Full Power: NVIDIA Attacks With nForce2 – THG.RU

<!–#set var="article_header" value="Full Power:
NVIDIA Attacks With nForce2″ –>

Full Power: NVIDIA Attacks With nForce2

NVIDIA is a company with an exemplary success story. While the chip sector struggles with difficult economic conditions, NVIDIA has been able to report increasing stock values. The nForce chipset, launched about a year ago, was the only thing that couldn’t keep up with the star career of the GeForce graphics accelerator – despite its many technical innovations and a famous brand name. What happened?

There was certainly nothing wrong with the features. Even today, the nForce420/D is the only chipset that has a dual-channel memory interface. Only Intel’s E7500 is also able to support the use of DIMMs in pairs, which concentrates the bandwidth and increases overall performance. However, this chipset is reserved for use in servers, and is therefore more expensive.

The delays at NVIDIA were significant – it was only months after the official release that nForce-based products were actually available. nForce motherboards didn’t come out until Q4 in 2001. Many manufacturers balked at the high price of the chipset, and with good reason, because the latest products from VIA and SiS at the time (KT266A, KT333, SiS745) were offering the same performance at considerably lower prices.

What made things even more difficult was that there were hardly any users who really wanted the integrated graphics offered by the nForce. Even though the performance was on par with the GeForce2 MX, this wasn’t interesting anymore, especially not six months after the product launch. On the one hand, its 3D performance was too weak, and on the other hand, the price of the chipset was too high to be used for PCs in an office context. Paradoxically, NVIDIAs fast product cycle for 3D chips was the thing that tripped up the nForce chipset.

With the nForce2, NVIDIA makes up for this in two aspects: for one thing, the chipset has been brought to the highest level of technology, which clearly targets the higher end of the market; also, NVIDIA has given the nForce2 a modularity that can meet the needs of all market segments.

Chipset Basics: Meaning and Functions

When purchasing a PC, the chipset is all too often overlooked, which is unfortunate because the performance of a computer depends on it. The chipset in a PC can make as much of a difference in performance as, say, the difference between using an Athlon XP 1600+ and an 1900+. And this is no joke: while the first DDR chipsets for the Athlon were faster than their SDRAM ancestors, they were always significantly outclassed by the generations that followed. In the end, pure processor power is not enough if the system itself puts on the breaks.

The actual function of a chipset is to communicate between all components of the modern PC, or, to put it in other words, the chipset is the infrastructure. The chipset contains numerous components, and moreover, it makes various interfaces available for connecting additional components (PCI, USB, AGP, IDE, etc). As a rule, the chipset consists of two units, the Northbridge and the Southbridge. The reason that it is made of two units instead of one is that it’s difficult to integrate all components onto a single chip. The use of two chips also allows for different combinations of Southbridge and Northbridge chips. Today, most manufacturers have a small building-block type of system from which the motherboard makers can choose components to suit customer requirements.

The Northbridge usually contains the CPU interface and the memory controller. With nForce and nForce2, the graphics unit is also on the Northbridge. This chip is clearly larger in size compared to the Southbridge, because hundreds of data paths are needed for the processor and the RAM. Because of this, all complex chips are packed in BGA (Ball Grid Arrays) casing, where all the soldered connections are located on the underside of the chip.

For years, the PCI bus was used as the connection to the Southbridge. However, its bandwidth is no longer sufficient for today’s requirements, so the chip makers all offer their own solutions (e.g., VIA’s V-Link, SiS’s MuLTIOL, Intel’s i-Link and AMD/NVIDIA’s HyperTransport).

Currently, the Southbridge contains at least a PCI controller, floppy/ IDE/ hard disk controllers, serial and parallel ports, USB support and power management functions. Through the years, the chipsets have gained extended functionality through the Southbridge, which is the reason that nowadays these units often also include interfaces and codecs for sound or network controllers. This means that motherboard manufacturers only need to provide simple controllers, which are much cheaper to produce than complete PCI network cards or sound cards.

