Introduction

Just a week ago we took an in-depth look at the 730S chipset from SiS, the first ever chipset for the Socket-A platform that featured integrated video.  The 730S proved to be a very feature rich solution as well as an option with mature drivers for its SiS 300 integrated video controller, however it did have its shortcomings. 

We mainly attributed those shortcomings to its memory controller, which resulted in sub-par performance when compared to the KT133 chipset.  The use of an external GeForce2 GTS graphics card effectively leveled the playing field between the KT133 and the 730S, allowing us to conclude about the strengths and weaknesses of the chip’s design itself as well as its integrated memory controller. 

If you recall back to our first look at the KX133 chipset, which is virtually identical to the KT133 in all performance aspects, you will remember that we criticized VIA’s memory performance.  That same criticism was present when we reviewed the Apollo Pro 133A, and even before then with the Apollo Pro Plus.  However we never realized how blessed we were to have such well performing chipsets from VIA until we came across the 730S that offered significantly lower performance. 

Concluding our review of the SiS 730S was a forward looking statement that warned of what a competing chipset from VIA would be able to offer over SiS in the near future.

Why should SiS be afraid?  If VIA was able to produce a solution that performed on-par with the KT133 yet featured an integrated video core, the 730S would be in a considerable amount of trouble since it painfully lost the performance battle to the KT133 in all situations.

Well, that near future we spoke of happens to be today and that competing chipset we mentioned happens to be VIA’s ProSavage KM133 which is closely based on the KT133 chipset.  Let’s have a look at VIA’s delayed solution for the Socket-A platform and see if it was really worth the wait. 

The one no one expected to succeed

By the time VIA had finally gained some ground in the Socket-7 chipset market, Intel was done playing with those “silly” sockets and had moved onto the Slot-1 platform for their new baby, the Pentium II.  VIA's first attempt at a Pentium II chipset left Intel with virtually no fears.  Not only was the Apollo Pro Plus incapable of offering performance equal to that of the BX chipset, but it was also not priced significantly lower.

As you will all remember, it wasn't really until the release of the Apollo Pro 133 that VIA was really taken seriously in the desktop PC chipset market for Slot-1/Socket-370 CPUs.  The reason being that the Apollo Pro 133 (and later the 133A) offered everything Intel failed to deliver with their i820 chipset.  For the end-user as well as the OEMs/system integrators, the 133/133A series of chipsets was exactly what they wanted: a BX replacement with 133MHz FSB/PC133 SDRAM support not to mention the useless AGP 4X support which was desired as well. 

Using the Apollo Pro 133A's design as a stepping stone, VIA's next major win was with the KX133, the first PC133 chipset for the Athlon which was met with instant success.  Combined with the fact that VIA essentially had the only Socket-A chipset at the launch of the first Socket-A processors, it didn't take long for VIA to essentially have the Slot-1/Socket-370 and the Slot-A/Socket-A markets at their control.

In spite of this string of success, VIA failed to deliver what AMD most definitely needed.  We discussed exactly why the Duron needs a highly integrated chipset to run on, and VIA realized this as they included plans for a value chipset for Socket-A platforms in their roadmaps, however this value platform, the KM133, took quite a while to make its way to production.  While we saw a few KM133 boards floating around at Computex, we hadn't seen anything surface outside of trade shows and company press material. 

While VIA was concentrating on making one chipset release at a time, in order to make certain they could meet product demand, the KM133's release slipped and slipped until the product was almost forgotten.  Then on September 26, 2000, VIA made the joint announcement with S3 Graphics that the KM133 was officially launched.  What kept the KM133 from being launched months earlier alongside the KT133 as well as the AMD Duron is still unclear, as VIA would be in an even more powerful position had they done so.  In any case, the KM133 is here and now armed with the SiS 730S to compare it to we can offer a solid comparison of what is currently available. 

The ProSavage Line

With VIA's acquisition of S3's graphics division, they gained the rights to use all of S3's graphics technology, the most recent contributions being the Savage4 and the Savage 2000 graphics cores.  VIA's ProSavage line of chipsets all take advantage of these graphics cores by featuring a graphics engine derived from a combination of the Savage4 and the Savage 2000. 

According to VIA, the ProSavage chipsets, which include the KM133, feature the Savage4's 3D core combined with the Savage 2000's 2D core making up what we like to call, the ProSavage graphics core.  The reason for not including the Savage4's 2D core is because of performance concerns, and the reason for not going entirely with a Savage 2000 core is most likely of cost/heat concerns. 

