Bitzenbytes.com

I’ve often remarked that the computer’s motherboard is the most important part you are buying...and the least talked about by sales staff, marketing materials, etc. In fact, with some vendors, it can be very difficult to find out exactly what motherboard you are getting in your new system.

Forum member cassandra suggested that I post a ‘whitepaper’ here to help everyone understand motherboards a bit better. I’ll include some graphics I’ve ‘gleaned’ from the web to help illustrate.

Despite the advice of those more sane than I, here goes...

What does a motherboard do?

Cool…let’s start out with the basics. A motherboard moves data between the Processor (CPU), the Level 1 (L1) and Level 2 (L2) caches, the Memory Subsystem and your devices. It performs this awesome feat through buses...electrical pathways by which these various parts communicate. There are a few different types of buses, each one running at different bandwidths; the amount of data they can carry, measured in bits (data capacity or width) multiplied by the speed at which they carry data, measured in megahertz. That will give you the bandwidth in bits; divide by 8 to get the bandwidth in bytes. The most familiar buses to most users are the ones named after the expansion slots that are connected to them (AGP, PCI, ISA, etc.) but there are other important buses too.

The chart above shows the Data Capacity (Width), the Speed and the Total Bandwidth (Width X Speed) for the various buses found in your computer. Keep in mind that the bandwidth is shared between all the devices on the bus; the total amount of data sent from all the devices plugged into a bus will never exceed the Total Bandwidth.

The Core-Logic Chipset:

The core-logic chipset…sometimes referred to as ‘chipset’ or ‘agp-set’ (we’ll call it ‘chipset’ here) is the thing that controls the movement of data between the various devices on your computer. The Intel 430TX, 440BX or the VIA KT133 are all examples of various chipsets. The chipset regulates the system speed and directs all instructions from the CPU and memory to the various buses.

How can you tell which chipset you have in your machine? You can quickly peek at the chipset/BIOS information that appears early in the boot process, or check your motherboard manual or motherboard manufacturer’s website.

The System Bus:

The System Bus connects the CPU to the core-logic chipset and coordinates the data flow for each device connected to the motherboard. The system bus on an Intel 440BX chipset, for example, is 64 bits wide and runs at 100 MHz, so the total bandwidth is 800MB per second.

The chipsets designed for AMD’s Athlon processor have a system bus speed of 200MHz. Because communication between the chipset and the system memory runs over its own channel, no other devices share this bandwidth. (That’ll be nice if they ever release any multiprocessor boards ).

The newer Intel chipsets support a 133MHz system bus, but they currently require the phenomenally expensive (and not as fast as promised) Rambus DRAM (RDRAM).

The Backside Bus:

The Backside Bus connects the CPU to the Level 2 (L2) cache. It keeps the CPU from wasting clock cycles retrieving data from the much slower system memory or hard drives. Data can only stay in the L2 cache for a few seconds before being refreshed, but that’s plenty of time for the backside bus to find the data and send it to the CPU. The backside bus is 64-bits wide.

For the more adventurous, you can disable the L2 cache in the BIOS setup and try to run some programs. That will give you a first-hand look at the performance gains due to the L2 cache.

On older systems the L2 cache was located on the motherboard.

On Intel’s Pentium II and Pentium III and on AMD’s Athlon processors, the L2 cache is located on separate chips near the processor on the same circuit board.

On Intel’s Pentium Pro, second-generation Celeron and ‘Coppermine’ Pentium III and on AMD’s K6-3, ‘Thunderbird’ Athlon and ‘Duron’ CPU’s, the L2 cache is on the processor die. The advantage of on-die cache is that it can run at the full processor speed; for example on an 850MHz chip, the L2 cache runs at 850MHz. At 64-bits wide, that’s...a lot...you do the math.

Off-die cache runs at half the processor speed, or less.

More to come later...

Okay...time for more…if anybody’s still awake.

The Frontside Bus:

The frontside bus (FSB) connects the CPU and memory. It also connects the CPU to the AGP, PCI, and ISA buses (while those peripheral buses are usually treated as separate buses, they are really part of the frontside bus).

When you run a program, the data is copied from the hard drive to the memory. The CPU runs it from the memory and uses the chipset to communicate with the peripheral buses.

The frontside bus, on a 66MHz system runs considerably slower than the backside bus, at 528MB /sec. It’s still a bottleneck on Celeron systems, which are hobbled at 66MHz.

The bandwidth of the frontside bus on the latest motherboards is equal to or greater than the backside bus...800MHz or better, depending on the chipset. Intel’s 440BX and 440GX chipsets were the first to offer a 100MHz FSB, newer chipsets like VIA’s Apollo KT-133A offer a 133MHz FSB (or 266MHz with DDR SDRAM).

This might be a good time to discuss the current state of memory or RAM. The latest chipsets use SDRAM (Synchronous Dynamic Random Access Memory), RDRAM (Rambus Dynamic Random Access Memory), or the newcomer in system RAM, DDR SDRAM (Double Data Rate SDRAM).

The chart below shows the Width, External and Effective Clock Speeds and Bandwidth for each type of RAM.

SDRAM is the memory most of us are familiar with. It comes in 66MHz, 100MHz and 133MHz flavors, and is named after the bus speed...PC66, PC100, and PC133. It synchronizes with the CPU and is fast enough to run the 133MHz bus. It is 64-bits wide.

