For the past couple of years AnandTech has failed to have a presence at one of the most interesting trade shows on the planet not to mention the third largest. The show in particular is none other than Computex which is held each year in the Taipei World Trade Center (TWTC) and the Taipei International Convention Center (TICC) in Taipei, Taiwan.

Finally, after waiting for the moons to align perfectly with the sun, AnandTech was finally able to free up an entire week in which we could cover the show. So a total of 18 hours in the air placed us half way around the world ready to tackle the wealth of information that Computex was able to provide.

For those of you that aren't familiar with the show, Computex can be considered to be much like the Fall Comdex held every year in Las Vegas with one major exception, at Computex you get a lot more of the motherboard manufacturers and a lot less of the random manufacturers producing mice and keyboards that we're really not concerned with (no offense to mouse/keyboard manufacturers, they just aren't the most interesting components).

Before we get on to the show we have some eye candy taken from about 37,000 feet in the air as we traveled over Canada and Alaska.

820E where were you?

The one most notable absence from the Computex 2000 show floor was Intel's "flagship" (if it can be called that) 820 chipset. Motherboard manufacturers did display boards based on the i820 chipset but they were very scarce and not nearly as popular at this show as they were 7 months ago at the 1999 Fall Comdex in Vegas.

Even more of a blow to Intel was the fact that the i820E was barely shown at all on the floor. Why is this so shocking? Mostly because the i820E was launched the first day of Computex, Monday, June 5th. For those of you that aren't familiar with the i820E it basically replaces the i820's I/O Controller Hub (ICH) with an updated revision of the part called ICH2 that adds Ultra ATA 100 support as well as support for 4 USB ports.

The main feature of ICH2 however is its integrated hardware LAN support which can be taken advantage of by the new CNR slot on motherboards that feature the new ICH2.

We asked every motherboard manufacturer we met with at Computex how their i820 boards were selling and not a single one (obviously) responded with anything other than "poorly."

Intel is definitely hurting the Taiwanese motherboard manufacturers with their current chipset situation. No one wants the i820 because of its lack of support for SDRAM and the still inflated cost of RDRAM, no one wants the i840 because of its expensive nature and its dependency on RDRAM as a memory technology as well, and no one wants to use an MTH based motherboard design because of all of the problems that have been associated with Intel's MTH.

This leaves three options for motherboard manufacturers as far as chipsets go. They can either 1) Run to VIA, 2) Produce more BX based designs or 3) Hope that Intel's Solano (815) chipset can save the day.

815 & 815E to the rescue

While we didn't see many 820/820E based motherboards, we did see quite a few 815 and 815E based motherboards on the show floor. The 815 and 815E chipset (the difference between the two being that the 815E uses the new ICH2) was announced at Computex on the 5th alongside the 820E but you won't see an official launch until June 19th. We will have a review of the chipset on the 19th in addition to a roundup of motherboards based on the 815 and the 815E that week.

As far as features are concerned, the 815 chipset basically has all of the features of 810E chipset (133MHz FSB, SDRAM support, etc...) with a couple of exceptions. In addition to featuring an integrated video core the 815/815E does boast support for an external AGP 4X connector meaning that you can use an AGP video card and disable the integrated video which was one of the major downsides to the i810E. The 815E, with its ICH2, becomes even more feature filled as it adds ATA 100 support, an additional USB controller and one of the best features, an integrated Intel 82559 10/100 LAN controller.

The integrated LAN controller will make adding onboard Ethernet very easy for a motherboard manufacturer, basically all they have to do is provide an external chip which consumes very little PCB space and the proper connector. The integrated LAN also supports HomePNA (1Mbps) so some manufacturers may choose to put an RJ11 (telephone jack) connector on their motherboards instead of the regular RJ45 LAN connector. If a motherboard manufacturer does not wish to take advantage of the integrated LAN controller they can simply outfit the board with a CNR slot (also supported by ICH2) which will allow for a relatively inexpensive riser card outfitted with either an RJ11 or RJ45 connector to be used for much less than the cost of a regular PCI 10/100 Ethernet card.

