On June 19, 2000, AMD released their first value processor based on a new architecture since they brought the K6 to the eyes of the public, over three years ago.  This processor was capable of delivering 90% of the performance of high performing Athlon while still being targeted at the value PC market segment.  At that time, it had been months since the market had seen a high performance yet cost effective processor.  This processor, the AMD Duron, had changed all of that.

It was instantly on the enthusiasts wish list.  Now selling for well under $100 per chip, in many cases the Duron was just barely more expensive than a K6-2 or a K6-III, which couldn’t even dream of offering performance competitive to that of the Duron.  In terms of competition from outside AMD, even a Celeron overclocked to speeds that we still won’t see for another three months couldn’t offer superior performance to the Duron.

But to this day, companies like Gateway don’t even offer a single system based on AMD’s Duron.  Companies offer Celeron, Pentium III and definitely promote their Athlon based systems, yet if someone wants to purchase a high performance, low-cost system, the Duron seemingly isn’t an option. 

Most AnandTech readers that build their own systems know the power and potential of the Duron -- just run over to the AnandTech Forums and read about TheNemesis’ new Duron 600 system or how Regalk got his Duron to run at 969MHz.  Why is it that these big retail manufacturers aren’t doing the same?  Do they just not see the potential of the processor? 

The fact of the matter is that they do understand the potential of the Duron.  The problem is, however, that they have no platform to run it on.  But what about these KT133 boards that everyone is building their Duron systems around, why can’t the retail guys simply use those?  Unfortunately, when building sub-$700 systems that include a free printer and a scanner along with 24/7 tech support, these larger system manufacturers have to make some sacrifices. 

You won’t see GeForce2 or even GeForce2 MX based graphics cards in these systems; most of the time you’ll see something listed as Intel 3D AGP Graphics.  And that Intel 3D AGP Graphics is nothing more than the integrated video present on the i810E chipset, which is exactly why the majority of the PCs in the value market segment are built using Intel Celeron processors. 

System integrators and OEMs love the idea of having a highly integrated platform.  If they can put together a system without having to include third party graphics accelerators, modems or sound cards, they’re happy since that’s money saved.  And guess what VIA’s KT133 chipset doesn’t provide - integrated video support.

On the other hand, Intel’s 810E chipset is perfect for these system builders since it’s a tried and true solution with integrated audio, modem and video support.  Not only that but the chipset support both the value Celeron processors as well as the mainstream 100MHz FSB Pentium III CPUs.  Using a single system design, a vendor can offer this idea of a “built to order” system to their customers, when all that’s being changed is how many sticks of SDRAM are installed, what hard drive they screw in and what CPU they place in the socket.

These system builders don’t care much for performance since their customers mainly purchase based on clock speed and brand name, so the Duron really doesn’t have any advantages for these guys. 

If AMD had the 3D technology to implement their own graphics core into a chipset they would have, then the Duron would be killing the Celeron in sales across the globe.  However, AMD still isn’t to the point where they can do something like that.  Intel developed the 3D core that went into the i810E and i815E chipsets with Real3D back in the i740 days when Intel had dreams of getting into the 3D graphics business.  While those dreams fell apart, the very capable 3D core found its way into Intel’s chipsets and is the reason for the continued success of the Celeron processor, even in spite of the Duron’s superiority when it comes to performance. 

Without a desirable platform to sell the Duron on, AMD had no choice other than to focus on shipping more Athlons, and that they did.  As of November 2000 approximately 24% of the retail desktop market belonged to the Athlon.  But will things turn around for the Duron?

In Europe they have; numerous vendors have picked the Duron over the Celeron, but in North America the same can’t be said.  We would be painting a different picture for you had VIA delivered the KM133, a KT133 with an integrated Savage4 3D graphics core, last year as originally planned.  However, VIA got caught up in producing chipsets for the much larger Socket-370 market which forced the KM133 onto the backburner, and also opened up the market for another manufacturer to step in and save the day for the Duron.

