Last year turned utterly fruitful for VIA. The company contrived to occupy good 50% of the market, since Intel pursued an inadequate policy in the chipset sector and there was hardly any rivalry on the part of other Taiwanese core logic manufacturers. However, VIA is bound to face strong competition this year and it will have to care a lot about its leadership, which promises to be a hard task to fulfill. First, as far as chipsets for Intel processors supporting SDR SDRAM are concerned, Intel got a fair chance to rock VIA's positions after it launched its i815EP in late 2000. It is a version of i815E missing the integrated graphics and is similar to Apollo Pro133A chipset by VIA. Second, although VIA has long been the only company shipping chipsets for AMD Socket A processors, it risks to lose its dominance. ALi, SiS and AMD itself have already launched their own chipsets for this platform. Moreover, a number of other companies like NVIDIA, Micron and probably ATI are expected to join the core logic manufacturers this year. This way, the future looks no longer peachy for VIA.

So, the company's major headache in 2001 will be to preserve its present achievements. Namely, in order to maintain the sales volume of 693A and 694X, VIA has made a price cut and re-positioned them as Low-End chips. That means that VIA will have to introduce a new mainstream solution to compete with i815EP and other follow-on chipsets for Pentium III CPU. Considering the latest trends, this chipset should support DDR SDRAM that has a greater memory bus bandwidth than the common SDRAM, but in theory it won't be much more pricy. In this review we will discuss VIA's new product that meets all these requirements - Apollo Pro266 chipset.

Chipset

The first chipset for Socket 370/FC-PGA processors supporting DDR SDRAM, VIA Apollo Pro266, differs tangibly from its predecessor, Apollo Pro133A. To make this point clear you may consult the following scheme:

As you can see, a distinguishing feature of Apollo Pro266 is the support of PC1600 and PC2100 DDR SDRAM. Since there is no great physical difference in the interface of the ordinary SDRAM and DDR SDRAM, this chipset can work with both memory types. Theoretically, it allows the manufacturers to design mainboards with 168-pin slots for SDRAM alongside with 184-pin DDR SDRAM DIMMs. But the stumbling stone is that it is quite hard to allocate such interface layout on the board. Therefore, just a few manufacturers have ventured to announce mainboards of the kind, although solutions like that offer greater flexibility and look rather promising.

Unlike AMD and ALi, VIA is experienced enough in creating products where the processor and the memory buses work asynchronously. No wonder that VIA Apollo Pro266 sticks to its forerunners and supports asynchronously working memory buses. This feature will let you use the cheaper PC1600 DDR SDRAM with the processors supporting 133MHz FSB and consequently PC2100 DDR SDRAM with older CPUs supporting 100MHz FSB. All this makes the memory controller of Apollo Pro266 highly flexible.

In developing its new chipset, VIA doesn't confine itself only to adding the DDR SDRAM support to the memory controller. Another innovation is a V-Link bus to connect the two Bridges of the chipset. VIA rejected to use the traditional PCI bus in this case for a very simple reason: the 133MHz memory bus bandwidth that the PCI bus provides is not enough to secure the required data transfer between the controllers implemented in the Bridges. Indeed, three fourths of the bandwidth will be needed to only transfer the data from the hard disk drive supporting ATA-100. Still, the South Bridge contains some other devices also requiring an intensive data transfer, for example, an integrated USB controller or a network controller. To solve this problem, it is quite natural to use new buses to connect the chipset bridges. Intel, for instance, has already switched over to using a new bus between the hubs in all the chipsets of the 800 family, which is two times faster. Now it is VIA's turn.

V-Link bus, which VIA has taken to connect the Bridges in Apollo Pro266, is a PCI-like 32-bit bus working at 66MHz. This way, the bandwidth of this bus is 266MB per second. The PCI controller that used to be hosted inside the North Bridge is now moved to the South Bridge, because its place is occupied by the V-Link controller. Subsequently, VIA Apollo Pro266 architecture is very much like that of new Intel chipsets, i.e. the North Bridge supports only the processor bus, system memory and AGP 4x.

