We just took a look at VIA's latest chipset, the Apollo Pro 133A in Part 1of this review. On paper, the solution seems the ideal alternative to Intel's 820 chipset; unfortunately that may not hold true in the real world. History has already taught us that VIA has generally offered slower chipset solutions than Intel although, for a while, VIA held the performance throne in the days of the simple Socket-7 platform, but in the past 2 years Intel has definitely separated from VIA in terms of performance.

VIA's original Slot-1 chipset, the Apollo Pro+, was noticeably slower than the BX chipset and offered no real feature advantages and was therefore a silly alternative to Intel's solution. However, with the price of motherboards outfitted with Intel's 820 chipset expected to be reasonably high and adding to that the costs associated with Rambus DRAM, migrating to the i820 platform doesn't seem to be a very cost effective alternative. Ideally using SDRAM with the i820 would help keep costs reasonable, but there is an inherent problem with this approach.

The i820 chipset does not natively support SDRAM, either because Intel wants to force manufacturers into pursuing RDRAM as the only memory option or simply because they want to save on cost. In order to gain SDRAM support on an i820 motherboard, the manufacturer must use a Memory Translator Hub (MTH) to allow for the SDRAM slots to be accessed by the Memory Controller Hub of the 820 that can only "talk to" RDRAM. The situation is like having a translator convey a message you speak in English to a friend that can only speak Japanese; the translation is often slower than if you could speak the original sentence to your friend in English. Two problems present themselves with this approach -- using SDRAM with an MTH will most likely be slower than SDRAM on a BX board and the cost of adding the MTH will also be reflected in the overall cost of the motherboard. Where does VIA come in?

The Apollo Pro 133A offers everything the i820 does, without the requirement for RDRAM. This difference leaves VIA in a very advantageous position, because they are providing a cheaper solution that supports the industry standard in memory (memory can still be run at 100MHz even if the FSB is at 133MHz) and the solution should theoretically be faster than the Intel BX and should theoretically prove to be a decent competitor to the Intel 820 chipset.

The Motherboard

Micron is proud to say that they are the first manufacturer to be shipping a complete system based on the Intel Pentium III 600B processor. How are they doing this? By using the Apollo Pro 133A as the heart of their system, allowing for them to sell a nicely configured system based on the 600B at under $2000 while being able to boast that they support the 133MHz FSB.

Click to Enlarge

Now, normally when a company like Micron is building a flagship system like their Millennia MAX 600, they go to Intel or another top tier motherboard manufacturer for their motherboard. However, with the unique nature of the MAX 600 (it requires a VIA Apollo Pro 133A chipset) they had to look elsewhere. Luckily, Tyan came to the rescue with their Apollo Pro 133A based motherboard, the S1854 Trinity 400.

The Trinity 400 features a total of 6 PCI slots (1 ISA slot optional), 1 AGP 4X slot, and 3 DIMM slots. Tyan wisely chose to integrate real hardware sound on the motherboard with the Creative ES1373 chip. The board features both a Socket-370 and a Slot-1 CPU interface that help it offer a greater range of configuration options in systems, making it ideal for Micron's line of products.

The Trinity 400 also features the elusive standardized front panel LED connector that we're beginning to see on motherboards. This is a definite step forward and we're more than happy to see Tyan supporting it.

The board itself is a final board and we were lucky enough to receive a sample in lab to perform all of our tests on. Unlike i820 motherboards, the S1854 supports the 66MHz FSB frequency courtesy of the Apollo Pro 133A chipset.

The Memory

We used three different memory types in the performance comparison. For the i820 test system we used Samsung PC800 RDRAM. We did not include slower PC700 and PC600 benchmarks simply because we already know the decreased performance these two frequencies offer, illustrated in our review of the i820 chipset.

The i440BX and Apollo Pro 133A test beds used, by default, Samsung PC133 modules equipped with the Samsung 'GA' chips. These chips are fully PC133 compliant; however, they are only rated at 133MHz - CAS 3. Because of this, all memory timings for SDRAM were lowered to 3-3-3.

The Apollo Pro 133A test bed also made use of our NEC Virtual Channel SDRAM rated at 133MHz. Pictures of the modules are featured below:

RDRAM

VC-SDRAM

Prelude to Performance

We already know from our review of the i820 chipset what we can expect from the 133MHz FSB. The performance increase the 133MHz FSB offered in most business and gaming applications was negligible and the only real increase came under 3D Studio MAX.

The performance of the Apollo Pro 133A is already somewhat known because of the benchmarks we published in our review of the AGP 2X version of the chipset, the Apollo Pro 133. In light of this knowledge, we geared our performance analysis of the Apollo Pro 133A more towards exploring the performance advantages of VC-SDRAM and the AGP 4X implementation of the chipset.

