Evaluating New Technologies for Test and
Measurement: PCI Express, Multicore
Processing, and Microsoft Windows Vista
NIDays 2007
Worldwide Virtual Instrumentation
Conference
10,000
Max Bandwidth (MB/s)
Increasing (Improving) Bandwidth
Evaluating Test and Measurement Buses
Good
Better
Best
PCI Express (x4)
1,000
PCI/PXI (32/33)
Gigabit Ethernet
100
10
Hi-Speed USB
IEEE 1394a
Fast Ethernet
USB 1.1
1
10,000
1,000
VME/VXI
GPIB (HS488)
GPIB (488.1)
100
10
1
Approximate Latency (μs)
Decreasing (Improving) Latency
3
0.1
Increasing Bus Bandwidth Opens New Applications
24
Multichannel
Audio
Number of Bits
20
IF
Communications
16
Data Acquisition
12
HighResolution
Digitizers
Instrument
Control
8
High-Speed
Imaging
PCI
ISA
1M
10M
PCI Express
100M
1G
10G
Sample Rate (S/s)
4
100G
PCI Express Overview
•
Evolutionary version of PCI
– Uses same software model as PCI, ensuring compatibility
•
•
•
Inside every new PC and notebook today
Low cost – built into PC chipsets
Serial interconnect at 2.5 Gb/s
–
–
–
–
PCI transactions are packetized and then serialized
Low-voltage differential signaling, point-to-point, 8 B/10 B encoded
Bandwidth is dedicated PER slot and in BOTH directions
Multiple lanes can be grouped together to form links
• x1 (by 1) has bandwidth of 250 MB/s/direction
• x16 (by 16) has bandwidth of 4 GB/s/direction
•
•
Scalable interconnect – chip-to-chip, backplane, or cabled
Roadmap for longevity with Gen-2 clocking (5 Gb/s)
5
Dedicated Bandwidth per Device
Total System Bus Throughput
(MB/s)
3000
2500
Ie
PC
x4
2000
1500
1000
PCIe x1
500
PCI (32/33)
Gigabit Ethernet
0
1
3
2
Number of Devices
6
Software Layer
• PCI software-model compatible
– 100% OS and driver-level compatible
– PCI enumeration, configuration, and power management
mechanisms
– Existing operating systems boot with no changes (including BIOS)
• PCI Express hierarchy mapped using PCI elements
– Host bridges
– P2P bridges
– All enumerated using the regular PCI device configuration space
• PCI capability pointer for PCI Express-specific extensions
7
Physical Layer
Data
Clock
•
•
•
•
•
•
•
Sequence
Number
Packet
Request
CRC
Frame
Device B
Device A
Data
Frame
x1 Lane
Frame
CRC
Packet Sequence
Request Number
Frame
Data
Data
Clock
Point-to-point, differential interconnect with two endpoints
Low-voltage signaling, AC coupled
Two unidirectional links, no sideband signals
Bit rate: >2.5 Gb/s/pin/direction and beyond
Clocking: Embedded clock signaling using 8 B/10 B encoding
Link widths (per direction): x1, x2, x4, x8, x12, x16, x32
Gen-2 (5 Gb/s) speed increase
8
PCI Express and PCI Slots on a Motherboard
3 x1
PCI Express
Slots
1 x16
PCI Express
Slots
2
PCI Slots
9
PCI Express Cards
NI PCIe-GPIB
Instrument Control
(x1)
NI PCIe-1429
Image Acquisition
(x4)
PCI Express Graphics Card
(x16)
Examples of Different PCI Express Link Widths: x1, x4, and x16
10
Up-Plugging and Down-Plugging
Up-plugging: Installing boards in higher-lane slots
• Allowed by PCI Express
• Example: Plugging a x4 module in a x8 slot
• Caveat: Motherboard vendors are only required to support a x1
data rate in this configuration
– Full-bandwidth support will be vendor specific
– Example: x16 slots may operate as a x1, even for x4 cards
Down-plugging: Installing boards in lower-lane slots
• Physically prevented by the design of the slots and connectors
for the desktop form factor
• Allowed in PXI Express and CompactPCI Express
11
ExpressCard – PCI Express for Laptops
• Both x1 PCI Express and Hi-Speed USB signaling on host
• 34 mm and 54 mm form factors
• PXI embedded controllers include ExpressCard/34 slot
12
PCI Express Industry Adoption
• First PCI Express desktops shipped mid 2004
• First ExpressCard laptops shipped early 2005
• PCI and PCI Express are side-by-side in all
Intel/Dell roadmaps
• Primary consumer driver is graphics processing
(gamers, video editing)
– PCI Express x16 replacing AGP
13
National Instruments Shipping Products
•
•
•
•
•
•
•
•
•
•
NI PCIe-GPIB (x1)
NI PCIe-6251 M Series (x1)
NI PCIe-6259 M Series (x1)
NI PCIe-1429 Camera Link (x4)
NI PCIe-1430 Camera Link (x4)
NI PCIe-8361 MXI-Express (x1)
NI PCIe-8362 MXI-Express (x1)
NI PCIe-8371 MXI-Express (x4)
NI PCIe-8372 MXI-Express (x4)
NI ExpressCard-8360 MXI-Express
14
PCI Express Advantages
•
•
•
•
•
•
•
•
Software compatibility with PCI
High bandwidth (up to >4 GB/s)
Scalable bandwidth
Dedicated bandwidth per slot
Low latency
Peer-to-peer communication
Internal and external operation
Long life (20+ years in the mainstream market)
