By Chris Angelini

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 


Intel rocked its infrastructure partners this past summer with a complete revamping of its platform architecture. Not that the move was a surprise. In fact, everyone had been expecting it for quite some time. However, making the transition from older 865PE- and 875P-based motherboards to 915P and 925X platforms may have actually felt like a huge leap for resellers.


After all, those older chipsets supported a long line of readily available AGP graphics cards. They worked with Socket 478 Pentium 4 processors. And they represented Intel’s commitment to DDR memory after fumbling around with RDRAM for so long. The latest hardware redefines all of those technologies to which we’ve grown accustomed, with PCI Express, the LGA775 processor interface, DDR2 memory, and even new power supplies with 24-pin, rather than 20-pin, ATX connectors.

Mobilizing For PCI Express

The graphics folks are saying they’ll keep making AGP cards, but as chipset manufacturers start eliminating support for the bus topology from their products, the motivation to buy AGP cards might evaporate faster than forecasted. And indeed, Intel claims that by the end of 2004, at least 50 percent of its shipping chipsets will support PCI Express exclusively. At the same time, Intel will undoubtedly promote sales of its LGA775 processors, and push for faster variations on DDR2 to procure additional memory bandwidth. It’s a never-ending game driven by a need for speed and higher earnings, isn’t it?

Not only did Intel’s move directly affect its own products, but it also started to hang cobwebs over the chipsets that backed AMD and its Athlon 64 family. Even if AGP is with us for another couple of years, an emphasis on PCI Express will result in less interest in AGP graphics cards and the platforms that support them. Intel is fine with that, but AMD and its partners, in light of the chipsets currently available, are not.

Fortunately, NVIDIA and VIA have announced the technical details of their PCI Express chipsets. Though availability will presumably be sparse through the end of 2004, it will improve in the beginning of 2005, in time for a fresh generation of high-end graphics hardware that will, according to representatives at ATI, emphasize PCI Express first and support AGP sometime later.

If you still aren’t convinced that now is the time to mobilize for a transition to PCI Express, consider that AGP’s prevalence moving forward will be amongst the do-it-yourself crowd interested in upgrading old technology. Those looking for completely new systems will be looking for PCI Express.

NVIDIA’s nForce4

The nForce4 chipset family from NVIDIA should be the first to materialize on third-party motherboards. There are actually three versions of the chipset, all of which will appeal to different needs and budgets. However, the entire family does share several characteristics as well. For example, the entire nForce4 line centers on a single-chip design, simplifying motherboard layout. Additionally, they all offer 20 lanes of PCI Express connectivity, native Gigabit Ethernet, NVIDIA’s built-in Firewall technology, some fantastic options for connecting different hard drives, and 10 USB 2.0 ports.

The top-end version, called nForce4 SLI, is a gaming enthusiast’s dream come true. Its 20 PCI Express lanes are configurable, so a single motherboard centering on the chipset will feature two x16 slots along with one or two x1 slots. When you plug a single graphics card into the board, it runs at the x16 slot’s full bandwidth. But the real magic happens with SLI mode enabled. Using two compatible NVIDIA GeForce 6-series graphics cards from the same vendor and of the same model, both slots switch to x8 PCI Express operation and are made to render 3D graphics cooperatively by way of an SLI connector, which NVIDIA claims will come with every motherboard centering on nForce4 SLI.

Now, throwing a pair of graphics cards into a $200 motherboard might seem like a specialty application that won’t catch on, but you’d be surprised by the amount of interest it’s generating. For starters, SLI scales incredibly well in games that are truly limited by graphics performance. NVIDIA claims that at 1600x1200 and with 4x anti-aliasing, a pair of GeForce 6800 Ultra cards propels top-seller Doom 3 at 71.7 frames per second – 68 percent faster than the 42.4 frames you’d get from a single card. The performance gains are very real, and while the price of acquisition is comparably high, SLI is just one more reason to be jazzed about PCI Express.

NVIDIA’s nForce4 Ultra chipset is nearly identical, only its PCI Express functionality is fixed at one x16 slot and three x1 connections, precluding SLI. The chipset’s features are still attractive, though. NVIDIA is already known for comprehensive storage options, and the nForce4 Ultra expounds on that by adding support for Serial ATA 3 Gbps hard drives, which are still in development. The chipset enables four channels of Serial ATA and two of ATA-133, for a total of eight devices across the two interfaces. RAID 0 striping, RAID 1 mirroring, and RAID 0+1 all work natively, and again, RAID arrays can span hard drives connected to different interface technologies.

Were you under the impression that RAID is only necessary in enterprise servers and workstations? Think again. Using a couple of like-sized Serial ATA hard drives, you can sell your customer the piece of mind that if one of those drives were to fail, the other in a RAID 1 array would prevent data loss. NVIDIA’s RAID technology makes it easier for resellers to support their systems as well. The Disk Alert System gives you a graphical illustration, in software, to identify a failed drive. How’s that for simplified troubleshooting? Moreover, the ability to designate a spare means that if a drive does go down and you can’t provide service immediately, the standby hard drive steps in to keep the array in top form.