NVIDIA: A Success Story

Founded in 1997 by a handful of ex-SGI employees, NVIDIA started out with the Riva128 chipset, which was a fresh stimulus in a 3D sector that had been dominated by 3Dfx until then. Shortly afterwards, Riva TNT and TNT2 caused the downfall of 3Dfx and its Voodoo 3D accelerators. Since 1999, the fastest graphics chips are called GeForce. It’s a success story that should soon continue with the fifth generation of graphics chips.

NVIDIA’s relationship with Microsoft was promising from the start, and it was an important factor in the company’s success. Today, NVIDIA has contributed its chipset plus 3D graphics unit to the Xbox. When the partnership was announced, NVIDIA’s stocks increased considerably, counter to the current trend. The reason for the increase, however, was not only due to its partnership with Microsoft, but also to its expansion into the areas of chipset and audio components.

Early on, NVIDIA saw that it had to address all three market segments (low-end, mid-rand, high-end). And so it began with a low-cost version of the Riva TNT (M64), continued on to the GeForce2 MX, GTS, Pro and Ultra, and up until the two latest series, the GeForce4 MX and Ti, both of which are divided up into three different performance categories.

It was only with the chipset that NVIDIA broke away from its own strategy, because the graphics controller was a fixed part of the first nForce. NVIDIA’s hopes that the chipset would meet with acceptance through its high 3D performance quickly dissolved. The customers were king, and they weren’t ready to shell out extra bucks for the graphics unit. The story was the same for both the end-user as well as the motherboard manufacturers. The latter was not ready to take the risk, in the case that the product could not be sold.

nForce & nForce 2: Only for AMD CPUs!

If you were hoping for a chipset for the Pentium 4 – sorry! nForce2, just like the nForce, is only for AMD Athlon XPs. It can only be speculated as to exactly why NVIDIA took this stance, but indirectly, NVIDIA also offers a technically similar chip for the Pentium III. The Xbox contains such a variant of the nForce technology. Because of this, rumors have it that NVIDIA might soon introduce an appropriate chipset for the Pentium 4.

nForce2: An Overview of Features

Before we look in detail at the individual features of the nForce2 and compare them to the established chipsets, let’s look at a summary of its specs.

Northbridge: IGP and SPP Compared

Apart from new aspects of the technology, one of the most important differences from its predecessor is in the Northbridge. While the IGP (Integrated Graphics Processor) of the first nForce has always been equipped with integrated graphics, the customer can now opt for an nForce2 withouth integrated graphics, namely, the System Platform Processsor (SPP). The SPP is a “pure” version of the Northbridge, without graphics, and therefore significantly lower in price.

The Force Is On Its Side: nForce2 With Dual DDR400

Up till now, dual-channel DDR mode has only been used by the nForce and Intel’s E7500 chipset. The nForce2 is the third member of this group, and, through further optimizing as well as increasing the the clock speed up to 200 MHz (DDR400), the system performance should increase accordingly.

Dual-channel DDR operation with DDR266, DDR333 or even DDR400 works with the IGP as well as the SPP. The memory controller of course has two separate data buses for dual-channel operation and ‘only’ three address buses, meaning that we won’t see any motherboards with 4 DIMM sockets. According to NVIDIA, dual-channel mode also works with three DIMMs.

As with the nForce, the memory interface of the nForce2 is twice as important. For one thing, the dual memory controller enables the bandwidth to double, because the interface can work with 128-bit instead of 64-bit. This means up to 6.4 GB/s with dual DDR400, 5.4 GB/s with DDR333 and 4.2 GB/s with DDR266. These are values that make the 3.2 GB/s of dual PC800 RDRAM seem quite low. Another aspect is that the graphics unit of the nForce2 IGP doesn’t have its own memory, so it always has to use a portion from the main memory. Today, AGP graphics cards have between 32 and 128 MB of dedicated memory, which usually works at much higher speeds than the system memory; so, systems with integrated graphics are almost always slower than those with separate AGP graphics cards. However, NVIDIA might not be able to solve all problems of shared memory, but, with the use of dual DDR400, it offers a solution thatpromises to perform even better than the GeForce4 MX420.