Move over VIA, SiS is back

In the later Socket-7 days, there were three competing chipset manufacturers that all had solutions for your K6, K6-2, or Cyrix processors.  While quite a few motherboards were still shipping with the Intel TX chipset, VIA, ALi and SiS were all fighting for a piece of the pie that Intel had left over when they abandoned the Socket-7 market.  VIA had their Apollo line of chipsets with the MVP3 eventually becoming the platform of choice for many.  ALi had the first official Super7 platform with their Aladdin V, but even before the MVP3 and the Aladdin V had the chance to run through their initial revisions, SiS was there with the 55xx line of chipsets. 

The 5591 from SiS was the last we had heard from them before they disappeared over two years ago.  On the MTech R581-A the 5591 made its debut as the first platform capable of running at the 90MHz FSB setting, which for a Socket-7 platform was incredible (these were in the pre-100MHz FSB Super7 days).  However, there was a lack of support for the 5591 and SiS in general, and eventually the company faded away along with ALi. 

At the end of November we declared ALi back in the game with the release of their MAGiK 1 chipset for the Athlon, and today SiS is back as well with their 730S.  However, unlike ALi, SiS isn’t attempting to get back into the game with a chipset aimed at the performance segment.  VIA already has a tight hold on that market for now and competing with the KT133 as well as AMD’s recently released 760 DDR chipset will be very difficult for a company that has had no experience with the Athlon platform. 

Instead, SiS is attempting to go after the market that has been completely ignored by ALi, AMD and VIA but is obviously present judging by the continued success of the Celeron.  We have been strongly hinting at this market for the past few months and throughout the introduction pages of this review.  We’re obviously talking about an integrated platform for the Duron and Socket-A processors in general, and that’s exactly what the SiS 730S is.

A chip called Homer

Going by the codename ‘Homer’ (not from the Simpsons but Homer as in the storyteller that brought us The Iliad and The Odyssey), the 730S is a very unique solution.  We are used to chipsets having two major components, a Northbridge and a Southbridge, or in the case of Intel 8xx chipsets, a Memory Controller Hub (MCH) and a I/O Controller Hub (ICH). 

The Northbridge or MCH houses the memory and graphics controllers and provides the interface to the CPU as well.  This leaves the Southbridge or ICH to handle the USB ports, IDE channels and PCI devices, among other tasks. 

The two are usually connected via the PCI bus running at 33MHz (133MB/s) or in the case of Intel’s Hub Architecture, the two chips are connected via a Hub Interface running at 133MHz (double pumped = 266MB/s bandwidth). 

The 730S deviates from this tradition, as it is a single chip solution.  This isn’t to say that the 730S doesn’t have a Northbridge and a Southbridge connected by a PCI bus, all it means is that the North and Southbridges are contained within a single chip.  This chip is the 672-pin 730S.

This obviously saves space on the motherboard since manufacturers only have to make room for a single chip in their design and thus drives the cost of the overall solution down right from the start.  Also, by moving to a single chip design, the traces that would otherwise be routed from the Southbridge up to the Northbridge are no longer necessary since all connections are now contained internally within the 730S chip itself. 

To bring the integrated nature of the 730S chipset up to the next level, the single chip also features an on-die graphics controller, which is exactly what the Duron needs in order to be successful in the cutthroat value PC market.  The 730S makes use of the SiS 300 graphics core, which, being a solution made by SiS themselves, is perfectly suited for use with the 730S.  As we discovered in our testing, the 730S’ drivers exhibited no problems and the integrated graphics worked flawlessly.  Performance-wise, you shouldn’t expect too much from the SiS 300 graphics core; basically it is equivalent to the i810’s integrated graphics. 

The one major selling point that the SiS 300 does hold over the competition is that it offers hardware motion compensation and iDCT support.  This helps reduce CPU usage while playing DVD/MPEG2 streams, which won’t be too big of a problem with even the slowest Duron processors, but it will help keep those CPU cycles free for whatever other tasks you may throw at your “value PC.”  For more information on what this support gives the SiS 300, read our article on DVD Quality, Features and Performance.