Just about every motherboard manufacturer we talked to stated that they were hoping the i815 and 815E would succeed. Otherwise Intel would be quite angry that pretty much their entire line of motherboards would shift over to VIA's stronghold.

VIA owned the show

Talking to a group of VIA representatives and product managers, the gist of things was that Computex was VIA's show and that was a fact which was reiterated by the presence of a VIA chip on close to every single motherboard that was on display at Computex. Unlike last year's Fall Comdex where there was a reasonable presence from VIA among the motherboard manufacturers but an equally great show from Intel with their i820 chipset, at Computex the VIA logo was boasting significantly better attendance than Intel's.

The Apollo Pro 133A chipset which was the talk of the town last year took somewhat of a back seat to VIA's onslaught of other chipsets and designs. This is not to say that manufacturers weren't displaying their 133A boards because they most definitely were, but the chipset is no longer as interesting and unique as it once was since it is now widely known that the BX chipset clearly outperforms it at 133MHz and Intel's 815E chipset is on the horizon as well.

One of the biggest problems with the 133A is that its performance is clearly not up to par with that of the BX chipset. VIA has never been known to be a performance leader, rather a leader of producing low cost chipsets, but if they are to lead the desktop chipset market while Intel reworks their platform strategy then the performance simply must improve. We did question VIA as to exactly why their 133A is so much slower than the BX running at the same frequency and unfortunately we didn't get a straight answer from them other than that the Intel BX is a very well designed chipset. So much for that.

One 133A implementation that did manage to catch our attention was the dual processor motherboards based on VIA's Apollo Pro 133A. You may remember from our Motherboards in 2000 Preview that the 694X North Bridge of the 133A chipset has a little something special about it that makes creating dual processor motherboards based on the chipset a much more cost effective effort. Let's take a look at a quick refresher from what we discovered at last year's Comdex:

A fact we recently learned about VIA's North Bridges, including the 694X of the Apollo Pro 133A from above, is that they include built-in SMP support. While it is true that the BX chipset also supports SMP, it requires the use of an additional ASIC in order to take advantage of SMP. The 694X North Bridge, on the other hand, does not. If you couple the features we just discussed that the Apollo Pro 133A supports along with the low-cost of the chipset, a dual Apollo Pro 133A board could become a very affordable entry level path into the world of multiprocessing.

We weren't the only ones to think about this possibility as we managed to catch at least one demo of a dual Apollo Pro 133A based Pentium III system, so you can expect to see dual Apollo Pro 133A boards surface sometime in the first half of 2000. Stability issues are what is keeping manufacturers from releasing dual 133A boards right now. As you can probably guess on your own, making a dual processor board a stable one takes a considerably greater effort than toying with a single CPU board and thus very few manufacturers are up to the job. When these boards do hit the market, expect them to retail for no more than $130. For a dual processor board, that isn't bad at all.

That particular manufacturer was Tyan and they were demoing their Tiger 133 motherboard which is a dual Slot-1 board based on the 133A chipset. We've actually had the Tiger 133 for months now but the boards we have received (yes, more than one) have been far from stable. We have yet to see a solid Tiger 133 board capable of running as a regular desktop platform much less a high end workstation/server motherboard.

Click to Enlarge

Before you go off blaming Tyan for not being able to produce a decent motherboard keep in mind that Tyan has quite a bit of experience in the high end workstation/server market and the majority of their motherboard designs are multiprocessor boards. The fact of the matter is that creating a stable dual processor motherboard based on the VIA 133A chipset is a bit trickier than doing so on the Intel BX or GX chipsets.

While we saw a handful of boards based on the dual 133A reference design at the show most manufacturers indicated that they were still having stability issues with their designs and they weren't ready for shipping. The only manufacturer that seemed to have a nearly final board ready was Gigabyte who claimed that they had not encountered any stability problems but we'll have to put that to the test when our sample arrives.

VIA's ProSavage PM133 chipset was the basis for quite a few motherboard designs on display at Computex. The PM133 is basically an Apollo Pro 133A with an integrated modified Savage4 AGP core. The Savage4 core was modified to feature the 2D engine of the Savage2000 while using the 3D core of the Savage4.