VIA has done nothing to increase the performance of the Savage4 or the Savage 2000 cores that are present in part on the KM133 North Bridge.  Looking back at our original performance analysis of the Savage4, you’ll remember that we found that its performance was approximately equal to that of a default clock TNT2 (125/150) in 16-bit and slightly slower than a TNT2 Ultra in 32-bit performance.  This is at lower resolutions, as you move to 800 x 600 and beyond, the performance drop was more pronounced and fell noticeably behind the TNT2.

With the ProSavage line of chipsets however, the 3D performance won’t be limited by the core, rather it will be severely crippled by a lack of memory bandwidth.  As with all Unified Memory Architecture (UMA) systems the ProSavage core must share the same memory bandwidth as the rest of the motherboard.  Meaning that with PC133 SDRAM, the 1.06GB/s of available peak memory bandwidth must be shared by the CPU, PCI devices, disk controllers, and the ProSavage video core as a means of getting to its frame buffer.  Unfortunately this means that the ProSavage core will have much less available memory bandwidth than the standalone Savage4 which truly matters when it comes to 3D performance. 

In terms of frame buffer size, the ProSavage core uses, by default, 8MB of system memory however that number can be changed by simply hopping into the BIOS and manipulating a setting.  With the solution already being memory bandwidth limited, there will be very little need for more than an 8MB frame buffer setting since most of the time you will be restricted to very low resolutions (640 x 480 x 16) in order to sustain playable frame rates.

Adding ProSavage to the KT133

By taking the ProSavage core and adding it to the KT133 chipset, VIA makes the ProSavage KM133 which is identical to the KT133 in every way (shares the same AGP and memory controllers, etc…) except for the fact that it has the integrated video support.

Just like the SiS 730S and Intel’s 815, the KM133 features its integrated video solution alongside an external AGP 4X slot for future video upgrades. 

Combining the fact that it does have external AGP 4X support with the fact that it shares the same base as the KT133, the performance of the KM133 should be identical to that of the KT133 when using an external graphics card. 

The 552-pin VT8365 Northbridge is pin compatible with the KT133’s Northbridge (VT8363) meaning that manufacturers can fairly easily use a single motherboard design and offer both KM133 and KT133 solutions based upon it.  For more information on what features the KM133’s Northbridge does have, read our description of the KT133’s feature set, which is identical to that of the KM133. 

As is the case with all VIA chipsets, the KM133 has the ability to use any of VIA’s Southbridges.  The most common companion for the 365 Northbridge will probably be either the 686A or the new 686B that offers ATA/100 support in addition to the rest of the features boasted by the 686 Southbridge.  These features include support for 4 USB ports (the SiS 730S supports 6), AC/MC’97 codecs and ATA/66 support by default (as we just mentioned, the 686B supports ATA/100).

The KM133 does boast PowerNow! support which could be very useful for mobile Duron implementations which should be arriving in the next few months.  From an OEM/system integrator standpoint the KM133’s flexible memory support is wonderful but it’s nothing that the competition doesn’t already offer. 

VIA KM133 Reference Board

Click to Enlarge

Other than that, the KM133’s design is pretty straightforward.  It is the same design we have seen countless times on KT133 based solutions, now adding integrated video support. 

In terms of cost, the KM133 is priced slightly higher than the SiS 730S at $40 versus $39 for the 730S, however that shouldn’t cause too much of a concern for pricing differences.  Interestingly enough, the KT133, which is the same as the KM133 minus the integrated video comes in at $34, $6 cheaper than the KM133.

Evaluating the Platform

We are conducting this review of the KM133 in the same manner as we did our SiS 730S Review, splitting up the benchmarks into two separate test sections. A full description of the procedure is provided below:

Obviously if we simply benchmarked the KM133 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 KM133-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 KM133 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 VIA KM133 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 KM133 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 KM133 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 ProSavage 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 and not to mention the KM133's direct competitor, the SiS 730S. 

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

Business/Content Creation Performance using GeForce2 GTS

As usual, we are starting out with SYSMark 2000 as a tool for evaluation Business/Content Creation application performance in an environment where only a single application is used at a time but it is used heavily.  In our original review of the 730S we criticized its poor performance here as it was over 11% slower than that KT133. 

While we predicted that the KM133 would come out just as fast as the KT133, it turned out to be 5% slower.  We questioned VIA about this and they indicated that because we were using a first revision reference board, the performance was not as high as the ASUS A7V we used as the KT133 test bed for this review.  They indicated that the final revision of the KM133 as found in motherboards that will be shipping very soon will perform identically to the KT133 which makes sense since the two chipsets are virtually identical. 