15
PXI Express – Integrating PCI Express into the
PXI Backplane
• Up to 6 GB/s backplane and 2 GB/s slot
bandwidth
• Backward compatibility
– Complete software compatibility
– Hybrid slot definition – install modules with either
PCI or PCI Express signaling in a single slot
• Enhanced synchronization capabilities
– 100 MHz differential clock, differential triggering
16
PXI and Hybrid Slots Ensure Compatibility
17
PXI Slots
18
Hybrid Slots
19
PXI Express Hybrid Slots
• x8 PCIe
(up to 2 GB/s)
• Differential
Clk. 100 &
Star Triggers
• Power
• Trigger Bus
• Star Trigger
• Clk. 10
• Reserved Pins
• Local Bus
(typically unused)
32/33 PCI
(132 MB/s
per system)
PXI Express Hybrid
PXI
20
Hybrid Slot Flexibility
PXI Express
Peripheral Module
32-Bit CompactPCI
Module
Hybrid Slot Compatible
PXI Module
21
NI PXIe-1062Q Hybrid Chassis
Hybrid Slot Configuration
PXI:
2
PXI or PXIe:
3
PXIe Only:
4
Hybrid Slots
22
6
H
5
7
H
8
PXI-8105 Dual-Core Embedded Controller
•
•
•
•
•
•
•
•
•
Industry’s highest-performance embedded controller
Up to 100% higher performance for multithreaded apps
2.0 GHz dual-core Intel Core Duo processor T2500
Dual-channel 667 MHz DDR2 RAM
Gigabit Ethernet
ExpressCard/34 slot
4 Hi-Speed USB ports
60 GB SATA hard drive
DVI-I video
23
NI PXI-1033 Chassis with Integrated MXI
Express Controller
• 110 MB/s sustained throughput with MXI-Express remote control
• Rugged, compact package with slots for five peripheral modules
• Quiet acoustic noise emissions as low as 38 dBA
• Kit includes chassis with integrated controller, host card (PCI Express
or ExpressCard), and cable
24
PXI Express Video Demo – NIWeek 2006 Keynote
Click box to start video demo
25
What Is Multicore Processing?
•
Multicore processors contain two or more cores, or computing engines, in one
physical processor
•
•
Multicore processors simultaneously execute two or more computing tasks
Why Multicore?  Because of power and performance issues, continuing to
rely solely on increases in processor clock rates to improve performance is not
feasible
26
Multi-core Programming
“One Holy Grail of computer science research has been finding a
way to let a compiler take care of parallelization. “
- Richard Wirt, Intel Senior Fellow
C
LabVIEW
27
Multicore vs. Multiprocessor vs. Hyperthreaded
Multiprocessor
• Multiprocessor systems include two or more physical processors
• Multiprocessor systems duplicate computing resources that are
often shared in multicore systems (front-side bus, etc.)
• Multiprocessor systems are, most often, higher cost than similar
multicore systems (single processor, processor socket, etc.)
Hyperthreaded
• A hyperthreaded processor “acts like” two physical processors
• Certain resources are duplicated (register set, etc.), but the
execution unit is shared
• Hyperthreaded systems include multiple logical processors
28
Multitasking
• Multitasking environments (Windows XP, etc.) allow
multiple applications to run at the same time
• With a multicore processor, these multiple applications
can simultaneously execute on the processor cores
29
Multithreading
• Multithreaded applications separate their tasks
into independent threads
• A multicore processor can simultaneously
execute these threads
30
Demo
Multithreaded Application Executing
on a Dual-Core Processor
LabVIEW 8.0
Multi-Threaded VI
PXI-8105 LabVIEW Benchmarks
PXI-8105
LabVIEW 8.0
Single-Threaded VI
PXI-8196
100%
PXI-8105
PXI-8196
0
50
25%
100
150
200
250
32
The Future of Multicore Processing
• Architecture improvements to further reduce
power and improve memory bandwidth
• Multiprocessor systems with multicore
processors
• More processor cores
• Quad-core processors will release in 2007
33
Microsoft Windows Vista Overview
•
•
•
•
•
•
Visualization and Search
Security Changes
.NET 3.0 API
Vista x86 versus Vista x64
Vista Availability
Vista System Requirements
34
Graphics and Visualization
35
Vista x86 versus Vista x64
Vista x86 (32-Bit)
Vista x64 (64-Bit)
WoW Emulation
Executes in
User Mode
32-Bit
Application
32-Bit
Application
64-Bit
Application
Executes in
Kernel Mode
32-Bit
Service
or Driver
64-Bit
Service
or Driver
64-Bit
Service
or Driver
NI Software 2007
After 2007
36
Vista System Requirements
• Minimum (XP-like experience)
– 1 GHz “Modern” Processor
– 512 MB RAM
– DirectX 9 Video
• Premium (“Aero” experience)
– 1 GHz “Modern” Processor
– 1 GB RAM
– DirectX 9 Video with 128 MB VRAM
37
Vista-ready LabVIEW 8.2.1 released on Monday, April 9th
38