Just like the i815/815E, the PM133 features support for an external AGP graphics adapter so the integrated graphics can be disabled by either using a PCI or an AGP video card. Providing the AGP slot on a PM133 based board is optional and although all the PM133 based boards we saw had an AGP slot it is still up to the particular motherboard manufacturer or OEM order as to whether or not there will be an AGP slot on board.

In order to offer motherboard manufacturers the greatest flexibility, VIA will begin supplying manufacturers with a new revision of the Apollo Pro 133A chipset's North Bridge, the 694Z (the current part is the 694X). The 694Z only differs from the 694X in that it is pin compatible with the 552-pin 605 North Bridge of the PM133 chipset. By giving manufacturers a pin compatible part VIA is allowing them to create one motherboard design and use either the 133A or the PM133 interchangeably.

Click to Enlarge

As far as pricing goes, the PM133 is set at $40 in OEM quantities which is slightly above the $39.50 price tag Intel's i810E carried upon its initial release but will most likely weigh in just under the OEM cost of an 815E. The true question is, would you rather go for the integrated video of Intel's 815 or the Savage4 core present on the PM133? Three words worth considering while answering that question: drivers, drivers and drivers.

On the AMD side of things, the KX133 chipset took a back seat to VIA's latest offering, the KT133. As you now know from our AMD Athlon "Thunderbird" review as well as our Socket-A Motherboard Preview the KT133 chipset is basically a Socket-A version of the KX133 with the only major changes being most likely timing related. Every motherboard manufacturer we visited had a Socket-A motherboard on display, some had two (ATX & microATX). Most of the boards were relatively similar to each other, we will provide you with pictures in our Day 2 coverage of the show tomorrow. The notable exception to the similarity "rule" about Socket-A motherboards is the QDI Kinetiz 7T board which QDI claims is able to adjust the clock multiplier of the once thought to be clock locked Thunderbird CPU.

A couple of points of interest we discovered about Socket-A motherboards and the Athlon platform in general:

• Motherboard manufacturers are having a very difficult time getting the Thunderbird/Duron platforms to run on 145W power supplies which is what is necessary for FlexATX cases (think of really small microATX). There is no official word from AMD on what can be done to put Thunderbird/Duron in a FlexATX system with a 145W power supply but one company we talked to mentioned that the workaround would be quite "interesting" to say the least.
  
  
• Getting Socket-A motherboards to work at 1GHz and above is quite tricky according to the motherboard manufacturers we spoke with. Iwill was convinced that unless a Socket-A motherboard is very well designed, getting 1.1GHz and faster Thunderbird CPUs to work will be a difficult task.
  
  
• VIA seemed to indicate that the reason current KX133/KT133 motherboards can't reach the 133MHz FSB is because they physically made it impossible to reach that FSB setting on current designs. According to the representatives we talked to they claim that they "didn't want people messing around with [the setting]."
  

Just as the ProSavage PM133 is the 133A with an integrated Savage4 core, VIA's KM133 is basically the KT133 with an integrated Savage4 core (with Savage2000 2D of course). The KM133 is already pin compatible with the KT133 so motherboard manufacturers can use their current KT133 designs as KM133 designs as well by simply replacing the North Bridge and including a 15-pin VGA connector in the place of one of the 9-pin serial ports on the ATX backplane of the board.

While no manufacturer had the chip on display, VIA will be releasing an updated revision of their 686A South Bridge, the 686B. The 686B will basically add ATA 100 support to the 686A and will be pin compatible with the 686A as well so it will basically be a drop in replacement for all current motherboard designs that use the 686A.

VIA's Future: DDR

Without a doubt VIA has been able to capitalize greatly on Intel's lack of any real 133MHz FSB solutions that were PC133 compliant in 1999 up through June 2000. The company is now in a very interesting situation, they have the potential to take advantage of their recent success and completely dominate the desktop chipset market for much longer. Obviously VIA is on the right track to achieving that success because if you look at their roadmap the next major announcement coming from them will be in the form of the first DDR solution for the desktop PC, the Apollo Pro2000.