Memory technology of the future: increasing the voltage from 2.5 to 2.6 or 2.65 V enables a memory clock of 200 MHz. Thanks to double data rate technology, this corresponds to a performance level that SDRAM could only achieve at 400 MHz: DDR can transfer double the amount of data per clock cycle. The nForce2 can manage two memory channels simultaneously.

It cannot be said that the bus width is simply doubled, because, as a matter of fact, both of the memory controllers cooperate in an intelligent manner. The goal is to minimize the latency (waiting cycles that occur when accessing SDRAM) through alternating use of a second memory controller. Also, StreamThru access (realtime data transfer, see details below) gets the highest priority. For example, while the first controller prepares for memory access, the read process can occur through the second controller at the same time. Theoretically, this would lead to a reduction of the latency by one half.

Moreover, NVIDIA gives the memory controllers various algorithems in order to predict and optimize memory access. This feature is called DASP (Dynamic Adaptive Speculative Preprocessor), which, according to the datasheet, has been significantly improved compared to the first nForce chipset. This remains to be confirmed by extensive testing.

Paving the Way for the Next GeForce Series: AGP 8x

Of course NVIDIA thought to integrate AGP 3.0 on the nForce2 – after all, the next generation of graphics accelerators will be based on this fast interface.

Up till now, NVIDIA has not been able to provide samples for testing, so a review of the new AGP interface will come at a later time. The press kits also don’t mention whether the graphics unit of the nForce2 IGP is connected via AGP 4x or AGP 8x, although this is an aspect that can very much affect the performance of the graphics unit.

As for the nForce2 SPP, the 3D performance depends solely on the graphics card that is used. Nevertheless, the nForce2 is theoretically the best platform for 3D applications because of its combination of dual DDR and AGP 8x.

nForce2 IGP: GeForce4 MX Integrated

At this point, we are unable to make a definitive statement about the performance of the graphics unit for lack of a testing sample. The 3D unit of the first nForce worked with 175 MHz. With the nForce2, there should be a significant increase in performance compared to that, but there are tighter limits to integrated chipsets compared to dedicated graphics chips, so we expect a clock speed between 220 MHz and 250 MHz (corresponds to 440-500 Mpixels/s).

The 3D features correspond to those of a GeForce4 MX, and include the following:

• 256 Bit 3D accelerator
• Shading Rasterizer with DirectX8 support
• No Pixel or Vertex shaders (only with GeForce 3 and GeForce 4 Ti)
• 32 Bit Stencil Buffer
• Accuview Anti-Aliasing
• Multiple-display support through nView
• Digital Vibrance Control
• Hardware DVI-Out for TFT displays
• Resolutions up to 1920×1440
• Integrated PAL/NTSC TV encoder for TV-out up to 1024×768
• MPEG-2 hardware decoding

For further details on the GeForce4 MX, see the article PC Graphics Beyond XBOX – NVIDIA Introduces GeForce4.

Southbridge: MCP and MCP-T

MCP is NVIDIA’s acronym for Media and Communications Processor. The “T” version simply means “Turbo,” which in this case does not translate to higher speeds, but to a more extensive range of functions.

Basically, the new MCP-T differs from the MCP in three aspects: the MCP-T offers an integrated FireWire controller and a second network controller, as well as NVIDIA’s Audio Processing Unit (APU). And be careful to note that MCP has different meanings with the nForce and the nForce2. With the first nForce, the APU was included in the simple MCP, whereas with the nForce2, the MCP-T is required.

UltraATA/133 – Soon To Be Outdated?

As is appropriate for a new chipset, the fastest hard disk interface, UltraATA/133, is also included. Currently, only Maxtor has hard disks for this new standard in their program, and we expect that Seagate and Western Digital will not be selling UltraATA/133 hard drives at all. The reason for this is simple: a majority of the manufacturers, including Intel, have decided against UltraATA/133. This is not meant as a blockade, but rather, is supposed to expedite the introduction of UltraATA/133’s long-awaited successor, Serial-ATA.

