At this year’s Computex VIA was more stressed than they had ever been.  Just prior to the show SiS had sent out their round of 735 reference boards that completely trampled VIA’s KT266 solution.  SiS didn’t exhibit a single problem that the original KT266 shipments demonstrated, the 735 was priced much cheaper and it outperformed VIA’s solution.  The most dominant force the graphics market has ever seen, NVIDIA, was also barking up VIA’s tree with promises of a truly high-class chipset for the Athlon platform.  With SiS attacking immediately and the threat of a very attractive nForce later this year, VIA’s tight grasp on the Athlon market was loosening. 

To top things off, VIA was playing with the idea of showing off a very controversial product of theirs: the P4X266.  With their engineers working night and day on a potential P4X266 demo at Computex tensions were definitely high at VIA; to top things off, the balloon fiasco made this one of the most difficult times for VIA in recent history.

Fast forwarding to the present day, VIA has learned a few tricks from their chief competitor Intel.  While it seems like the market always favors the underdog, VIA has been doing to SiS what we have all been accusing Intel of doing to VIA.  From promising higher performing KT266 BIOSes to offering chipset rebates to manufacturers that wouldn’t promote the SiS 735 chipset, there are some very good reasons that you don’t see more than one SiS 735 based motherboard on the market. 

Luckily with VIA it’s not all about marketing pressure and today is the perfect example of that.  Instead of using political muscle to push the already abused motherboard manufacturers around, VIA has responded to recent threats with what may quite possibly be the highest performing Athlon chipset.

More than just a letter

We first introduced you to the KT266A in our Computex coverage as the KT266 rev 2.  The goal of this chipset revision was simple; to outperform the SiS 735 and provide competition for NVIDIA’s nForce.  It definitely played to VIA’s advantage that motherboard manufacturers in Taiwan were very willing to share performance figures of NVIDIA’s reference nForce boards as well as their own designs.  Along with most of you all, VIA had already seen what the SiS 735 could do so it wasn’t too difficult to aim for a target with hopes of outperforming it. 

The end result is the VIA KT266A which from an architectural standpoint is nothing more than the original KT266 with an updated North Bridge.  The VT8366A (the original KT266 North Bridge was the VT8366) has two major enhancements: an improved memory controller and deeper internal buffers. 

VIA is being very tight lipped about the exact improvements surrounding the new memory controller however we have some educated guesses.  We know that the DDR memory controller in VIA’s recently released P4X266 chipset features some enhancements compared to the KT266.  These enhancements most likely have to do with sizes of internal data pathways and maybe even an intelligent write cache.  Two other recently announced chipsets have included such features: Intel’s 845 and NVIDIA’s nForce.  It isn’t too far of a stretch to assume that VIA would pursue a similar option to improve the performance of their memory controller for the P4X266; remember that the Pentium 4 is a much more memory bandwidth dependent platform than the Athlon, so a high speed memory controller is absolutely necessary for the chipset to be successful.  When applied to the KT266, these enhancements would give the Athlon a bit of a performance boost as well.

There’s no question about it that the P4X266 featured deeper internal read/write buffers than the KT266.  The benefit of deeper buffers is that the CPU can get back to processing quicker after having sent data out to be written to main memory.  At the same time, a more uninterruptible stream of data can be guaranteed from main memory to the CPU.  As CPUs and their FSBs increase in frequency, the reliance on deeper buffers is necessary to ensure performance.  When applied to the KT266 chipset, the Athlon can definitely benefit from better write buffering; remember that the Athlon is no lightweight when it comes to FSB traffic.

Finally, there are a number of timings that can be improved on over time as a manufacturer becomes more familiar with how their chipset performs.  This is a confirmed source of some of the performance gain offered by the KT266A. 

No New Name

Marketing departments always want to run wild with establishing catchy new names and phrases for the great technology that comes out of semiconductor companies like VIA.  So what kept the KT266A from being called VIA’s GoliathForce Turbo?  A big concern that motherboard manufacturers have to face whenever dealing with a new chipset is the time to market for the solution.  If a new chipset however attractive the features may be, requires a new motherboard layout then it’s immediately a longer time to market and a pain to implement. 

Luckily, VIA has always been friendly to their neighboring motherboard manufacturers in Taiwan.  Pin-compatibility between chipsets is something VIA strives for and it’s not a surprise that the KT266A’s North Bridge is pin-compatible with the KT266 North Bridge.  Although VIA calls this a feature of their VIA Modular Architecture Platforms (V-MAP), the reality is that “V-MAP” has been around for much longer than even the KT133.  The important things to take away from the KT266A is improved performance; it was always expected that the solution would be pin-compatible with the KT266.

The rest of the chipset remains unchanged.  It’s still connected to the same VT8233 South Bridge using a V-Link connection.  For more information on the chipset check out our original review of the VIA KT266. 

The Board

We tested on a VIA KT266A Reference Board but shipping boards should be able to offer similar performance. The first boards should hit the market next month.

Click to Enlarge

The Test

There are two contenders that we would have really liked to include in this review but couldn’t.  The new ALi MAGiK1 B0-stepping chipset is performing quite well (although not as well as the KT266A) but Iwill was unable to provide us with a sample board in time for this review.  NVIDIA’s nForce is also ready for action but NVIDIA isn’t quite prepared to let us publish benchmarks yet, it won’t be much longer for that one either.

Memory Bandwidth

The KT266A’s claim to fame is an improved DDR memory controller; what better way to verify that than with synthetic memory benchmarks?

We were very impressed when we first saw the 735’s performance scores under Sandra’s STREAM tests.  A 21% advantage over the SiS 735 is exactly what the KT266A is able to achieve, not to mention a 48% increase over the original KT266. 

