Systems and Technology Group
Power Efficient Processor Design and the Cell Processor H. Peter Hofstee, Ph. D.
[email protected] Architect, Cell Synergistic Processor Element IBM Systems and Technology Group Austin, Texas
© 2005 IBM Corporation
Systems and Technology Group
Agenda Power Efficient Processor Architecture System Trends Cell Processor Overview
2
© 2005 IBM Corporation
Systems and Technology Group
Power Efficient Architecture
3
© 2005 IBM Corporation
Systems and Technology Group
Limiters to Processor Performance Power wall Memory wall Frequency wall
4
© 2005 IBM Corporation
Systems and Technology Group
Power Wall (Voltage Wall) 1000 – Active power – Passive power • Gate leakage • Sub-threshold leakage (sourcedrain leakage)
Power Density (W/cm2)
Power components:
100
Active Power
10
Passive Power 1 0.1 0.01 1994
10S Tox=11A Gate Stack
2004
0.001 1
0.1
0.01
Gate Length (microns)
NET: INCREASING PERFORMANCE REQUIRES INCREASING EFFICIENCY Gate dielectric approaching a fundamental limit (a few atomic layers) 5
© 2005 IBM Corporation
Systems and Technology Group
Memory wall Main memory now nearly 1000 cycles from the processor
– Situation worse with (on-chip) SMP Memory latency penalties drive inefficiency in the design
– Expensive and sophisticated hardware to try and deal with it – Programmers that try to gain control of cache content, but are hindered by the hardware mechanisms Latency induced bandwidth limitations
– Much of the bandwidth to memory in systems can only be used speculatively – Diminishing returns from added bandwidth on traditional systems
6
© 2005 IBM Corporation
Systems and Technology Group
Frequency wall Increasing frequencies and deeper pipelines have reached diminishing returns on performance Returns negative if power is taken into account Results of studies depend on issue width of processor – The wider the processor the slower it wants to be – Simultaneous Multithreading helps to use issue slots efficiently
Results depend on number of architected registers and workload – More registers tolerate deeper pipeline – Fewer random branches in application tolerates deeper pipelines 7
© 2005 IBM Corporation
Systems and Technology Group
Microprocessor Efficiency Recent History: –Gelsinger’s law • 1.4x more performance for 2x more transistors
–Hofstee’s corollary • 1/1.4x efficiency loss in every generation • Examples: Cache size, OoO, Superscalar, etc. etc.
Re-examine microarchitecture with performance per transistor as metric –Pipelining is last clear win 8
© 2005 IBM Corporation
Systems and Technology Group
Attacking the Performance Walls Multi-Core Non-Homogeneous Architecture – Control Plane vs. Data Plane processors – Attacks Power Wall 3-level Model of Memory – Main Memory, Local Store, Registers – Attacks Memory Wall Large Shared Register File & SW Controlled Branching – Allows deeper pipelines (11FO4 … helps power!) – Attacks Frequency Wall 9
© 2005 IBM Corporation
Systems and Technology Group
System Trends
10
© 2005 IBM Corporation
Systems and Technology Group
System Trends toward Integration Memory
Northbridge Memory
Accel
Processor
Cell Processor IO
Southbridge
IO
Increased integration is driving processors to take on many
functions typically associated with systems
– Integration forces processor developers to address offload and acceleration in the design of the processor – Integration of bridge chip functionality Virtualization technology is used to support non-
homogeneous environments
11
© 2005 IBM Corporation
Systems and Technology Group
Next Generation Processors address Programming Complexity and Trend Towards Programmable Offload Engines with a Simpler System Alternative
GPU NIC CPU
Security
… Media
Hardwired Function
12
CPU
Streaming Graphics Processor
64b Power Processor
Network Processor
Synergistic Processor
Security Processor
…
Media Processor
Programmable ASIC
Mem. Contr.
.. .