As an upgrade path, the Apollo Pro 133A will offer support for many of the features which the BX inherently cannot. Many 820-weary users will be looking for a viable alternative presenting us with the big question to answer -- is the Apollo Pro 133A faster than the good ol' Intel BX chipset? Let's find out.

The Test

Click here for detailed descriptions of all of the tests run.

Business Winstone 99 is a good test for measuring performance of different CPUs under business applications, but the test varies too greatly to provide accurate indications of performance differences among very similar configurations. By truncating the score down to two digits (tenths of a Winstone point are meaningless in terms of real world performance) we get the breakdown of scores, as seen above.

The improvement of the 133MHz FSB and newer memory types in business applications is less than 4% at most. Realistically, you won't notice the performance difference at all. The point that could be made here is that the Apollo Pro 133A is as fast as the Intel 820, but the BX is virtually as fast as either of those two options so discussing performance here is pretty much worthless.

If all you're going to be running are business applications then you don't need to be even considering a 133MHz FSB platform, the Celeron exists for a reason you know :)

In overall system performance the addition of Virtual Channel SDRAM brings the Apollo Pro 133A is slightly above the 820 + RDRAM combination. The performance advantage isn't huge but neither is the performance advantage the 820 + RDRAM offers over the good ol' BX. This brings us to point number two; even while factoring in content creation and office applications, overall system performance is left relatively unaffected by the 133MHz FSB and newer memory types. The greatest performance difference illustrated here is an amazing 3%, not worth your time nor your money.

If SYSMark 98 and Business Winstone 99 characterize your usage patterns, then don't even bother with either of the two 133MHz FSB platforms.

Here is where we begin to see some indication of a performance advantage brought to the table by the two 133MHz FSB chipsets. Intel's Naturally Speaking benchmark does not portray the performance of RDRAM in the brightest light, which is why it is conspicuously absent from Intel's performance disclosure of the i820 platform.

In this test, the Apollo Pro 133A uses its lower latency SDRAM to its advantage which grants it the small performance advantage over the i820. In reality, with the entire range of performance values falling within about 7% of each other, arguing over performance here is pointless as well. The main thing to point out is that the Apollo Pro 133A comes out on top once again, with the addition of VC-SDRAM offering a very tiny improvement to the performance of the platform.

Once again, the Apollo Pro 133A comes out on top and, once again, not by a very large margin. The lower latency SDRAM has the advantage in this test over the RDRAM on the i820 platform and the addition of Virtual Channel architecture increases performance by another percent, nothing too revolutionary at all.

Adobe Photoshop was used as Intel's tool to illustrate the true performance advantage the i820 + RDRAM platform offered, and it does so in this test as well. Even when armed with VC-SDRAM, the Apollo Pro 133A can't beat the i820, but once again the performance difference is only 2%, not a very large difference.

While we're not huge fans of synthetic benchmarks we needed a way to test VIA's AGP 4X implementation so we used Intel's Platform Bandwidth Test that simply runs a quick test to measure the throughput of the AGP bus. As you can see, the Apollo Pro 133A falls close to 30% behind what the i820 is capable of pumping through the AGP bus. The addition of VC-SDRAM accounted for an extra 3% increase in performance, but nothing that could hope to bring the Apollo Pro 133A up to par with the 820 in AGP performance.

Using the AGP 4X Transfer Test (a part of Intel's AGP test suite) we can begin to see how bombarding even the GeForce 256 with textures can force it to almost rely on the AGP bus, in this case the Apollo Pro 133A is 17% slower than the i820 and 10% slower if VC-SDRAM is used, but it is still not up to par with the i820.

NVIDIA claims that their GeForce 256 works at peak efficiency in AGP 4X systems, so in order to use this to our advantage in a test situation we fired up Quake 3 Arena and tweaked a few settings. The first thing was to bump up the texture detail slider in the options screen so the most detailed and thus largest textures were used. The second tweak was to implement the special higher polygon settings offered in Quake 3 that come without a performance penalty to the GeForce 256 courtesy of its hardware transforming & lighting engine. Those settings can be found below:

r_lodbias -2
r_SubDivisions 1
r_lodCurveError 10000
r_PicMip 0

It turns out that 640 x 480 was the best setting for illustrating potential performance differences between the platforms, and as you can see above the performance differences were minimal. The BX is obviously slower than the rest because of it's AGP 2X and 100MHz FSB limitations, and the 820 is on top as a result of its superior AGP implementation to the Apollo Pro 133A.