The INT-STREAM results were no fluke, the standings do not change much at all here.  The KT266A still holds a 20% advantage over the SiS 735.  But then again it’s extremely rare that these results translate into real world performance advantages right?

Cachemem is much more of a theoretical bandwidth test than Sandra since it doesn’t perform as many functions on the data being transferred.  Here we see the true advantages of the KT266A’s memory controller as it is 30% faster than the SiS 735.  This is almost 80% efficiency of the KT266A’s DDR memory controller; it would be surprising to see even NVIDIA’s nForce offer significantly higher results in this particular test.

Again the KT266 holds a 30% lead over its closest competitor which, in this case is the AMD 760.

Finally we have the memory latency results where the KT266A offers a bit of an advantage over the SiS 735 but not much.  This confirms our suspicions that the majority of the performance improvement came at the hands of deeper buffers and optimizations for greater amounts of traffic coming in over the FSB.  Although VIA’s timing improvements are definitely seen by the close to 30% reducing in latency over the KT266, it still does not hold a huge latency advantage over the 735.

It’s obvious that the KT266 will excel in bandwidth intensive situations, not necessarily latency sensitive applications.  With that said, it’s on to the real world…

Business & Content Creation Performance

If you had heard the expletive that left our mouths after we saw these results you’d be shocked.  We expected the KT266A to be competitive but not this competitive.  The KT266A sustains a greater than 10% lead in Business Winstone 2001.  We’ve made it a point to mention exactly how bottlenecked Business Winstone 2001 is by disk performance and the fact that the KT266A can outperform its closest competitor by 10% on the same exact platform is beyond impressive.  We reran this test over and over again before we believed the results.

The performance improvement drops to about 7% in Content Creation Winstone 2001; not as impressive as the first results but not disappointing by a long shot.

Business & Content Creation Performance (continued)

Although we had always assumed that the Internet Content Creation suite of SYSMark 2001 was very bandwidth intensive because of Windows Media Encoder, we proved the contrary in our P4X266 review.  Thus it isn’t surprising that the KT266A does not achieve a huge advantage over the competition here. 

The same holds true for the Office Productivity suite under SYSMark 2001; the SiS 735 is even able to outperform the KT266A once again.

The overall performance picture painted gives the KT266A a slight advantage over the SiS 735. 

IT/Enterprise Computing Performance

It’s unheard of to get any real variation among platforms using the same configuration at the baseline test settings.  The reason being that we’re simply dealing with basic office application performance here; there are no real bandwidth dependencies that cannot be satisfied by PC133 SDRAM and latency should not be a major issue either.  In spite of this, the KT266A maintains a 17% performance advantage over the rest of the DDR SDRAM platforms.  The only possible explanation for this are the internal optimizations in the KT266A North Bridge; it’s a shame VIA didn’t let out more information as we’d be very curious to find out exactly what they did.  For that matter, so would SiS and NVIDIA.

If the KT266A dominated with no load on the system, there’s no doubt that it will continue to do so as we crank up the number of simultaneous tasks.  A 10% lead is what VIA can command over the SiS 735 here.

The picture doesn’t change a bit as the load increases, the KT266A defends its 10% lead.  Only the 4.2GB/s of memory bandwidth that the nForce can offer has a chance of beating VIA here.

3D Rendering & Animation Performance

We’ve already proven that 3D Studio MAX and other 3D rendering programs aren’t memory bandwidth intensive enough to show a difference between these platforms.  Even the KT133A is sufficient for these needs.

Normally we don’t see much differentiation under SPECviewperf because you’re mostly CPU and GPU limited, however the KT266A is able to offer a 0 – 17% performance improvement over the next best alternative.

3D Gaming Performance

To round out the performance spectrum we have our usual suite of gaming tests.  AMD and VIA first raised the bar on gaming performance with their DDR solutions; SiS did it again with the 735 and VIA seems to have pulled performance out of thin air (or thin silicon – very poor joke, we’ll never do it again) with the KT266A. 

Granted, in a video card bottlenecked game you won’t see this kind of a performance difference between the platforms.  With even faster video cards less than one month away from release don’t expect to be too video bottlenecked at even 1024 x 768 in many games.

Can anyone say, 15% performance advantage?

Final Words

If the VIA KT266A in its current form had been the platform to introduce DDR to the world last year, even Intel would be putting Rambus out to dry.  The performance of this platform is utterly amazing; without a single increase in theoretical memory bandwidth, VIA has taken the KT266 from an easy pushover to NVIDIA’s biggest performance nightmare.  It’s going to take a lot for NVIDIA to outperform VIA’s latest offering.  At least we know that VIA won’t give up without a fight.

The Athlon has at least another year left in it before it is replaced by AMD’s Hammer line as the platform of choice for performance users.  That is ample time for yet another Socket-A chipset to be crowned so let’s look at the future and see what options currently exist.

Obviously the biggest threat to the KT266A is NVIDIA’s nForce-420 which offers twice the amount of memory bandwidth as the KT266A.  There is one thing that few fail to realize and that is the Athlon itself can still use only half of that bandwidth, so you shouldn’t expect 2x the performance out of nForce.  It may turn out that the performance enthusiast that already has a good Ethernet and Sound Card will be better off upgrading to a KT266A motherboard, while nForce is better suited for the user starting from scratch. 

Both ALi and SiS can revamp their Socket-A chipsets to compete with the KT266A but it is much more likely that they will focus on the Pentium 4 instead.  And of course we still haven’t heard anything from ATI who is also scheduled to enter the chipset market shortly. 

It seems like we’re slowly returning to the market of a few years ago where CPU choices are plentiful as are platform choices.  The real question is, what companies in the future will make the same mistakes that left us with only two choices not too long ago?