The SiS 300 features no integrated frame buffer making it a unified memory architecture (UMA) solution.  This means that it shares your system memory using the 64-bit 100/133MHz memory bus.  This gives the SiS 300 a very limited amount of memory bandwidth compared to today’s graphics accelerators.  More specifically, depending on whether PC100 or PC133 SDRAM is used, the SiS 300 will have a maximum of either 800MB/s or 1.06GB/s of available memory bandwidth.  Compared to the “low” 2.7GB/s that the GeForce 256 using SDR memory has, the SiS 300 core obviously won’t be pushing any high frame rates in games.  The 300 will pretty much be limited to 640 x 480 x 16 as a playable resolution regardless of the game. 

Luckily, the 730S, like the i815E, supports an external AGP 4X graphics port.  Making use of this port will disable the on-chip graphics, but it allows for a direct upgrade path for those users that have the 730S in their system.  The only downside to this is that SiS must include all of the pins for the external AGP 4X connector on the 730S chip itself.  Combine that with the fact that the chip must also house both a North and a Southbridge and you can see that the chip is already going to be more expensive than the i810E. 

If a motherboard manufacturer or system integrator doesn’t mind giving up the flexibility of the external AGP 4X slot, they can opt to use SiS’ 301 Video Bridge that allows for a secondary video output to either a digital flat panel monitor, a TV or a secondary CRT.  As we alluded to however, this can only be used in place of the AGP 4X slot, you cannot have both. 

The 730S boasts PC133 SDRAM support with no support for DDR SDRAM; we won’t see a value DDR chipset from SiS until the end of next year with their Odyssey chipset (SiS 740).  (It seems like someone over at SiS’ engineering department has been reading a lot of Greek mythology lately judging by their codenames.)  A maximum of 1.5GB of PC133 SDRAM may be installed in a 730S board.  With the chipset being targeted at the value PC market segment, this limitation doesn’t seem to be that big of an issue.

The SiS 730S Reference Board

Click to Enlarge

 

Something to keep in mind is that SiS has much less experience with PC133 memory controllers than VIA.  We have criticized VIA heavily in the past for having poor memory performance, but VIA has had since late 1999 to perfect their PC133 memory controller whereas the PC133 memory controller present in the 730S is generally new territory for SiS.  Keep this in mind as we move into the benchmarking section of this review.

The 730S claims support for both the 100 and 133MHz DDR FSB frequencies (effectively 200/266MHz).  While we could get both settings to work, the 133MHz setting was noticeably less stable even to the point where none of our Windows 2000 tests would complete.  Chances are that the issues will be sorted out in future revisions of the chipset if not simply requiring a BIOS update on our particular test board. 

The 730S has support for 6 USB ports in addition to the usual AC’97 and Modem Riser card support.  Like the i815E, the 730S has built in support for 10/100 Ethernet or 1Mbps HomePNA with the inclusion of an external controller chip, the SiS 900 in the case of 10/100 Ethernet support. 

Conventional Chipset Design

The final feature of importance surrounding the 730S is its ATA/100 controller.  In conventional chipset designs, the IDE controller must share the 133MB/s bandwidth that the PCI bus offers with the rest of the PCI devices as well as your USB ports, etc…  Intel’s Hub Architecture attempts to solve this problem by offering a wider 266MB/s bus connecting the I/O Controller to the Memory Controller (essentially the Northbridge).  The 730S addresses this issue by providing the IDE channels with a dedicated 133MB/s pathway to the Northbridge.  This should help alleviate any bottlenecks that may arise as ATA/100 drives come closer and closer to peaking at that 100MB/s limit of the specification. 

SiS 730S Chipset Design

With all of the features that the 730S incorporates into a single physical chip, it is amazing that SiS can keep the price down to $39 in 10K quantities.  This is almost identical to the cost of AMD’s 760 chipset, but this covers the cost of integrated video as well as all the normal functions of the chipset.  This is approximately $6 more expensive than Intel’s 810E chipset, which can be expected since the 730S is a physically larger chip than both the GMCH and the ICH of the i810.  This translates into more expensive system costs when you look at the big picture where hundreds upon thousands of these systems will be shipping.  Whether this will, in the end, affect how competitive the Duron will be in comparison to the Celeron in the retail market has yet to be seen. 