Remember that in these tests we are only comparing the chipsets using an external AGP 4X graphics accelerator to identify the chipsets’ strengths and weaknesses before evaluating the performance of the solution in an actual value PC. 

Under Business Winstone 2001 the business application tests that are being run have no problem fitting within the Athlon’s large L1/L2 caches (384KB total), making memory performance less of an issue.  In this case, the KM133 even comes out slightly slower than the 730S, but still only 3% slower than the KT133. 

Content Creation Winstone 2001 focuses much more on those applications that are used to generate, view and modify “content” such as HTML (Dreamweaver), images (Photoshop) as well as documents (MS Word, Excel, PowerPoint, etc…).  This set of application tests, like Business Winstone 2001 focuses on multitasking performance however these particular tests happen to stress the main memory a little more than those in Business Winstone 2001 resulting in a slight advantage for the KM133 over the 730S. 

You will notice, however, that the performance difference between the KM133 and the 730S is just 2%, which isn’t noticeable at all.  Keep that in mind, as we will definitely call upon that data later in this review.

3D Gaming Performance using GeForce2 GTS

In order to shorten the length of this review we’ve compiled all of the gaming benchmarks into the following page.  All of the performance tests were conducted at 640 x 480 x 32 under Windows 98SE. 

As we predicted in the SiS 730S review, the KM133 is performing on par with the KT133 when using the external AGP graphics instead of the integrated ProSavage video.

There is some variance in performance here but the trend continues, the KM133 is still on top of the 730S and on-par with the KT133.

The trend continues…

…and finally with Expendable, a very memory performance-dependent benchmark, the KM133 is again performing identically to the KT133.

Professional OpenGL Performance using GeForce2 GTS

It was under the professional OpenGL tests that the 730S really lost a lot of ground, however as you can see the KM133 is performing within 0.5% of the KT133.

Again, the KM133 is no more than 1% slower than the KT133.  Everything looks quite promising for the KM133 thus far. 

The Real World System Test

This part of the review focuses on the real world performance of the KM133 in a configuration that you can expect to see it in.

Business/Content Creation Performance using Integrated Video

Here’s where things start to get interesting.  Remember the 5% performance lead that the KM133 held over the 730S in SYSMark 2000 when we first looked at the chipset’s theoretical performance?  That performance lead has disappeared and instead turned into a one point deficit, the only difference being that in this case we are using half as much memory and are using the integrated video instead of an external display as would be the case with an OEM system based on this chipset.

The 5% performance lead the KT133 held over the KM133 has extended to an 8% lead, just by forcing the KM133 to use its ProSavage integrated video core instead of the same graphics accelerator as the KT133. 

So much for hoping that the KM133 would do the Duron justice.

In our SiS 730S Review we complained about the 730S being 10% slower than the KT133 + TNT2 M64 combo here.  However we never thought that the KM133, when relying on its integrated video could be another 10% slower than the 730S. 

When paired up with the KM133 chipset, the Duron 800 is actually slower than the Celeron 766.

Under Content Creation Winstone 2001 the performance difference isn’t as noticeable as it was under Business Winstone 2001, however the KM133 isn’t dominating the charts as we had originally expected it to. 

This brings up a very important question, why? 

With the only real variable changed between the theoretical chipset performance benchmarks and the real world system performance benchmarks being the use of the integrated video, there is only one thing we can attribute the poor performance of the KM133 to: 2D performance.

When was the last time that we were ever concerned with 2D performance?  It’s been quite a while indeed, however, one of the worst performers when it came to 2D has been S3.  The Savage4 was horrible and the Savage 2000 did not improve performance to the point where the solution was competitive enough with the rest of the market. 

Is this what’s holding back the KM133?  Let’s have a look at its 2D performance using Winbench 99.

Starting out with some basic business 2D performance tests we see that the 2D performance of the GeForce2 GTS is almost 70% greater than that of the ProSavage graphics core found on the KM133. 

And when compared to the SiS 300 which is the graphics core used in the 730S chipset, the ProSavage is still at least 10% slower. 

The High End graphics tests don’t seem to widen the gap any more between the ProSavage and the GeForce2, however the SiS 300 happens to gain a 20% advantage here. 

This confirms our suspicions that the KM133’s 2D performance is what is holding it back and what is actually allowing the SiS 730S to outperform it, in spite of the results of the first set of theoretical benchmarks we took you all through.

3D Gaming Performance using Integrated Video

Under Quake III Arena, the Savage4 that is a part of the ProSavage graphics core proves to be a higher performing 3D solution than the 300 chipset which the 730S uses.  However the close to 33% performance advantage here doesn’t mean much for OEMs and system integrators since very few value PC users will be purchasing either of these chipsets to play games with high performance expectations. 

