by Shui-Che Lim

Background

Most people consider the CPU the "brains" of a modern PC. This perception, while logical, is not entirely correct. I prefer to think of the CPU as the engine which drives the whole system. The real brains of the modern PC is in a relatively simple semiconductor component called the core logic chipset. The chipset acts as the "traffic cop" of the system, resolving requests for access to and from the CPU to main memory, cache memory and a host of I/O functions.

Most core logic chipset devices are relatively simple affairs containing anywhere between 50,000-70,000 logic gates, as opposed to the millions of gates present in a modern CPU. However, it is this device which determines the capabilities of the entire system. Modern chipsets contain a memory controller, cache controller, PCI bus controller, real-time clock, keyboard controller as well as other system logic.

Did you ever wonder why certain systems only support FPM and EDO DRAM and not the newer SDRAM? How about the newer Ultra DMA specification hard drives? Well, these are among some of the features defined and supported by the core logic chipset.

The Plot Thickens

In the past, there were many chipset vendors supplying a wide variety of core logic solutions. It was not uncommon in the era of 486 and even during the early Pentium years to find motherboards using chipsets from companies such as UMC, SiS, Opti, ALI and VIA.

Over the past several years, Intel has taken measures to become the premier supplier of chipsets for most motherboards. Whomever controls the chipset, controls the specifications for the entire motherboard and therefore the system. The methods Intel employed were deceptively simple and the results were brutally effective.

Intel, as the premier supplier of CPUs, reasoned with motherboard and system makers that they could supply a less buggy and more feature laden chipset together with the introduction of newer CPUs. Third party vendors such as UMC, SiS, Opti, ALI and VIA would always lag several months behind the introduction of newer CPUs. Thus, using Intel chipsets would allow motherboard and system makers to bring newer products to market with shorter leadtimes.

This sounded like a Good Thing ™and many motherboard and system makers bought this story. The end result is that Intel now controls about 90% of the chipset market. The remaining 10% is being fought over by the remaining third party vendors, some of which have gone out of the chipset business entirely. Therefore, just as in the CPU market, there is now also less variety and choice for the end user when it comes to chipsets. But is this really so bad?

Conspiracy Theories

The hardcore flight simulation community, more than any other segment of the gaming public, is constantly riding the bleeding edge of technology. But lately, more and more of us are complaining about the speed at which these changes are occurring… and is there any wonder? The time between introduction of ever newer CPU generations is shrinking, and Intel is currently pushing everyone onto their latest Slot-1 Pentium II platforms, all but abandoning the Socket 7 infrastructure.

Control of the core logic chipset has allowed Intel to control the migration of motherboard and system vendors and ultimately the end users. Intel succeeded in migrating the infrastructure from 486 to Pentium architecture with their 430FX, 430HX, 430VX and 430TX chipsets. They are now attempting to do the same with the 440 series of chipsets and migrate the infrastructure from Socket 7 to Slot-1.

Since Intel now controls the chipset market, switching the motherboard makers and system vendors over to Pentium II is as simple as announcing that they will no longer supply the Socket 7 430TX chipset. Manufacturers that don’t tow the line often have "allocation issues." This kind of control means the ability to determine direction in the marketplace.

Socket 7 vs. Slot-1

The real question, though, is what exactly does Slot-1 and Pentium II bring to the party that Socket 7 does not? Intel has been putting in overtime trying to convince everyone that Pentium II is far and away a better architecture than Socket 7. But is it?

AGP Support

In case you’re not quite up to date on the latest technological trends, AGP stands for Accelerated Graphics Port. It is another Intel specified bus specifically defined for graphics use. The AGP video bus standard was defined specifically to address 3D graphics bandwidth issues. Intel’s marketing machine is currently making AGP another one of those "must have" items, in the same manner as MMX marketing six months ago. Whether or not AGP will live up to all its hype is another story...

I won’t go into a lot of detail since sites like Tom’s Hardware has already done quite a bit of in depth articles and benchmarks on AGP vs. non-AGP video cards. In short, AGP works very much like the current PCI bus on standard PCs. However, instead of the 33MHz bus speed of the PCI bus, AGP defines a 66MHz bus speed.

In theory, an AGP video card should have roughly double the performance and bandwidth of non-AGP video cards on the PCI bus. Furthermore, the AGP 1.0 specification defines a 2x mode where AGP enabled software video drivers are able to access the card on both the rising and falling edge of the 66MHz clock signal. This effectively doubles the performance to 133MHz.

However, in practice, AGP stores most of it’s textures in local system memory instead of on fast local video memory like current PCI video cards. This has by and large obviated any real advantage of the 66MHz bus, since local video memory (such as 100MHz or 125MHz SGRAM, WRAM, etc.) has much higher bandwidth than AGP to system memory. Moreover, full 133MHz AGP support won’t generally be available until Windows 98 ships.

Finally, AGP shows its strengths only when there are very large textures that need to be manipulated. While games with larger textures can be expected in the future, current games such as Longbow 2 and F22: ADF cope with 2MB or 4MB texture memory limits quite reasonably.

100MHz Host Bus

This is another performance enhancing feature that Intel is touting for Slot-1. Currently, the CPU to memory bus is running at 66MHz for most CPUs in the market today. Increasing the host bus from 66MHz to 100MHz would increase CPU to memory performance by roughly 50%. Most users who have overclocked their CPUs and system buses to higher speeds such as 75MHz or 83MHz will understand what this does for the overall performance of the system.

Since the CPU spends most of it’s time reading and writing from cache or main memory and executing instructions, anything that increases the CPU to memory bandwidth will have the greatest effect on overall system performance. The big difference here is that 100MHz will be fully and officially supported by future CPUs and system board makers.

Go to Part II

© 1997 - 2000 COMBATSIM.COM, Inc. All Rights Reserved.