Evaluating the platform

Obviously if we simply benchmarked the 730S in the same manner that we have done all of our other Athlon chipset reviews, it wouldn’t be too useful of a comparison.  Realistically speaking, no one is going to purchase a 730S-based system with a 1.2GHz Athlon and 256MB of SDRAM.  However, without doing a comparison with the fastest processor and a decent amount of memory, we can’t fairly evaluate the 730S as a chipset alone without introducing potential bottlenecks created by using a slower CPU and/or less memory.  So we ran two sets of benchmarks designed to evaluate the SiS 730S as: 1) a chipset and 2) as a platform solution for the Duron processor.

The first set of benchmarks follow our usual chipset review format: they pit the 730S against the KT133 and the ALi MAGiK1 chipsets to show any strengths/weaknesses of the chipset itself.  In these tests we use the external AGP 4X slot to house our test bench GeForce2 GTS in order to eliminate any sort of bottlenecks that may be caused by the chipset’s integrated video.

The second set of benchmarks is designed to place the 730S in the type of system configuration you can reasonably expect to see it used in.  We make use of a Duron 800, 128MB PC133 SDRAM as well as its integrated SiS 300 video core instead of our own GeForce2 GTS.  In this section, the system is compared to an identically configured system using the KT133 chipset and a NVIDIA RivaTNT2 M64 video card (which is commonly found in the lower priced Athlon/Duron systems) as well as a Celeron 766 system using an i810E motherboard. 

So if you’re interested in seeing the chipset compared to other Athlon chipsets, take a look at the first set of benchmarks, otherwise skip ahead to the second set. 

The Chipset Test

As we’ve explained in the past, SYSMark 2000 is a combination of a Business and Content Creation benchmark.  The benchmark runs through a set of 12 application tests ranging from business performance under Microsoft Word to content creation performance under Adobe Photoshop.  SYSMark 2000 only runs a single application test at a time and the suite as a whole is representative of the way some may use their systems provided that multitasking among multiple applications isn’t characteristic of their usage.

Our ALi MAGiK1 platform wouldn’t complete these tests under Windows 2000 so they are left off the chart, however the data we have is enough to see that the 730S’ memory controller is definitely inferior to that found in VIA’s KT133.  The VIA KT133 vs SiS 730S match here is perfect since both are using 100MHz FSB settings and PC133 SDRAM, and in these tests, both setups are using the same GeForce2 GTS graphics accelerator. 

The KT133 is just over 10% faster than the SiS 730S, the only factors that can be attributed to this lead being the disk and/or memory controllers.  Let’s continue with the benchmarks and see if any other clues pop up.

ZDM’s Business Winstone 2001 is more of a real world business application performance test.  The test involves multitasking among various business applications such as MS Word and Excel as well as while browsing locally stored web pages using Netscape, unzipping files, etc…  This obviously makes the test very disk intensive and thus disk limited. 

The 730S is just approximately 1% slower than the KT133 here which is definitely acceptable, it’s a much better margin than what we saw under SYSMark 2000.  This could be because of the fact that Business Winstone 2001 is more disk limited than SYSMark 2000, introducing a limitation that isn’t allowing the performance gap to grow between the KT133 and the 730S.

Another possibility is that these tests, being relatively simple in nature, aren’t taxing enough of the memory bus to bring any differences between the two chipsets to light.  In order to answer this question let’s have a look at a slightly more memory intensive test from ZDM, Content Creation Winstone 2001.

It looks like our theory held true.  Content Creation Winstone 2001 follows the same benchmarking model as Business Winstone 2001 however instead of running business applications, as the name implies, CC Winstone 2001 runs content creation applications such as Adobe Photoshop and Macromedia Dreamweaver. 