Networking is another focus of nForce4 Ultra. The nForce3 Ultra that ruled previously included some killer Gigabit Ethernet performance. That’s just one of the benefits of a single-chip implementation with low internal latencies and a high-speed bus, I guess. The only downside was greater-than-average host processor utilization when the network traffic was running at full tilt. NVIDIA responded with what it calls the Secure Networking Engine, which accelerates a number of the controller’s functions, alleviating several host processor duties. NVIDIA also claims that its revised Firewall is completely hardware-based and hardened with standard filtering techniques and anti-hacker technologies.

Even if AGP is with us for another couple of years, an emphasis on PCI Express will result in less interest in AGP graphics cards and the platforms that support them. Intel is fine with that, but AMD and its partners, in light of the chipsets currently available, is not.

The lower end nForce4 chipset includes the same PCI Express connectivity as the Ultra, sacrificing the advanced Secure Networking Engine and SATA 3 Gbps second support in favor of a lower price target. If you’re building systems, NVIDIA anticipates that the nForce4 would be a good match for Athlon 64 and Sempron processors.

One other advantage that NVIDIA wields over its competitors is extensive software support for its hardware products. As it pertains to nForce4, this includes a proven track record of respectable drivers and the nTune configuration utility. The former is an important part of evaluating the robustness of any hardware. And at least from a week of usage, the drivers for nForce4 would appear to be well-developed. The latter, nTune, helps the enthusiast tweak their platform for optimal performance using memory timing adjustments, bus overclocking and, in systems with GeForce FX or newer graphics cards, GPU overclocking controls. Because nTune offers an unprecedented level of configurability, it’s probably something you won’t want to preinstall, but you’ll certainly want to let enthusiasts know that such a tool exists and works with all models of the nForce4 family.

VIA’s K8T890

Early on in the life of AMD’s Athlon 64, VIA earned itself a fair bit of popularity by going after enthusiasts—the folks who were impressed with the processor architecture’s alacrity in gaming and content creation—while NVIDIA chased down favor with workstation users. VIA had native Serial ATA, a high-quality audio option, and faster HyperTransport connections, which combined to make its K8T800 chipset attractive early on. To a degree, history might be repeating itself. Again, it looks like NVIDIA’s interest is in storage and networking, while VIA adds more multimedia.

VIA’s design approach is traditional, involving north and south bridges, and initially only one model of each. The K8T890 north bridge features 20 PCI Express lanes, a 1 GHz,

It’s particularly interesting that VIA doesn’t offer Gigabit Ethernet with the K8T890. Instead, representatives claim that users will see better performance from a Gigabit add-in card plugged into one of its PCI Express x1 slots.

16-bit interface to its host processor, and an Ultra V-Link connection to its complementary south bridge. K8T890 isn’t compliant with NVIDIA’s SLI graphics technology, so the component supports one x16 slot and four x1 peripherals.

There’s even more action on the upcoming VT8251 south bridge, which includes two more PCI Express x1 lanes for peripheral connectivity. VIA adds support for Intel’s High-Definition audio specification as well, giving the chipset a leg-up on nForce4. NVIDIA’s solution actually supports eight-channel AC’97 codecs, but the audio fidelity is less than what you’d get from the HD standard. Of course, that also means you’ll need to ensure a third-party motherboard actually includes the feature, which won’t always be exposed, especially on budget boards.

Like NVIDIA, VIA also spent some time improving its storage subsystem. The VT8251 includes native support for four Serial ATA and four ATA-133 drives. There aren’t any provisions for the SATA 3 Gbps speeds, but the controller does feature NCQ (Native Command Queuing) for a modest performance improvement with compatible hard drives. RAID 0, 1, 0+1, and JBOD configurations all work solely with the Serial ATA controller rather than spanning both serial and parallel drive interfaces.

It’s particularly interesting that VIA doesn’t offer Gigabit Ethernet with the K8T890. Instead, representatives claim that users will see better performance from a Gigabit add-in card plugged into one of its PCI Express x1 slots.

Two more variations of the K8T890 north bridge are currently scheduled for production this year, including one higher-end product that purports to support two graphics slots, similar to nForce4 SLI, and a more mainstream solution with an integrated graphics core. VIA hasn’t yet laid out the technical details of its dual-slot chipset; however, that one will go by the name K8T890 Pro. The K8M890 will feature S3’s DeltaChrome graphics technology, appealing to business customers who don’t necessarily need a discrete graphics card.

In Retrospect

Though a long time in coming, AMD’s Athlon 64 is finally getting support for PCI Express and drawing even with Intel’s 900-series chipsets, at least with regard to features. It couldn’t happen at a better time either, as graphics card manufacturers will focus primarily on the new bus topology from here on out. AGP will still exist, of course, but the real action will take place on these new platforms.

     
  Back to top
     
 
   
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form.