for fast data transfer: 48GB/s
•2,560 bit wide VDRAM bus!!!!
Polygon Drawing Rate 75 Million /sec (small polygon)
50 Million /sec (48 Pixel quad with Z and A)
30 Million /sec (50 Pixel triangle with Z and A)
25 Million /sec (48 Pixel quad with Z, A and T)
Process Modes
Rendering Engine
Visualization
-3D Polygon (triangle, quad, mesh)
-2D Sprite
-2D/3D Line
-Particle
-Point
-Photo/Movie (MPEG-2)
Sony optimizes their instruction set
to speed up these specific operations
Inside the Beast
•No. of Pixel Engines 16 (in Parallel)
•Display output NTSC/PAL
•Silicon process technology 0.25 µ 4-level metal
•Total number of transistors 43 Million
•Die size 279mm2
•Embedded DRAM 4 MB of multi-port DRAM
Input: Highly Parallel
Pixel Proccess Instructions
PE PE PE PE PE PE PE PE
n=2,560
PE PE PE PE PE PE PE PE
Output: Geometry
Engine
Talking Shop:
Pixel Configuration
Pixel Configuration:
64-Bit (RGB, Alpha, Z-Buffer (24,8,32)
24-Bits are used for RGB colors, 8-Bits
are used for transparency effects, and 32Bits are saved for the Z-Buffer which
keeps track of everything in the 3D
environment.)
Geometry engine:
•Alpha channel
•Anti-aliasing
•Bezier surfacing
•Gouraud shading
•Mip mapping
•Z-buffer
Talking Shop:
GEOMETRY ENGINE – Alpha Channel/Aliasing
Alpha Channel/Aliasing:
Alpha channel
Transparency effects are created using
high density dark gray Alpha bits.
Anti-aliasing
Aliasing: The jagged “stair steps” that
occur when images are painted from
pixels in straight lines.
Anti-aliasing: Drawing gray pixels
around the lines of an image,
“blurring” the lines, minimizes
the stair steps and makes
an object appear more realistic.
Talking Shop:
GEOMETRY ENGINE – Bezier Surfacing/Gouraud Shading
Bezier Surfacing/Gouraud Shading:
Bezier Surfacing
A DSP that, approximates a 3D real-world object
Into a 2D polygon structure. This involves
*pixel  triangle  polygon  3d-object*
transformation.
Gouraud Shading:
This is a technique (also DSP) that applies shading
For smoothness to a Bezier Surface 3d-object.
Talking Shop:
GEOMETRY ENGINE – Mip Mapping
Mip Mapping:
When it doesn’t look the same up close as it does
from far away, you’re using MIP Mapping!
Compare and Believe:
GAMECUBE:
Polygons per second: 12 million polygons.
PLAYSTATION2:
Polygons per second: 75 million polygons.
XBOX:
Did we mention that
You can run Linux on a PS2?
Polygons per second: 45 million polygons.
•Open Market – January
•Linux Beta – This year
•Linux Emulation
for Cluster Parallel Proc. –
*Now*
Mega Polys per Second
Polygons Per Second v. System
100
80
60
40
20
0
CUBE
PS2
XBOX
System
http://ece.eng.rowan.edu/~fost1582/PS2