It is also worth mentioning that VIA plans to bring the frequency of the V-Link up to 133MHz to double its bandwidth. However, it is most likely to be introduced only in Pentium 4 chipsets at the earliest.

North Bridge		South Bridge

Using V-Link as a new bus to connect the Bridges required a new South Bridge as well. VIA Apollo Pro266 chipset comprises VT8633 North Bridge and VT8233 South Bridge. The functions of VT8233 South Bridge resemble those of the widely used VIA686B, but unlike the latter, VT8233 allows connecting up to six USB devices and supports ACR, an alternative to CNR, which is backward compatible with AMR.

Unfortunately, we should point out that VIA Apollo Pro266 doesn't support Pentium III processors with Tualatin core manufactured with 0.13-micron technology that are due in Q3 2001. To compensate for this drawback, VIA should launch an improved chipset version, Apollo Pro266T, featuring a new VT8653 North Bridge. Working alongside with its partner-chip, it will support up to four 64-bit PCI slots working at 66MHz. A chipset with such characteristics would enable VIA to strengthen its positions as a server chips manufacturer. Although even the current Apollo Pro266 will also support dual-processor configurations.

DDR SDRAM and Intel Pentium III

Before we turn to the tests we ran with a VIA Apollo Pro266-based system, let us try to figure out what one expects from the use of DDR SDRAM with Intel Pentium III CPU. If we compare the memory and processor bus bandwidths in systems built on Athlon with those in systems built on Pentium III, we will see pretty well that using faster memory with higher bus bandwidth is more efficient for Athlon platforms. The matter is that the bandwidth of this processor bus (1.6GB per second with 200MHz frequency) is greater than that of PC133 SDRAM (1.06GB per second). That's why in Athlon systems with PC133 SDRAM the memory bus is a real bottleneck and increasing its bandwidth should have a positive effect on the overall system performance.

As for Pentium III, the bandwidth of the processor bus is equal to that of PC133 SDRAM and makes 1.06GB per second. In this case, DDR SDRAM is used, the system turns unbalanced since there appears a bottleneck: the bandwidth of the CPU bus doesn't suffice to make full use of the DDR SDRAM advantages, as DDR memory has a bandwidth of 1.6 or 2.1GB per second. As a result, it is evident that DDR platforms for Pentium III will by no means provide the same performance gain as Athlon-supporting systems.

As you can see from the chart above, VIA Apollo Pro266 is the only chipset with FSB bandwidth lower than the memory bus bandwidth.

In one of our latest reviews dedicated to Gigabyte GA-7DX mainboard working with Athlon CPUs and PC2100 DDR SDRAM we reported a 20% performance gain. Nothing of the kind is to be awaited from Pentium III. So, the use of DDR SDRAM in Pentium III systems is justified only if DDR SDRAM and common PC133 SDRAM modules get more or less close in price.

Then why did VIA stake on a DDR chipset for Pentium III having launched it first? First, there are no competitors in this field. Apart from VIA, only ALi thinks of manufacturing a DDR chipset for Pentium III (its mass production will start no earlier than in Q2), while Ali and AMD have made DDR chipsets for AMD's processors. Besides, ALi Aladdin Pro 5 chipset supporting DDR SDRAM and Socket 370 CPUs isn't so smart as Apollo Pro266 from VIA. Namely, it supports only six memory banks versus eight banks supported by Apollo Pro266 and it features no ECC support. So, the climate on the market is mild enough for Apollo Pro266 to appear.

Second, like many other companies, VIA has been misled by Intel's plans. Let us remind you that up to the very end of 2000 Intel intended to increase FSB frequency of its Pentium III processors up to 200MHz together with switching them over to the 0.13-micron technology and the new Tualatin core. Should it happen, using more efficient memory than PC133 SDRAM would be highly welcome. Then Intel was going to introduce the DDR SDRAM support in its on-coming Pentium IIIchipset, Almador.

