Dux Computer Works
http://duxcw.com/
The nVIDIA nForce Motherboard Chipset:
a different perspective
by Larry F. Byard
Last updated: 6/28/01
 Introduction. For
many years I have cautioned customers not to buy all-in-one motherboards
(see Advice
on Buying a Motherboard) or computers that use them. These are
motherboards with the display adapter and soundboard functions built into
the motherboard. Although, there have been attempts at establishing
industry standards, many all-in-one motherboards are manufacturer-specific
and computers built with them often cannot be upgraded with generic motherboards. Also,
in the event of failure, these motherboards can be costly to replace--if
one can find a replacement. Furthermore, many of them have been designed
for low-end, cheap computers. Some of them are poorly designed, lack
other features, such as sufficient expansion board slots, and cut corners
with the number and quality of parts used to make them.
In the past, all-in-one motherboards
were designed with separate chips and software drivers from multiple vendors. The
integration of these components was done by the motherboard designer and
has often been poor. Sound
functions, in particular, can add more unchangeable interrupts to a computer
and can spell problems even if the function is deactivated. I have
seen computers with blown integrated display adapters that could not be repaired
or jumpered-out. Often the add-on sound and display chips are from
the low-end of the manufacturers’ product lines--perhaps, so that the motherboard
will not compete their expansion board offerings. There are motherboards
with on-board video that hog main memory and memory bandwidth for functions
that would otherwise use video memory on a plug-in display adaptor, and very
noticeably slowing everything down.
Recently, several motherboard chipset
manufacturers have started to integrate the display and sound functions into
the motherboard chipset. The chipset manufacturer instead of the motherboard
designer started doing integration. However, these products are mainly
being produced by partnerships (or acquisitions of one company by another)
between motherboard chipset and display adapter manufacturers. Although,
this solved some of the problems with integration, the sound and display
adapter functions are generally still not nearly as good as those found with
expansion boards.
All
of this is about to change. Enter Microsoft. In 1999
Microsoft told Game developers that it was going to build the world's greatest
game machine, the Xbox. In March of 2000 Microsoft announced that
nVIDIA would design a customized version their high-performance GeForce3
graphics engine, eventually called the Xbox Graphics Processor Unit (XGPU),
for the Xbox. Later nVIDIA was chosen to design another principal
Xbox chip, the MCPX or Media Communications Processor. The MCPX integrates
typical motherboard Southbridge chip functions, such as the IDE disk controller
and USB, with networking and a high-end audio processing capability. The
$299 Xbox console will sport a Pentium3 733mhz CPU, 64MB of DDR memory,
an 8 Gbyte hard disk drive, a DVD drive, 3D Dolby Digital Surround audio,
Ethernet interface, etc. It should appear on store shelves this October. You
can read more about it on Microsoft’s
Xbox web site and the many other web sites on the Internet devoted
to the Xbox.
Last September nVIDIA announced that
it would use the integrated chipset development efforts for the Xbox as a
foundation to grow into the motherboard chipset business. This signaled
a fundamental change in motherboard chipsets and motherboards. Instead
of a chipset manufacturer designing-in the graphics engine from a graphics
board company, the graphics board company was going to design the entire
chipset. Graphics board designers do more than glue motherboard functions
and components together with a chipset, they are driven to compete in market
geared towards performance. And that’s what nVIDIA did: designed performance
into a chipset from the ground-up. They put some of the essence of
the Xbox in a PC architecture and introduced it as the nFORCE motherboard
chipset at Computex in Taipei, Taiwan on June 4th.
The
nFORCE chipset replaces the Northbridge and Southbridge chips comprising
the typical motherboard chipset (right) with the Integrated Graphics Processor
(IGP) and the Media and Communications Processor (MCP), respectively. The
chipset architecture and both chips differ significantly from their predecessors,
as will the motherboards built with them.
Here are feature highlights of each chip:
The IGP consists of about 20 million
transistors that are organized into three principle units:
Twin Bank Memory Architecture,
which is implemented through dual-independent, 64-bit memory controllers. It
provides a 266 MHz, 128-bit wide access path to Double Data Rate (DDR) memory. The
result is a 4.2 GBytes/sec. peak memory bandwidth. This compares
to 2.1 GBytes/sec. for other DDR chipsets. Towards the high end of
the PC graphics performance spectrum, the nVIDIA’s GeForce3 display adapter
has four 64-bit memory controllers and a 256-bit access path to its on board
DDR memory.