Config. Synergistic IO Processor
Cell
© 2005 IBM Corporation
Systems and Technology Group
“Outward Facing” Aspects of Cell Cell is designed to be responsive .. to human user – Real-time response – Supports rich visual interfaces
.. to network – Flexible, can support new standards – High-bandwidth – Content protection, privacy & security
Contrast to traditional processors which evolved from “batch processing” mentality (inward focused). 13
© 2005 IBM Corporation
Systems and Technology Group
Cell Overview
14
© 2005 IBM Corporation
Systems and Technology Group
Key Attributes of Cell
Cell is Multi-Core – Contains 64-bit Power Architecture TM – Contains 8 Synergistic Processor Elements (SPE)
Cell is a Flexible Architecture – Multi-OS support (including Linux) with Virtualization technology – Path for OS, legacy apps, and software development
Cell is a Broadband Architecture – SPE is RISC architecture with SIMD organization and Local Store – 128+ concurrent transactions to memory per processor
Cell is a Real-Time Architecture – Resource allocation (for Bandwidth Measurement) – Locking Caches (via Replacement Management Tables)
Cell is a Security Enabled Architecture – SPE dynamically reconfigurable as secure processors
15
© 2005 IBM Corporation
Systems and Technology Group
Cell Chip Block Diagram
SPU SPE
SXU
SXU
SXU
SXU
SXU
SXU
SXU
SXU
LS
LS
LS
LS
LS
LS
LS
LS
SMF
SMF
SMF
SMF
SMF
SMF
SMF
SMF
EIB (up to 96 Bytes/cycle)
L2
PPE
MIC Dual XDRTM
L1 16
BIC FlexIOTM
PXU © 2005 IBM Corporation
Systems and Technology Group
Cell Prototype Die (Pham et al, ISSCC 2005) S P U M I C
P P U
S P U
S P U R B R I A C C
MIB
S P U
17
S P U
S P U
S P U
S P U
© 2005 IBM Corporation
Systems and Technology Group
Cell Highlights Observed clock speed – > 4 GHz
Peak performance (single precision) – > 256 GFlops
Peak performance (double precision) – >26 GFlops
18
Area
221 mm2
Technology
90nm SOI
Total # of transistors
234M © 2005 IBM Corporation
Systems and Technology Group
Element Interconnect Bus EIB data ring for internal communication
– Four 16 byte data rings, supporting multiple transfers – 96B/cycle peak bandwidth – Over 100 outstanding requests
19
© 2005 IBM Corporation
Systems and Technology Group
Power Processor Element PPE handles operating system and control tasks
– 64-bit Power ArchitectureTM with VMX – In-order, 2-way hardware Multi-threading – Coherent Load/Store with 32KB I & D L1 and 512KB L2
20
© 2005 IBM Corporation
Systems and Technology Group
Synergistic Processor Element SPE provides computational performance
– – – –
21
Dual issue, up to 16-way 128-bit SIMD Dedicated resources: 128 128-bit RF, 256KB Local Store Each can be dynamically configured to protect resources Dedicated DMA engine: Up to 16 outstanding request
© 2005 IBM Corporation
Systems and Technology Group
User-mode architecture
SPE Highlights
– No translation/protection within SPU – DMA is full Power Arch protect/x-late LS SFP
Direct programmer control
DP
SPU LS
FXU EVN
– Branch hint
VMX-like SIMD dataflow
FWD
GPR
– Broad set of operations
LS
CONTROL
FXU ODD
– Graphics SP-Float – IEEE DP-Float (BlueGene-like)
LS
Unified register file
CHANNEL
SBI
SMM
BEB
DMA
– DMA/DMA-list
ATO RTB
14.5mm2 (90nm SOI)
SMF
– 128 entry x 128 bit
256kB Local Store – Combined I & D – 16B/cycle L/S bandwidth – 128B/cycle DMA bandwidth
22
© 2005 IBM Corporation
Systems and Technology Group
SPE Organization (Flachs et al, ISSCC 2005)
23
© 2005 IBM Corporation
Systems and Technology Group
SPE PIPELINE (Flachs et al, ISSCC 2005)
24
© 2005 IBM Corporation
Systems and Technology Group
I/O and Memory Interfaces I/O Provides wide bandwidth
– – – –
25
Dual XDRTM controller (25.6GB/s @ 3.2Gbps) Two configurable interfaces (76.8GB/s @6.4Gbps) Flexible Bandwidth between interfaces Allows for multiple system configurations
© 2005 IBM Corporation
Systems and Technology Group
Cell Processor Can Support Many Systems Game console systems
XDRtm
Workstations (CPBW)
XDRtm
XDRtm
XDRtm
HDTV
XDRtm
XDRtm IOIF
XDRtm XDRtm
XDRtm XDRtm
CELL Processor
CELL Processor
BIF
XDRtm XDRtm
XDRtm XDRtm
IOIF 26
IOIF1
CELL Processor
IOIF0
SW
CELL Processor
CELL Processor
BIF
CELL Processor
IOIF
IOIF IOIF
IOIF
Supercomputers
CELL Processor
BIF
Home media servers
© 2005 IBM Corporation
Systems and Technology Group
Cell Processor Based Workstation (CPBW) (Sony Group and IBM)
First Prototype “Powered On” 16 Tera-flops in a rack (est.) – ( equals 1 Peta-flop in 64 racks )
Optimized for Digital Content Creation, including • • • •
27
Computer entertainment Movies Real-time rendering Physics simulation
CELL Processor Board High BW Sys Storage Networks Mgmt I/O Bridge CELL Processor (2-Way SMP)
Memory
16 TFlop rack
…
© 2005 IBM Corporation
Systems and Technology Group
Cell Processor Example Application Areas Cell is a processor that excels at processing of rich media content in
the context of broad connectivity – Digital content creation (games and movies) – Game playing and game serving – Distribution of (dynamic, media rich) content – Imaging and image processing – Image analysis (e.g. video surveillance) – Next-generation physics-based visualization – Video conferencing (3D?) – Streaming applications (codecs etc.) – Physical simulation & science 28
© 2005 IBM Corporation
Systems and Technology Group
Summary Cell ushers in a new era of leading edge processors optimized for digital media and entertainment Desire for realism is driving a convergence between supercomputing and entertainment New levels of performance and power efficiency beyond what is achieved by PC processors Responsiveness to the human user and the network are key drivers for Cell Cell will enable entirely new classes of applications, even beyond those we contemplate today 29
© 2005 IBM Corporation
Systems and Technology Group
Acknowledgements Cell is the result of a deep partnership between SCEI/Sony, Toshiba, and IBM Cell represents the work of more than 400 people starting in 2001 More detailed papers on the Cell implementation and the SPE micro-architecture can be found in the ISSCC 2005 proceedings
30
© 2005 IBM Corporation