We’ve already received the first controllers, but we still have to wait a few weeks for the new hard disks. Serial-ATA brings you bandwidths of up to 150 MB/s (although half of today’s hard disks can’t even make use of even half of its potential), and it means getting rid of those impractical and touchy flat cables. As implied by the name, Serial ATA does away with parallel operation, so it only needs a handful of wires. Moreover, there’s much less chance of electrical interference. Without the flat cable, it is also easier to ensure proper air circulation in the PC case.

Even if UltraATA/133 will be perfectly sufficient for the time being, it is still curious as to why NVIDIA, for all its apparent perfectionism, didn’t go directly to integration of Serial ATA in the chipset.

Two Network Controllers (Optional), Thanks To DualNet

Here, NVIDIA offers a feature that might not seem to make much sense at first glance, namely DualNet, provided by the MCP-T. As the name suggests, DualNet consists of two network controllers. Upon closer scrutiny, this proves to be a welcome feature. Because the traditional modem is being replaced by more modern technologies, such as DSL, a network card is pretty much indispensable these days. However, a single interface is only sufficient if the PC is to be connected to either the Internet or a a local network, but not both. If you need to connect the PC to both, then you definitely need a second port.

In any case, from a financial point of view, integrating a second network port makes sense, because it only costs you a few dollars per port. Also, the nForce2 MCP-T allows PCs to function directly as a gateway between two networks – regardless of whether one is LAN and the other is the Internet, or if both are LAN. In this case, a PC equipped with this feature could function as a server for two autonomous networks, taking on the corresponding routing functions. A configuration with a third network card on the PCI bus is also possible, if you want to connect two networks to the Internet, for instance. In short, there’s a wide range of possibilities with this feature.

HyperTransport and StreamThru

The fact that NVIDIA uses a technology that originates from AMD shows you just how closely the two companies are cooperating: NVIDIA uses HyperTransport as the link between the Northbridge and the Southbridge. With up to 800 MB/s, HyperTransport far surpasses the bandwidth of all other technologies. A few years ago, both units of the chipset were connected via the PCI bus (e.g., on the 440BX with 133 MB/s). Today, Intel uses its own hub architecture, and it’s a similar situation with VIA and SiS (V-Link and MuLTIOL, with 533 MB/s).

In practice, these links should be enough, but HyperTransport has the added advantage of being able to work isochronously – NVIDIA calls this feature “StreamThru.” This means that data transfer can happen in real time (iso = equal, chronos = time), where the access for an uninterrupted datastream is guaranteed by the chipset (per Single-Step Arbiter of the nForce2). This is particularly important for applications that rely on constant datastreaming (e.g., CD recording, video editing, hard disk recording).

Luckily, the multifaceted functionality, as well as the better performance, don’t require enormous technological effort – the link between Northbridge and Southbridge is created with relatively few circuits (only 8-bit bus width), thereby helping motherboard manufacturers to keep costs down.

External Connection: USB 2.0

A total of six ports are available for the MCP as well as the MCP-T. The MCP chip of the nForce chipset only supported USB 1.1, while the Southbridge of the nForce2 also supports USB 2.0. With a maximum bandwidth of 480 MBit/s, the USB connection is no longer meant for scanners, printers and digital cameras exclusively, but, because of its high data rate and its ability to quickly transfer large amounts of data, it can also be used for external storage systems such as CD recorders and hard disks.

External Connection: IEEE1394 / Firewire

If you want FireWire, you need the nForce2 with the MCP-T unit. This gives you an integrated FireWire controller with up to 400 MBit/s. Nowadays, FireWire is the most important interface for the video segment. Good digital camcorders always feature this interface because when compared to USB, it offers the invaluable advantage of working isochronously, just like the HyperTransport bus between the Northbridge and Southbridge of NVIDIA’s StreamThru architecture. All in all, nForce2 seems the ideal candidate for video editing, because apart from the external FireWire connection and the internal data transfer via HyperTransport, its dual DDR with DASP (Dynamic Adaptive Speculative Pre-Processor) also offers a good basis for continuous datastreams.

nForce2: Overclocking Made Easy

NVIDIA also took power-hungry users into consideration – the integrated clock synthesizer should allow you to overclock the three most important elements (CPU, system memory, AGP) independently from one another. However, it will depend heavily on the motherboard makers as to how this will work in practice and if this sort of functionality will be integrated in the BIOS.