Should the need arise for higher gaming performance, regardless of what chipset you’re using, the best option would be to make use of the external AGP 4X slot and disable the integrated video.

The performance advantage drops slightly to just over 21% in favor of the KM133 as the resolution increases and memory bandwidth becomes a limiting factor. 

The KM133’s performance advantage is cut into a third under UnrealTournament which is a very memory bandwidth intensive game considering that it employs high texture usage. 

Because of this, both the KM133 and 730S begin to perform much closer together as they are still bound by the same amount of memory bandwidth.

The TNT2 M64 takes a pretty big performance hit here as it must now rely on system memory instead of its 8MB on-card frame buffer.  Once again, the KM133 and 730S are performing very close to one another.

DVD & Hard Disk Performance

As you’ll remember from our SiS 730S Review, one of the major selling points for the chipset was that it’s integrated SiS 300 graphics core supports both motion compensation as well as iDCT (Inverse Discrete Cosine Transformation) in hardware.  This helps keep the CPU utilization while playing a DVD stream down to a minimum, provided that the decoder (in this case WinDVD) supports the features. 

Unfortunately, it seems as if the KM133 still uses the Savage4’s hardware motion compensation, which isn’t helping much in this case, providing extremely high CPU utilization figures.  You would be better off using a plain software decoder or trying to dig up an older software DVD decoder that would’ve shipped with a Savage4 based card that can properly take advantage of its hardware motion compensation engine. 

Another big issue that came up with the 730S Review was the extremely poor disk performance of the 730S under Windows 2000.  It turns out that the poor figures were due to poor driver support under Windows 2000, while under Windows 98SE the 730S can offer almost 80% of the maximum ATA/100 specification (which is reasonable once you take into account driver/OS overheads).

However, the KM133 would not provide us with greater performance than a 35.2MB/s burst rate regardless of what OS we were running under (using the 686A Southbridge).  The KM133 definitely needs to be paired up with the 686B Southbridge in order to compete with the 730S in terms of hard disk performance.

At the same time, the 730S needs improved driver support under Windows 2000 so that there isn’t such a great performance discrepancy between the two OSes. 

Final Words

Although the KM133 did meet our very high expectations in a theoretical sense, where it failed was where it most needed to succeed, in the real world tests.

Offering the same memory performance as the KT133 chipset, the KM133 chipset was set to be an instant success when its closest competitor was the SiS 730S. However, VIA's recent marriage to S3 left the KM133 with one of the slowest 2D graphics in the industry, in spite of using the newer Savage 2000 as the basis for its 2D core. This combined with the poor driver support that we have seen surrounding the Savage4 and Savage 2000 solutions from back when they were standalone graphics chips make the KM133 not as desirable as we originally hoped it would be.

In this case, the SiS 730S becomes much more appetizing than when we originally considered it. With all the cards on the table, it seems like SiS holds the better hand. With superior DVD and 2D performance as well as superior disk performance (under Windows 98SE for now), the resulting overall performance of the 730S is between 0 and 10% greater than the KM133. Factor in the driver maturity of the SiS 300 graphics core and you have a much better solution from SiS than what VIA is presenting with the KM133.

Can VIA turn things around? It is possible, however quite a bit of attention must be devoted to driver support for the ProSavage line of chipsets and they must invest a considerable portion of that attention to searching for 2D performance optimizations for the Savage 2000 2D core present in the KM133.

If you have to choose between VIA's KM133 and SiS' 730S, the better choice definitely lies with SiS. While the performance of the 730S isn't as high as we would like it to be, it's the best that is currently available in its price class whereas there are situations in which the KM133 with a Duron 800 is even slower than a Celeron 766 on an i815.

As far as the future goes, SiS will be releasing an updated version of their 730S, called the 740 by Q4-2001 which will boast DDR SDRAM support. And VIA will be releasing another UMA solution for the Socket-A platform later in 2001 with integrated Trident Blade 3D graphics instead of the ProSavage core.

Is there any other hope for the value Socket-A platform? Unfortunately it doesn't seem so. ALi won't be releasing any UMA chipsets for the Socket-A platform in 2001, at least not according to their roadmap. Instead, ALi will continue to use the Aladdin 7 as their UMA platform of choice for value PCs which, for those of you that aren't familiar with it, is a Super7 chipset for Socket-7 processors.

There you have it folks, the two UMA solutions for the Duron have been presented to you in their entirety. Now it's up to AMD to keep prices on the Duron low and the performance competitive, and let's watch the value PC market segment get competitive once again.