These applications and their normal usage patterns are inherently much more memory intensive than the usual crop of business applications simply because most of them can’t just fit within a large L2 cache and must store/retrieve the bulk of their data from system memory. 

This seems to help exemplify the differences in the memory performance of the KT133 and 730S chipsets, allowing the KT133 to pull away from the 730S with a 7% performance lead.  Again we remind you that in these tests the 730S isn’t using its integrated video, making the only difference between the 730S and the KT133 platforms the actual chipsets themselves. 

Under Windows 98 we were able to get scores with the 730S running at the 133MHz FSB along with it running at 100MHz, so we add another bar to the graph. 

Quake III Arena is a bit more than your typical 3D first person shooter since its engine happens to be one of the most well written and stressful when it comes to games of its class.  Quake III Arena has always appreciated increased memory bandwidth as well as a high bandwidth FSB.  Taking these characteristics into account the benchmark results make quite a bit of sense considering what we’ve seen thus far about the 730S’ memory performance. 

Clock for clock we see that the 730S is 12% slower than the KT133, and even when running the 730S at the 133MHz DDR FSB setting (effectively 266MHz) it’s still 7% slower than the VIA KT133 clocked at 100MHz DDR. 

At 1024 x 768 x 32 the performance is limited by the memory bandwidth available on the test bed’s GeForce2 GTS accelerator and thus creates a fairly uniform performance distribution.

UnrealTournament is a very texture intensive benchmark, making it also a very memory intensive benchmark.  However in terms of complexity it is much more like Quake III Arena yet even more prone to falling prey to performance limiting factors thus keeping the 730S within 10% of the KT133. 

Increasing the resolution increases the dependency on the graphics card which cuts the performance differential between the KT133 and 730S chipsets in half down to “only” 5% now.

MDK2 is slightly less complex than Quake III Arena and thus we get some higher performance scores.  The standings remain the same however and the 730S, clock for clock, jumps to just shy of 20% slower than the KT133. 

Again, in a memory bandwidth limited situation the performance remains virtually the same across the board.

There are three different classes we can place SPECviewperf’s benchmarks into, those that are heavily video card, memory or CPU dependent.  This particular comparison is specifically exploiting those benchmarks that are heavily dependent on memory performance since all of the platforms here use the same CPU (with the exception of the i815 platform that uses a Pentium III and the two PC2100 platforms which use a 266MHz FSB Athlon processor). 

SPECviewperf is split into a total of 6 viewsets, each of which measure performance that is representative of usage patterns under certain professional 3D applications such as MCAD and other visualization applications.

The Awadvs-04 viewset is obviously very influenced by a fast memory controller.  Not necessarily memory bandwidth dependent as the AMD 760 is only 3% faster than the VIA KT133, but the performance here is dependent on fast memory timings.  Unfortunately it is also here that the SiS 730S falls back the furthest when compared to the KT133 which happens to be 12% faster here. 

The DX-06 viewset paints an even worse picture for SiS as the performance delta extends to close to 24% in this scientific data visualization/analysis test.  Remember that these two platforms (KT133 & 730S) are identically configured, the only difference being the actual chipsets themselves and of course the motherboards.

The Design Review viewset (DRV-07) is a 3D modeling benchmark that specifically addresses performance when modeling structural elements including piping, beams, etc…  Again, the 730S comes out approximately 20% slower than the KT133.  Luckily for SiS’ sake, the 730S is directed at the value PC market.  However if VIA could ship their KM133 which will use the same memory controller as the KT133, SiS could be in quite a bit of trouble.

First we start off with the MedMCAD-01 test which exemplifies performance under MCAD applications such as Pro/ENGINEER and SolidWorks.  Again the KT133 comes out noticeably faster than the 730S, a 16% lead this time. 

In fact, the closest the 730S comes to the KT133 in performance here is under ProCDRS-03 with an 11% performance differential. 

While the 730S obviously targeted at this market there isn’t a reason why it shouldn’t be offering performance closer to that of the KT133.  We aren’t even testing the integrated video in this case, when paired with a GeForce2 GTS there is no reason that the SiS 730S should be this much slower than the KT133. 