But then again, in late 2000, Intel decided not to use a faster bus in the Tualatin family. Consequently, Almador was canceled as well. As for VIA, it failed to react quickly enough that's why the Apollo Pro266 family remained one of its basic products for this year.

Closer Look

For our tests we received a reference board based on VIA Apollo Pro266 chipset. Here it is:

This mainboard is designed in ATX form-factor and is equipped with three 184-pin DIMM slots for DDR SDRAM. Furthermore, the board features one AGP, four PCI and one ACR slot.

It makes no sense to describe the features of this mainboard in detail, as it is just an engineering sample meant for analysis and control of VIA Apollo Pro266 chipset performance. Thus, the low stability the board showed in our tests (especially when tested with PC2100 DDR SDRAM) in now way means that the real boards built on this chipset, which are to enter the market in Q1, will also be unstable.

The chart below gives the main characteristics of real mainboards built on VIA Apollo Pro266:

The list is really impressive. Now one can only wait for VIA to start shipping its Apollo Pro266 in mass, and believe us, various mainboards built on this chipset will hardly keep you waiting for long after that.

Testbed and Methods

We compared the performance of a VIA Apollo Pro266 based mainboard working with DDR SDRAM with the performance of systems built on different Pentium III chipsets supporting PC133 SDRAM: i815E and VIA Apollo Pro133A.

For our tests we also selected a former performance leader - a classical chipset, i440BX, working in an over-nominal mode (at 133MHz FSB frequency). In this mode the AGP bus works at 89MHz instead of the ordinary 66MHz, but we have chosen a graphics card which performs well in these conditions. As for the potential performance gain the overclocked AGP bus provides, keep in mind that although i440BX supports only AGP 2x mode, the AGP bus bandwidth is still greater in case we have mainboards supporting AGP 4x.

Our site has never neglected the interests of extreme overclockers. That's why we have tested a system built on i440BX chipset, clocked up to 150MHz FSB. Don't forget that in this case the AGP bus works at 100MHz (compare with the nominal 66MHz) and the memory bus reaches the top of 150MHz. Well, the system of the kind had to be assembled only of high-quality components to endure its work at over-nominal frequencies.

So, we assembled the following testbeds for the investigation:

All the systems worked with Microsoft Windows 98 SE.

As it goes from the table, the CPU on i440BX-based platform with the FSB clocked up to 150MHz performed at a little bit lower frequency than the CPUs in other systems. We had to arrange it this way because it was impossible to set the frequency of the CPU at 1GHz with the FSB running at 150MHz.

Performance

As always, we start our tests with SiSoft Sandra benchmark. In the case discussed this synthetic test is especially interesting to run, since there is a platform with PC150 SDRAM built on i440BX chipset - the fastest core logic for SDR memory hitherto.

The first test brings the first surprise. The platform built on i440BX with PC150 SDRAM has surpassed VIA Apollo Pro266 with PC2100 DDR SDRAM in performance. Obviously, it is due to the well-optimized memory controller with low latency. It gives i440BX-150 a tremendous boost when it comes to memory operations via ALU.

When the data transfer is carried out via FPU, that is in Double Words, everything goes the right way. VIA Apollo Pro266 with PC2100 DDR SDRAM proves the best. However, the results of this chipset with PC1600 DDR SDRAM are disappointing. VIA Apollo Pro266 with this type of memory turns out even slower than VIA Apollo Pro133A. Evidently, the asynchronous work of the memory and the FSB buses is not that easy to implement in DDR chipsets.

This test models the system functioning in standard office applications. It indicates that it makes no sense using the high-performance DDR SDRAM for business tasks. Combining PC2100 DDR SDRAM with Pentium III CPU provides a really insignificant performance gain in this type of applications.