Dynamic
Adaptive Speculative Pre-processor (DASP). The DASP uses
intelligent, pre-processing technology to exploit unused memory bandwidth
to load its cache (1024 bytes) with application instructions and data the
CPU is expected to request later. When the CPU requests the data,
it is returned to the CPU immediately rather waiting for it to be accessed
from memory. Translation: the DASP is a smart memory cache’ system.
Graphics Processor Unit (GPU). The
GPU uses GeForce2 MX core. With the Twin Bank Memory Architecture it
is capable of a maximum graphics throughput of
350 megapixels/sec. and does not have the stellar imaging performance
of the sup’d-up GeForce3 GPU in the Xbox, but it is certainly faster than
many run-of-the-mill display adapters and the graphics cores currently integrated
into other Northbridge chips. The GPU does use system memory for its
frame buffer (32 Mbytes) and shares memory bandwidth with the rest of the
system like other chipsets and motherboards with built-in graphics, but this
is mitigated by Twin Bank Architecture’s 128-bit memory pipe, the use of
DDR system memory, and the current, price-driven transition of mainstream
PC’s from 128 to 256 MBytes of memory. Also, because of the integration
of the GPU with other units in the MCP, complex graphics calculations can
be performed directly in the GPU itself without having to communicate back
and forth across the AGP bus with the CPU. This frees-up the CPU for
other tasks.
The IGP has an external 4X/8X AGP bus
to accommodate higher-performance, plug-in display adapters. It also
has a TV encoder interface which multiplexes on the external AGP bus, supports
HDTV resolutions, and supports simultaneous TV and monitor displays.
The
MCP consists of about six million transistors. The MCP Audio
Processing Unit (APU) is a very significant improvement over the audio
functions integrated into other chipsets and is
essentially the same audio technology found in the Xbox. It
has Five Data Signal Processors (DSPs) with
a combined processing power of about one billion operations per second. According
to nVIDIA, it is two or three times more powerful than a Creative SoundBlaster
Live soundboard. It is Microsoft
DirectX 8.0 compliant, provides real-time processing of up to 256-simultaneous
stereo audio streams, or 64 3D and stereo (2D) streams, and can output
to 2, 4, or 6 speakers. It features a new Dolby Digital Interactive
Content Encoder, which does real time encoding for interactive 3D positional
audio and Dolby Digital 5.1 channel audio for the kind of sound you will
find in a movie theater when plugged into a home theater audio system.
In addition
to other Southbridge functions, such as two ATA/100 channels for disk drives
and an Audio
Codec (AC’97) 2.1 Compliant Interface to implement basic audio
and MODEM functions, the MCP incorporates an IEE 802.3 Media Address Controller
(MAC) for 10/100 Ethernet and Home PNA 2.0 phoneline networks. The
chip has support for six USB 1.1 ports, but no USB 2.0 or IEEE 1394 firewire
support. The ISA bus is history. The LPC (Low Pin Count) Bus replaces
it for connectivity to the keyboard controller, floppy disk controller,
and serial, parallel, game, and MIDI ports.
Bandwidth. Other
than the features already mentioned, what really sets this chipset apart
is connectivity and bandwidth. As mentioned above, it “talks” to
DDR memory through two 64-bit memory controllers. The IGP is connected
to AMD DDR CPUs with a 133 MHz Front Side Bus (FSB) capable of 64-byte
DDR266 burst transfers. Meaning that data is transferred on both
the leading and trailing edges of the FSB clock pulse for an effective
266 MHz rate. Finally and not least, instead of hanging the Southbridge
off the PCI bus, with a theoretical maximum burst speed of 133 MBytes/sec.,
like most chipsets (the other exception being the SiS 737 single-chip solution--heretofore
the winner in benchmarks I have seen comparing DDR Athlon chipsets--with
it’s Built-in 1.2GBytes/s Multi-threaded I/O Link between its internal
Southbridge and Northbridge sections), the MPC is connected to the IGP
with AMD's HyperTransport technology. This
implementation of HyperTransport is 8-bits wide and has a total peak throughput
of 800 MBytes/second. According to nVIDIA this is more than enough
bandwidth for all of the Southbridge I/O functions to operate simultaneously
without interruption.