This just helps illustrate what having more experience with the platform can give a manufacturer, keep in mind that this is SiS’ first attempt at an Athlon chipset while VIA has already had quite a bit of experience not only with the memory controller used in the KT133 but with Athlon chipsets in general. 

At the start of this article we mentioned that the 730S is positioned perfectly to achieve the maximum theoretical bandwidth out of its ATA/100 controller because of the dedicated 133MB/s pathway from the Northbridge to the IDE channels.  The perfect way to test this is to use a benchmark that can test burst read performance where you are more likely to reach the limitations of the controller. 

We used the same IBM 75GXP Ultra ATA/100 hard drive for all of these tests.  Out of the five compared chipsets only the KT133 doesn’t officially support ATA/100 (we tested using the 686A Southbridge which is ATA/66 only).  We already know why the AMD 760 and the MAGiK1 are stuck at a 60MB/s, it seems as if Microsoft has not properly implemented ATA/100 support under Windows 2000, a problem which will be fixed with the upcoming Service Pack 2 release for the OS.  Surprisingly enough, Intel’s storage drivers do implement and enable ATA/100 properly as the i815E is able to surpass the ATA/66 limitations and come ahead with a 76MB/s burst speed.

Even more interesting is the fact that the 730S is only able to come forth with 47.4MB/s in this test.  We hope that this is due to poorly implemented Windows 2000 drivers; While this won’t hurt performance today (no IDE drive is capable of sustaining transfers at more than 40MB/s) it will hinder performance of the drives of tomorrow which will be even faster and even more likely to burst at above 66MB/s.

The System Test

Here our 730S platform is just 7% slower than the KT133 + TNT2 M64 setup and a full 11% faster than the Celeron 766 system.  An 800MHz (100MHz FSB) Celeron on an i810/815E platform could give this Duron 800/730S combination a run for the money.  This isn’t the same lead the Duron was boasting on the KT133 when we reviewed it last. 

Once again, 10% slower than the KT133 platform and approximately 7% faster than the Celeron 766 setup.  The Duron loses its appeal in the retail market if these are the performance figures that it’s boasting.  While it’s still the perfect option for enthusiasts that will be tossing the chips on KT133 or AMD 760 boards and overclocking them, a 7% performance advantage over a lower clocked Celeron isn’t really doing much for the average consumer. 

For the first time we see the Duron’s dominance restored courtesy of its 200MHz FSB and its large L1 cache.  Even on the 730S platform the Duron holds an impressive 20% lead over the Celeron.  An 800/100 Celeron can definitely cut into that lead but not enough to jeopardize the Duron’s standings.  Unfortunately for AMD, this incredible lead won’t be one to continue throughout the rest of the benchmarks.

Starting off at 640 x 480 x 16 the SiS 300 graphics on the 730S is offering approximately half the performance of the TNT2 M64.  At this low of a resolution the TNT2 M64 is benefiting from the fact that it has a much faster on-board frame buffer than the SiS 300 which must go all the way over to the shared system memory in order to swap textures. 

The 730S platform is offering decent performance just about equivalent to that of the i815 with a Celeron 766.  An 800MHz Celeron in this case would be able to give the Duron quite a bit of competition.  This is obviously painting a much more competitive picture than when the Duron is resting nicely on a KT133 platform of its own. 

There are two forces at work here that are holding the Duron back.  We know for a fact that the Duron 800 is considerably faster than the Celeron 766 as was evident from our Celeron 766 Review in which the Duron 700 was able to noticeably outperform the processor.  So why is it that the Celeron 766 is within striking distance of the Duron 800 now?

First, the Quake III performance of all of the platforms here is being limited by the performance of the integrated video solutions.  In this case, going from a Duron 600 to a Duron 800 wouldn’t result in a great performance increase at all.  However, overclocking the memory bus from 133MHz as it is now to 150MHz (if such an overclock were possible without changing any other settings) would result in a much larger performance boost.

