A quick review of Intel's chipsets is needed to determine the state of today's market. Ironically, the Intel 440BX, introduced more than two years ago, continues to be Intel's mainstay chipset for the desktop/business segment. The 820 chipset was suppose to be the 440BX's successor. But due to technical difficulties, the 820 didn't launch until November/1999. Originally, the 820 was to be paired up with Rambus memory; but the cost of Rambus as compared to SDRAM was about five times as much. And the performance gain of Rambus memory was marginal at best. Thus Intel introduced their Memory Translator Hub (MTH) chip to be use with Intel 820 based motherboards; this allows the 820 chipset to communicate with SDRAM memory.

Unfortunately, there was a performance hit associated with the 820/MTH combination. According to some benchmarks, the 820/MTH is slower than a 440BX system. And when Intel launched their next generation "Coppermine" Pentium III processors running at the faster 133MHz front side bus (FSB), only two Intel chipsets were able to support it: Intel 810/810E and Intel 820. But the 810 series came with the integrated Intel Direct AGP video and did not offer an AGP slot. The 820, with ATA-66 and AGP 4X support, was the ideal chipset to be used with the Coppermine processor. The downside is, if you choose 820/Rambus then you'll be paying a high price, while selecting the 820/SDRAM means you may take a performance hit.

At this time, the 440BX chipset was starting to show its age: it couldn't officially support the 133MHz FSB, and only had ATA-33 and AGP 2X support. But still, the 440BX equipped with a 100MHz Coppermine CPU and PC100 SDRAM was a viable system for most people. It was during this time that VIA, a Taiwanese based chipset maker, decided to fill the void. VIA launched their Apollo Pro 133 (VIA 693X) chipset that allow system builders to use the 133MHz FSB Pentium III processor with SDRAM running at PC133, and offering ATA-66 and AGP 2X. While Intel continued to struggle with their 820 problems, VIA quickly followed up with the Apollo Pro 133A (VIA 694X) chipset, which added AGP 2X/4X support.

Finally, the blow that killed the 820/SDRAM solution was the recall of all 820 with MTH based motherboards due to possible SDRAM timing issue. 820/Rambus systems were not affected by the recall; while prices of Rambus memory are slowly coming down, they are still about three times the cost of SDRAM. By late June 2000, Intel will release their new 815 "Solano" chipset (which will have native PC133 SDRAM, ATA-100, and AGP 2X/4X support) to fill the gap left over by the 820/SDRAM recall. But in the meantime, if you're looking for fast and affordable PC133 SDRAM, ATA-66, and AGP 4X support, the VIA Apollo Pro 133A is the only game in town.

Here's the summary of the components used in the test system:

System Board	Premio V94S (MS-6309/210630)
Chipset	VIA Apollo Pro 133A (694X)
BIOS	AMI version 1.4 - 05/17/2000
Processors	Intel Pentium III FC-PGA @ 100MHz and 133MHz FSB
Memory	128MB PC133 SDRAM
Hard Drive	Western Digital 20GB 7,200RPM ATA-66
CD-ROM	Creative Labs 52X ATAPI
Video Card	Matrox Millennium G400 AGP
Operating Systems	Microsoft Windows 98SE and NT 4.0 Workstation SP6

The table below is the list of Pentium III FC-PGA processors tested. The 100MHz FSB processor are listed in blue, while the 133MHz FSB CPUs are listed in green. The "B" designation after the CPU speed means the CPU is running at 133MHz.

Processor	L2 Cache	CPU and Memory Bus Speed
550	256K	100MHz
600	256K	100MHz
650	256K	100MHz
700	256K	100MHz
750	256K	100MHz
800	256K	100MHz
850	256K	100MHz
533	256K	133MHz
600B	256K	133MHz
667	256K	133MHz
733	256K	133MHz
800B	256K	133MHz
866	256K	133MHz
933	256K	133MHz

Ziff-Davis Winstone 99 measures overall system performance by running through a series of commonly used business and engineering applications. Higher number means better performance. Winbench 99 measures the performance of a specific part of the system; such as CPU or video performance. For both test suites, higher numbers mean better performance.

The chart above measures a system's overall performance under Windows 98SE. It is interesting to note that the fastest 100MHz FSB processor at 850MHz is outperformed by the 800MHz processor running at 133MHz FSB. To compare the performance difference between the 100MHz and 133MHz FSB processors, examine the numbers for the 600/600B and 800/800B. Is this correct? Take a look at the numbers below for Windows NT 4.0.

The graph above shows the relative system performance under Windows NT 4.0. A similar situation exists under NT 4.0. The 850 clocks in at 28.8, while the slower speed 800B is logging 29.8. But how can this be? By common sense, a higher clock speed processor should always outperform a slower one. Yes, that would be true if and only if raw CPU clock speed is the only factor. Since the Premio V94S is based on the VIA Apollo Pro 133A chipset, one must also take into account the CPU host and memory bus speed. For the 800B processor, both the CPU and memory bus are running at 133MHz. While the higher clock speed 850 processor runs both buses at a slower 100MHz speed. This is a perfect example of bus speed being more important than raw CPU speed. Keep this in mind when you're configurating a system.

Next