The idea of this test is almost identical to that of the previous one. The only difference is that Content Creation Winstone 2001 models the system functioning in content creation applications, such as Adobe Photoshop and Macromedia Dreamweaver. This benchmark shows that Intel chipsets with PC133 SDRAM work faster than VIA Apollo Pro266. Nevertheless, in comparison with the previous chipset by VIA, Apollo Pro133A, the freshly made VIA Apollo Pro266 proves to be 3% tougher.

SYSmark 2000 measures the average CPU performance in 12 standard applications and gauges the result obtained in every one separately. As we have expected, Intel chipsets are the best again. Anyway, this test doesn't reveal any great performance difference between all the systems. Let's take a closer look at the results in SYSmark 2000.

In this benchmark VIA Apollo Pro266 appears very close to Intel chipsets. By the by, a curious fact is that the leadership in office applications belongs to Intel 815 but not to an overclocked i440BX.

In content creation applications VIA Apollo Pro266 has nothing to boast. This DDR chipset got the lowest score of all the chipsets with PC133 SDRAM, although they have smaller memory bus bandwidth.

This time we have also paid attention to the professional 3D-modelling software called 3D Studio MAX R3. To estimate the performance of our review participants we checked how long it takes them to render Anisotropic Wheel scene at the resolution set to 800x600. So, the smaller value denotes faster performance. Surprising as it might seem, almost all the platforms demonstrate equal abilities, since the major workload during the scene rendering in 3D Studio MAX R3 is laid upon the processor FPU and not upon the system memory bus.

This is one of the few applications where VIA Apollo Pro266 performs well. Data compression requires transferring considerably larger amounts of data, and systems with the more efficient DDR SDRAM are much better at it. However, the platform built on i440BX with a bus clocked to 150MHz keeps pace even with Apollo Pro266 chipset supporting PC2100 DDR SDRAM.

We have also taken the pains to check how fast the tested systems are at encoding mp-3 files in a popular AudioCatalyst application. For this purpose we made use of a specially created 100MB wav-file. The results obtained here are very much similar to the preceding diagram. There is nothing to wonder at: mp-3 encoding is actually a kind of compression as well though following a specific algorithm.

As a whole, we can admit that DDR SDRAM grants no appreciable performance gain in real office and business tasks, as well as in content creation applications. So, a feeble 3% performance gain is the maximum Pentium III users will be able to achieve in these applications when they move to DDR SDRAM. That is another proof for the fact that increasing memory bus bandwidth is unjustified without increasing that of the CPU bus in its turn.

Now we suggest taking a glance at how things stand in gaming applications, but first let's inquire about the results of the synthetic 3DMark2000:

It's natural to compare the results shown by the boards built on VIA Apollo Pro266 and VIA Apollo Pro133A chipsets in order to see performance gain in Pentium III systems obtained due to DDR memory. 3DMark2000 states a 7% gain in this case. Of course, it is twice as low as in case of a DDR SDRAM system built on AMD-760 core logic with Athlon CPU, but still the gain exists. At the same time don't forget that VIA chipsets are known to have lower performance than those from Intel. This way, if we take VIA Apollo Pro266 and, for instance, Intel 815 to compare, we shall see just a 3% difference. And the still faster i440BX-133 nearly reaches the level of the PC2100 DDR SDRAM platform.

At higher resolutions the performance of the graphics subsystem comes to the forefront. That's why the memory type chosen doesn't influence the overall system performance that much. In the long run, all the platforms show similar results in this test.

While testing DDR platforms for AMD Athlon processor we noticed that DDR SDRAM works most successfully in Quake3. Systems built on AMD-760 chipset were 20% faster in Quake3 than their rivals with PC133 SDRAM.

In this case we again face some performance divergence between the platforms with different memory types. However, the performance of VIA Apollo Pro266 is only 7% higher than that of VIA Apollo Pro133A. For this reason, the better optimized i440BX chipset with both 133MHz or 150MHz memory bus overtakes the system built on VIA Apollo Pro266 and equipped with PC2100 DDR SDRAM.

Higher resolutions make the gap not so wide, but the main trends don't change at all.