The second factor is that, as we proved in the earlier benchmarks, the 730S features a memory controller that is between 10 and 20% slower than the KT133’s PC133 SDRAM controller which is already a few percent slower than the MCH on the i815E. 

As the resolution increases, the memory dependency does as well.  This cuts the TNT2 M64’s performance advantage almost in half since the card only has 8MB of SDRAM on it with a 64-bit memory bus.  In 32-bit color the performance drop would be even more significant to the point where both the 730S and the 815E could surpass the KT133 + TNT2M64 setup.

Again the 730S is just barely faster than the 815 with the slower Celeron 766. 

As we mentioned the first time around, UnrealTournament is a very texture intensive game, meaning that it fills up the 8MB frame buffer on our add-in TNT2 M64 card very quickly thus forcing the KT133 to resort to AGP texturing in order to continue trucking along.

This helps level the playing field as the 730S platform only turns out to be 11% slower at this low of a resolution. 

Cranking up the resolution another notch forces the TNT2 M64 to depend even more on AGP texturing and loses its once admirable performance advantage.  Now the SiS 300 on the 730S goes to the head of the class with a slight 3 fps advantage over the i815E platform. 

We mentioned before that the SiS 300 graphics core integrated into the 730S chip had hardware motion compensation and iDCT support.  If you look at the above CPU utilization numbers you can see exactly how much headroom those features buy you.  The SiS 300 manages to cut the CPU usage numbers in at least half for the less complex DVD streams and even more than that for the more complex streams. 

While this doesn’t matter as much for a processor like a Duron 800 versus a K6-2 350, when you’re talking about using 50% of your CPU versus only using 15% the difference can be noticeable regardless of what processor you happen to have. 

Final Words

The SiS 730S holds a lot of promise and it is definitely a step in the right direction, since the Duron desperately needs a highly integrated platform to run on.  The motto we associated with the Duron when we first reviewed it back in June was: “value does not have to mean slow.”  Unfortunately the SiS 730S seems to combat that motto quite a bit. 

While we did expect lower performance in 3D applications and games out of a UMA platform like that 730S, there is still no excuse for the degraded memory performance we saw here.  If you factor out the integrated graphics and use an external GeForce2 GTS card, the 730S was still 10% slower in 2D applications and up to 25% slower in 3D applications. 

Obviously SiS’ memory controller isn’t as mature as that which is present in VIA’s KT133 chipset.  However SiS has every reason to worry since the same memory controller which is dominating the 730S now is present in VIA’s KM133 chipset which, unlike the KT133, is targeted at the value market segment. 

Can the 730S possibly succeed if VIA’s KM133 is introduced into motherboard designs in a reasonable timeframe?  If you’re looking at 2D application performance alone, chances are that it won’t, however VIA does have something working against them with the KM133. 

While the 730S features SiS’ tried and true 300 graphics core, the KM133 relies on the Savage4 graphics core VIA acquired in the S3 deal.  The Savage4 wasn’t a great performer, but our biggest complaint about the Savage4 wasn’t its performance rather it was the incredibly poor driver support.  The 730S’ drivers worked flawlessly during our testing, and since the 300 was made by SiS it was obviously 100% compatible with the 730S chipset design. 

However, some of the biggest problems we ever encountered with the Savage4 chip as an external graphics solution were with VIA chipsets.  Granted that a lot of time has passed since we last touched a Savage4 solution, but if VIA hasn’t invested a good deal of time into driver development for their KM133 part, they could be in a bit of trouble.

As far as SiS is concerned, the 730S is definitely a good first attempt at a value solution for the Athlon platform.  And its exactly what the OEMs have been demanding, its cost effective (just a few dollars more expensive than the i810/815E chipsets), highly integrated and has good driver support (not to mention impressive DVD performance). 

Unfortunately it is taking away quite a bit of the performance advantage that the Duron originally boasted over the Celeron.  It is now up to VIA to deliver a low cost yet better performing value platform for the Duron, if they can do that then SiS will be in a bit of a bind, if they cannot however, the Duron may lose quite a bit of its appeal in the retail market.