It also remains to be seen if the asynchronous operation of all components is worth it. A direct, one-to-one clock correspondence (e.g., 133 MHz FSB, 133 MHz RAM) is certainly the best choice. Pseudosynchronous relationships (e.g., 133 MHz FSB, 166 MHz RAM) result in significantly longer latencies. We will be investigating how asynchronous clock modes (e.g., 145 MHz FSB, 166 MHz RAM) influence performance as soon as we get a motherboard that allows this sort of tuning. However, features such as the dual memory controller or the isochronous HyperTransport protocol might be able to make up for this possible disadvantage.

Music Is In The Air: The Audio Processing Unit (APU)

NVIDIA defines the term “Audio Processing Unit” as the following: a sound unit with hardware acceleration for 256 voices, 64 of which are 3D voices, as well as an extensive 3D sound logic based on multiple signal processors, support for DirectX 8 and output via Dolby Digital 5.1. The signal must be encoded in the hardware, of course.

The APU itself consists of four parts, which NVIDIA calls “processors”:

• Setup Engine
  This determines the parameters for the other three processors. The Setup Engine is also responsible for memory management, mapping and DMA resources.
• Voice Processor
  This contains a DSP (Digital Signal Processor), which has several functions for processing the voices and mixing the results in a buffer.
• Global Processor
  This unit consists of a programmable DSP. It adds other effects, including programmable effects, produces the audio stream, and outputs to the operating system.
• Dolby Interactive Content Encoder
  The name is pretty self-explanatory. Again, this involves a programmable DSP, which encodes Dolby Digital 5.1 and outputs the datastream to the SPDIF-out. This enables a complex 3D audio stream to be sent over a single digital cable and transmitted via 5+1 loudspeakers (five plus subwoofer).

You can read about further details in the article NVIDIA nForces Success In New Market: APU – The Audio Processing Unit.

Chip Set Comparison: Is nForce2 In The Lead?

Conclusion: The Winning Scores

After the first nForce was introduced a year ago, it was only to be expected that NVIDIA would not be content with a simple overhaul of the product. As a matter of fact, the nForce2 is a completely new product that harks on the technology of its predecessor.

NVIDIA is not the only one to support DDR400 – VIA is set to go with its KT400, and SiS isn’t standing by idly either. However, the latter two products should work with only one memory controller, in contrast to nForce, which provides two.

AGP 8x is gradually becoming the standard, but again, the nForce2 is not alone in providing support for this feature – KT400 and P4X400 (for Pentium 4) also function well with graphics cards of this elite caliber.

The first nForce failed to attain the desired success, due to delays as well as its unfortunate positioning (i.e., the graphics unit was first included). This time around, however, NVIDIA seems to have learned from its past mistakes. The various Northbridge chips (IGP and SPP) allow NVIDIA to fulfill the demands of individual manufacturers: nForce2 with IGP, for example, can be targeted towards business and home users, while nForce 2 with SPP can be geared towards workstations and gaming PCs that have their own, faster 3D graphics card. Moreover, there are two Southbridges available, which can be mixed and matched with the two Northbridge chips. The MCP aims for the masses, while the MCP-T targets multimedia PCs with high-end sound and FireWire.

nForce2 is also the only chipset to natively support USB 2.0 and FireWire. Another unique feature is DualNet: no other chipset is equipped to control two network adapters. The nForce2 even seems to be well-suited for low-end server platforms.

The sound system is not exactly new, but it’s still one of the best that the market has to offer. Note here that the APU is part of the chipset, and not a separate sound card.

For all its efforts to reach perfection, it’s still not quite clear to us why NVIDIA decided not to go for Serial ATA. Promise, Seagate, Adaptec, Highpoint and others already have fresh new products ready to go. It wouldn’t have hurt to delay the nForce2 in order to integrate Serial ATA, but it would seem that NVIDIA had other priorities.