Furthermore,
nVIDIA's StreamThru Networking/Broadband Architecture includes an advanced
isochronous transport system, which makes use of the HyperTransport and
features in the IGP and MCP to noticeably improve time-dependent applications
such as media streaming, file downloading and multi-player games. What's
that? Isochronous?
"A form of data transmission that guarantees
to provide a certain minimum data
rate, as required for time-dependent data such as video or audio. Isochronous
transmission transmits asynchronous data over a synchronous data link so
that individual characters are only separated by a whole number of bit-length
intervals. This is in contrast to asynchronous transmission,
in which the characters may be separated by arbitrary intervals, and with synchronous transmission."
Source: Free
On-Line Dictionary of Computing
What's it mean? Better networking
and music, and fewer, if any, buffer under runs when burning a CD.
So, even if
you are game fanatic with a GeForce3 plug-in display adapter, you may still
choose a motherboard with this chipset for it’s memory architecture and
Northbridge/Southbridge bandwidth, and sound capabilities. Others
will probably find that the IGP has all of the graphics power they need
or that it is sufficient to get them going until they are ready to upgrade
to a higher-end graphics capability.
The entire
chipset (graphics, sound, etc.) requires only one software driver from
one vendor.
Each of the nFORCE chips will come in
two flavors: IGPs with a single 64-bit memory controller or dual 64-bit controllers
and MPCs with or without the Dolby Digital 5.1 encoder. The nFORCE420
chipset will taste the best with a fully-equipped IGP and MCP. The
nFORCE220 will consist of the crippled chips, will cost about $10 less in
quantity than the 420, and may become a hallmark of El Cheapo, all-in-one-computers.
Larry
Please see our CONTACT page
if you have any comments or corrections that would make this article better. Please
use our Forums if
you need help with a computer or network problem.
IGP FEATURES
TwinBank Memory Architecture
- Integrated 128-bit Memory Controller
- Concurrent Simultaneous Access for IGP, MCP,
and CPU
- Supports 2.5V, 133/100MHz (266/200MHz) DDR
SDRAMs
- Supports 3.3V, 133/100 SDR SDRAMs
- Supports 64, 128, 256, and 512Mbit x8 and x16
DRAMs up to maximum 1.5GB
- Allows Different Speeds and Sizes Per DIMM
(Unbuffered, Non-ECC)
- Concurrent DRAM Write-back and Read-around
Write
- Software Configurable Timing and Configuration
Parameters
Dynamic Adaptive Speculative Pre-processor
(DASP)
- Enhances CPU Performance
- 8 Way Prediction of Memory Accesses
Integrated GeForce2 GPU
- High-Performance Two-Pixel Pipeline
- NVIDIA Shading Rasterizer (NSR)
- Integrated 2nd Generation T&L Engine
- 32-bit Color with 32-bit Z/Stencil
- Cube Environment Mapping
- 300MHz Palette – DAC
- External AGP4X/8X with Fast Write
HyperTransport™ Technology
- High Speed (800MB/sec.)
- Low Voltage
- Differential
- Low Pin Count Interface
- Isochronous Link between IGP and MCP
CPU Interface
- Supports AMD Athlon/Duron CPU
- 133/100/66 MHz FSB Clock
- DDR266/DDR200 Support 64-byte (Cache Line)
Data Burst Transfers
TV Encoder Interface
- Multiplexed with AGP Interface
- Clock Speeds Up to 80MHz in Slave Mode
- Supports 12-bit Single Clock with Dual Edge
DVI Interface
- Multiplexed with AGP Interface
- Clock Speeds Up to Full DVI Spec. of 165MH
- Simultaneous Display with CRT
Integrated Clock Synthesizer
- Internal PLL’s Derive All Internal and External
Clocks
- Spread Spectrum Capable on All
- Output (Except DVI and LDT Clocks)
PERFORMANCE
- Dynamic Adaptive Speculative Pre-processor
(DASP)
- 350Mpixels/sec. on Two Pipelines
- External AGP4X Option with Fast Write
- TwinBank Memory Architectur
- 128-bit DDR Simultaneous Memory Access
- 4.2GB/sec. Peak Bandwidth
- HyperTransport Technology 800MB/sec.