Quake3: Team Arena works with more complicated textures, so the difference in results is a bit slighter than in a common Quake3. In this test VIA Apollo Pro266 is not more than 5% faster than VIA Apollo Pro133A and falls behind both i440BX based systems again.

At the resolution set to 1024x768x32 we can observe practically the same situation as in Quake3.

Unreal Tournament demands huge resources from the PC, so the memory bus bandwidth matters a lot. This helps the system built on VIA Apollo Pro266 chipset with PC2100 DDR SDRAM show good results in Unreal Tournament, where it keeps the second position after the overclocked i440BX based system. There is nothing extraordinary about that, however. The i440BX based system with 150MHz bus frequency, which can boast fine results in all the tests, owes a lot to its fastest CPU bus working at 150MHz.

When the resolution gets higher, the amount of data transfers rises as well, enabling VIA Apollo Pro266 with PC2100 DDR SDRAM catch up with i440BX-150.

The results of the platforms in MDK2 are quite predictable. The performance of the VIA Apollo Pro266 based platform falls between those of Intel 440BX-133 and Intel 440BX-150.

As a whole, we can observe the same thing as in Quake3.

Our last gaming test is a demo version of Mercedes-Benz Truck Racing. This truck racing simulator is a vivid representative of the latest generation games that use a lot of textures and a complex geometric and physical model. However, in this demo at 640x480x16 resolution the system with PC2100 DDR SDRAM lags behind all the platforms built on Intel chipsets with PC133 SDRAM.

Like in most other cases, higher resolutions lead to more memory requests, including those via AGP bus. So, it's high time to make use of a bus with a higher bandwidth than that of the CPU bus. That is why in this test VIA Apollo Pro266 makes a happy change and improves its latest results. The only chipset running ahead of the DDR system is i440BX-150 with an overclocked up to 150MHz CPU bus that consequently demonstrates higher performance.

Conclusion

VIA Apollo Pro266 is the first chipset for Intel Pentium III/Celeron processors with DDR SDRAM. However, as practice shows, Pentium III systems do not need higher memory bus bandwidth so badly as those built on AMD Athlon. The trick is that as soon as a Pentium III platform gets faster memory, its CPU bus becomes a sheer bottleneck that hinders any further increase in performance. The whole affair ends in a distress, as the top performance gain one can get in Pentium III systems with DDR SDRAM doesn't exceed 7%, while the gain on Athlon platforms reaches the peak of 20%.

To say more, since VIA chipsets work slower with the memory than Intel chipsets, very often VIA Apollo Pro266 turns out no faster than Intel chipsets supporting PC133 SDRAM. In general, we can state that systems on VIA Apollo Pro266 with PC2100 DDR SDRAM perform close to those on i440BX with the FSB overclocked to 133ÌHz. As for the same i440BX based platforms with a FSB clocked up to 150MHz and equipped with PC150 SDRAM, they are unreservedly faster than VIA Apollo Pro266 in almost all the tests. That is another striking demonstration of how important it is to increase the bandwidth of both memory and CPU buses.

As far as PC1600 DDR SDRAM on Pentium III platforms goes, our tests illustrate it clearly that the asynchronous work of DDR memory and CPU bus leads to outrageously poor results. Pretty often VIA Apollo Pro266 with PC1600 DDR SDRAM was even slower than VIA Apollo Pro133A. Now it is no longer surprising why most mainboard manufacturers give up the support of an asynchronous memory bus in their products.

As a final word, we wish to highlight the notable progress VIA has revealed in its newly made VIA Apollo Pro266 chipset. This core logic doesn't only support the high-performance DDR SDRAM, but is also the first representative of the chipsets with a new architecture featuring the innovative V-Link to connect the Bridges. Unfortunately, VIA didn't find the right application field for its new product. The systems on VIA Apollo Pro266 (with DDR SDRAM, of course) will hardly become a popular solution unless their price gets down and becomes more or less close to that of PC133 SDRAM systems.