3D Studio MAX is a very memory bandwidth hungry application which is why the 820 leads by such a large margin. The Apollo Pro 133A has a difficult time keeping up with the competition, even with its 133MHz FSB frequency. The use of VC-SDRAM bumps up the performance significantly, but it still barely outperforms the BX which is running at its 100MHz FSB.

Conclusion

As a purely gaming solution, the Apollo Pro Plus 133A isn't what we hoped it would be. The AGP performance tests indicate that VIA's AGP 4X implementation is around 30% slower than Intel's AGP 4X implementation on the i820. But wait a minute, isn't everyone always saying how big of a joke AGP is? The fact of the matter is that AGP still isn't used intensely during today's games and the AGP texturing benchmarks Intel supplied us with are exaggerations of what actual gameplay would involve, simply to illustrate the performance potential AGP 4X offers over AGP 2X.

In the future this may very well change and gamers may begin to truly notice a performance increase on AGP 4X platforms versus AGP 2X platforms. But as time goes on, the amount of local memory on video cards will definitely increase, with most boards shipping with 32MB now and the trend towards 64MB+ in the future, will we ever really become dependent on AGP?

Chances are that we won't, and VIA's slower AGP implementation will mean that only the slow points in games take the performance hit, when textures are already being sent over the AGP bus. So, instead of having a range of frame rates from 30 fps to 90 fps, the lower end of that range would drop to 20 fps.

We had hoped that the use of VC-SDRAM would clear up any performance issues in games with the Apollo Pro 133A, but, unfortunately, that was not the case. The performance improvement in games was minimal and it did not even raise the performance to beyond what the BX is already capable of achieving. The hard core gamer will probably end up footing the cost of RDRAM and migrating to the i820 platform, which isn't a bad decision, it's just not going to be a cheap one.

For the average user that is looking to take advantage of the new 133MHz FSB processors, the Apollo Pro 133A makes quite a bit of sense. If you're not too hung over on the issue of frame rates in games and are more concerned with overall system performance and cost then this may be just what the doctor ordered. The sub-par AGP 4X implementation of the 133A will keep it from being the gamer's dream, but not everyone is a gamer and there are people that are looking for better overall system performance, not necessarily in Quake 3 alone. For this type of a user, the Apollo Pro 133A makes perfect sense because it is cheaper than i820 (considering memory costs). It acts as more of a bridge between the current BX platform and the future platforms without requiring a large investment on your part (you can just re-use your old SDRAM).

For businesses looking to take advantage of the latest 133MHz FSB CPUs, the Apollo Pro 133A once again makes more sense than the BX chipset, since it keeps the upgrade path open to future Pentium III E processors (Coppermine) without requiring a large up-front investment.

From the standpoint of a professional user, BX is still the way to go if you're on a budget, otherwise it seems like the i820 is the only other single-processor option. The 3D Studio MAX benchmarks show that even with the added performance of the 133MHz FSB, it took the use of VC-SDRAM on the Apollo Pro 133A to beat out the BX at 100MHz by a measly one-tenth of a second. The Apollo Pro 133A is just not a high-end solution, but then again, it was never meant to be one.

In response to the question we posed before entering the performance analysis: is the Apollo Pro 133A faster than the good ol' Intel BX chipset? When using the 133MHz FSB, the answer is generally yes, but even with that FSB advantage it isn't faster by a large enough margin to justify upgrading, which is the consensus we came to on the issue of the 133MHz FSB when we took a look at the i820 chipset and the Pentium III B.  

The next question is: does Virtual Channel SDRAM 'buy back' some of the performance deficit the Apollo Pro 133A holds in comparison to the i820?  In some cases it does, such as the 3D Studio MAX test where memory bandwidth and the efficient use of it is critical to a high performing system.  But in other cases, like in our gaming and AGP performance tests it doesn't make up for the slower AGP implementation of the Apollo Pro 133A.

Of course, if performance is not your top concern, then the BX platform is still a very good platform to stick behind. The initial AGP 4X benchmarks have been very disappointing and although NVIDIA claims that their GeForce 256 works optimally in AGP 4X mode, the performance differences we've been able to note have been minimal at best. The performance improvement the 133MHz FSB offers is not great enough to justify a major upgrade and the current 133MHz FSB processors are offering a very small performance improvement if any at all over their 100MHz counterparts. While the latter will change with the Pentium III E, the performance of a Pentium III 450 on a BX platform is in no way sub-par.

It's funny to think that, while Intel is releasing 133MHz FSB processors there are still a large number of users out there that are more than happy with their 66MHz FSB Celerons that they picked up for under $100 and for them, the BX is doing just fine. Price is king in the world of competition.