- 266 MHz DDR Front Side Bus Support
QUALITY
- NVIDIA Unified Driver Architecture (UDA)
- WHQL-Certified Windows 2000, Windows NTŪ, Windows
98, Windows 95, and Windows ME
- Complete Linux Drivers, Including Full OpenGL
MCP FEATURES
Audio Processing Unit (APU)
- Dolby Digital Interactive Content Encoder
- Hardware DirectX 8.0 Audio Processor
- 256 Total Voices
- 64 3D Voices
- Downloadable Sample Version 2 Acceleration
(DLS2)
- 32 Bin Mixer (8 voice volumes mapped to each
bin)
StreamThru
- IEEE 802.3 MAC (Media Access Controller)
- Supports 10/100 BaseT Ethernet/Fast Ethernet
- Supports HomePNA 2.0 PHYs
- Carrier Sense Multiple Access/Collision Detect
Compliant with IEEE 802.3 and 802.3u (MII)
- ACR Interface Support
- CNR Interface Support
- Isochronous Controller Paired with HyperTransport
Results in Fastest Networking Performance
Complete Communications Suite
- Supports 6 USB Ports
- Configurable as 3/3 or 2/4
- Supports Full Speed (12Mb/sec.) and Low Speed
(1.5Mb/sec.) per Port
- ACR Interface Support
- CNR Interface Support
- Allows USB Concurrency
- Supports 10/100Base-T Ethernet/Fast Ethernet
- Supports HomePNA 2.0 PHYs
HyperTransport™ Technology
- High-speed (800MB/sec.)
- Low Voltage, Differential, and Low Pin Count
Interface
- Communication Link Between IGP, MCP, and CPU
Dual ATA-100 Disk Drive Controllers
- Supports Ultra DMA Mode 0, 1, 2, 3, 4, and
5 (Ultra DMA-100/66/33)
- Industry Standard PCI Bus Master IDE Register
Set Separate Independent IDE
- Connections for Primary and Secondary Interface
(5V Tolerant)
AC’97 2.1 Compliant Interface
- Supports 2, 4, or 6-channel Audio
- Dual AC-Link – Supports Up to Two CODECs
- 16-bit Stereo Output and Input Streams
- Supports Input, Output, and GPIO Channels for
Host-Based Modems
- Separate Independent Functions for Audio and
Modem
- Six Independent DMA Controllers
- ACR Interface Support
- CNR Interface Support
- S/PDIF Output (Stereo or AC-3 Output)
Internal Fast PCI to PCI Bridge (Version 2.2
Compliant)
- Supports Up to 5 External PCI Slots at 33MHz
- PCI Master and Slave Interfaces
- Bi-directional Write Posting Support for Concurrency
Internal Fast PCI-to-LPC Bridge
- Subtractive Decode
- Can Be Connected to External Super I/O, Flash
BIOS, and LPC-to-ISA Bridge
- Supports LPC DMA and Mastering
- Serial Interrupt Protocol Support
Integrated Clock Synthesizer
- MCP, LDT, PCI, IDE, and USB Clocks
Dual System Management Bus (SMBus)
- 2.0 Interface
- PEC and ARP Support
- SMBus Host and Slave Support
- SMBus Alert Support
PERFORMANCE
- 256 Total Stereo Voices
- 64 3D Voices Plus 192 Stereo Voices
- HyperTransport Technology 800MB/sec.
- Dual ATA-100 Disk Drive Controllers
- PCI Support at 33MHz
QUALITY
- NVIDIA Unified Driver Architecture (UDA)
- WHQL-Certified Windows 2000, Windows NTŪ, Windows
98, Windows 95, and Windows ME
- Complete Linux Drivers
- Fully Compatible with Microsoft DirectX 8.0
|