As you probably know by now VIA's Apollo Pro2000 will be a P6 bus solution meaning that the first DDR chipset from VIA will be a Pentium III/Celeron solution. Shortly following the Apollo Pro2000's release we should be seeing an Athlon version of the chipset which will most likely be called the KT266, the 266 coming from its support for PC2100 DDR SDRAM otherwise known as DDR266 (133MHz DDR).

The DDR memory controller on the Pro2000 and on the KT266 (they will share the same memory controller) will simply be an upgraded version of the current memory controller on the 133A, meaning that if there are any inherent performance limitations caused by that memory controller design they will most definitely be carried over to the DDR platforms. Although it is obvious that the Pro2000 and KT266 should easily outperform any SDR platform available today.

There were banners all over the Computex floor advertising VIA's DDR266 support but we failed to see a single motherboard, not even a mockup or a prototype, that used either the Apollo Pro2000 or the KT266 chipset. Later on we found out from a motherboard manufacturer that VIA's chipset isn't too close to being complete. While the only ALi supporter we talked to at the show claimed that VIA didn't even have a pinout for their upcoming chipset ready, all the other manufacturers claimed that the chipset was in very early alpha stages.

The KT266 in particular seems to be up and running in some labs but the chipset is currently only running with DDR200 (or PC1600) SDRAM, at least that's what our sources revealed to us. Unfortunately VIA was not able to guarantee that the KT266 would be ready by Q4 2000 meaning there is a slight chance that the KT266 could hit the streets in 2001. We'll just have to wait and see how long it takes VIA to crank out the chipset. The good news is that the Thunderbird is running just fine at the 266MHz FSB (133MHz DDR) but we expected that in the first place.

VIA's future chipsets will also be using their answer to Intel's Accelerated Hub Architecture (used on all of the i8xx chipsets) called V-Link. While we couldn't get much information about V-Link, VIA claimed that it would increase the transfer frequency between the South Bridge and the North Bridge to 133MHz which is up from the 33MHz of the current implementation.

ALi

When AMD launched their Athlon 10 months ago ALi was on the list for releasing a supporting chipset, a chipset that never really surfaced on any motherboards other than prototypes and mockups presented at trade shows As a company, ALi has been spending quite a bit of time redirecting their efforts at the IA (Internet Appliance) market instead of the performance/value desktop PC segment they had been concentrating on during the time of the Super7 platform.

In spite of this, ALi did promise to deliver a DDR Athlon platform and if everything goes according to plan it should arrive sometime this year. We asked VIA if they thought ALi was capable of delivering on this claim, in response they stated "[we] think that ALi is really good at PowerPoint." That's a pretty low blow from VIA but it also helps to illustrate the point that not only is designing an Athlon chipset difficult but designing one with DDR SDRAM support is quite an ordeal as well.

We asked every single manufacturer if they planned on using ALi's solution, especially if it became available before VIA's KT266 and every single manufacturer except for one had no plans whatsoever for using the ALi chipset (M1647). Each manufacturer had their own reasons for not pursuing an ALi based DDR Athlon solution ranging from a "good relationship with VIA and no reason to disrupt it" to a more serious "we are still on not so friendly terms with ALi."

The one manufacturer that did announce plans for a DDR Athlon motherboard based on the ALi solution was Iwill and they were more than proud to display their DDR Athlon mockup, which was actually the only Athlon board with 184-pin DDR memory slots that we saw on the floor. ALi told us that there would only be one manufacturer that would be producing a board based on their M1647 which is extremely disappointing, for ALi that is, since one motherboard manufacturer, especially Iwill (they're a great company but they happen to be one of the smallest motherboard manufacturers in Taiwan), isn't enough to get people using their chipset.

Click to Enlarge

Click to Enlarge

Here we have a better picture of the North Bridge and AMD's Duron.

Click to Enlarge

An overhead view of the three 184-pin DDR SDRAM slots.

Click to Enlarge

DDR SDRAM in all its glory.

Click to Enlarge

And finally we have one last shot of the DDR SDRAM.

Iwill told us that they expect to see samples of their KA266-R in September of this year, it will be interesting to see if ALi/Iwill can deliver or if VIA was right in saying that ALi is indeed really good at PowerPoint.