RDRAM uses a proprietary memory protocol (Rambus) to pass data to the CPU. It is ‘double-pumped’, meaning that it sends data on both the rising and falling edges of a clock pulse, effectively doubling the speed. RDRAM is (sort of) named after the effective clock speed…PC800 runs on a 400MHz clock (800MHz effective speed), PC700 runs on a 356MHz clock (712MHz effective speed) and PC600 runs on a 266MHz clock (532MHz effective speed). It is only 16-bits wide however...I’ll get back to that.

DDR SDRAM is also ‘double pumped’. It, like SDRAM, is an ‘open’ design. The naming convention can be a little confusing; it was originally named after the bus speed, like SDRAM, bringing us PC200 and PC266 DDR SDRAM. After RDRAM used the effective clock speed, which sounds faster, they switched to using the bandwidth...PC1600 and PC2100.

While those who have jumped on the Rambus bandwagon would have you believe that it is the fastest memory available (and be careful...Rambus might sue you if you disagree ), tests have shown that it is in fact slower that the current PC133 SDRAM. If these tests are correct, it will be considerably slower than DDR SDRAM. The Rambus side says that the tests are flawed...I’ll leave the decision up to you.

I will caution you about Benchmarketing, however. Certain testing labs use certain benchmarking software to compare systems. Designers and manufacturers know this, and sometimes design a product to do well on those benchmarks, while cutting corners elsewhere. Then the marketers push those numbers at you to make you think their product is something it’s not.

Remember the Apple commercials where they claimed that their chip runs faster than a comparable one from Intel? Well, it does...when running native Apple code against the same program ported to Wintel. Tests were done reversing the situation, where the Wintel box running native code beat out the Apple running the ported stuff. These are the things companies do to try and get your money...but I don’t need to tell you that.

So how could PC800 RDRAM, with its 1.6 GB/sec bandwidth, possibly be slower than PC133 SDRAM, with its mere 1.064 GB/sec? Rambus is a memory protocol. Think of it as a tiny Ethernet card sending 16-bit wide packets of data that must be reassembled into 64-bit wide chunks in order to be used by the processor. This adds latency…just like the ‘lag’ experienced by online gamers. Data must also pass serially through each RIMM (stick), so adding more memory actually increases the latency.

The battle rages on...  ...more information about this can be found at Tom’s Hardware and Real World Technologies

Now on to the peripheral buses…we’ll start with the oldest...

The ISA Bus:

These guys were the slots you found in the original IBM PC, and at a measly 16-bit width and 8.3 MHz speed are starting to look a little tired. If your motherboard still has an ISA (Industry Standard Architecture) bus, the mouse, keyboard, serial and parallel ports are connected to it. The PC99 initiative calls for the end of the ISA bus, and most new motherboards don’t have them. They do, of course, offer legacy support for those devices listed above.

The PCI Bus:

The PCI (Peripheral Component Interconnect) bus was originally developed for video cards. Video has always needed more bandwidth than the ISA bus could supply. ISA video cards would often cause problems for the system...they didn’t always ‘play well with others’.

The PCI bus is 32-bits wide and runs at 33MHz, making the bandwidth 132 MB/sec. Along with any PCI cards in the system, the IDE Controllers, USB and FireWire run on the PCI bus.

The AGP Bus:

As video cards and CPU technology advanced, the bandwidth of the PCI bus became insufficient, and Intel developed the AGP (Advanced Graphics Port) bus. The standard AGP bus is 32-bits wide and runs at 66MHz, raising the bandwidth to 264 MB/sec. AGP video cards directly access system memory, so it can store and manipulate high-resolution textures.

There are currently three flavors of AGP...Standard (or 1X), 2X and 4X.

AGP 2X transfers two 32-bit pieces of data for each 66MHz clock cycle, doubling the transfer rate to 133MHz and the bandwidth to 528 MB/sec. AGP 4X doubles it again to 1.06 GB/sec. Keep in mind that your motherboard and system memory must support these faster designs before you will see any improvement, PC100 SDRAM has a maximum bandwidth of 800 MB/sec.

Some AGP cards also use what is called sidebanding, which allows the graphics processor to queue new data requests while the main bus is gathering data.

I have not mentioned the MCA (MicroChannel Architecture) and VLB (VESA Local Bus), two failed bus systems that are (hopefully) long since dead. If your motherboard has either of these two buses, you might want to consider an upgrade...now.

That’s enough for tonight...but I do have a ‘pretty picture’ that helps illustrate the various buses:

Ready for more?

When we refer to ‘chipsets’, we usually mean the pair of chips located on the motherboard that connect and control the various buses. These chips have been called ‘Northbridge’ and ‘Southbridge’ chips (since the Intel 440BX, at least).

The Northbridge:

The Northbridge chip connects and controls the CPU, memory subsystem, the AGP bus and the PCI bus.

The Southbridge:

The Southbridge connects and controls the IDE bus, the USB (Universal Serial Bus), Plug ‘n’ Play support, the PCI-ISA Bridge, the keyboard and mouse controllers, Power Management features and other goodies.

Integrated Video:

In their zeal to lower costs (in order to separate you from your money ), some low-cost computer makers have resorted to using motherboards with integrated video chipsets that share resources with system RAM. A system called UMA (Unified Memory Architecture) allows this to work.

Integrated video has been a bottleneck because the AGP bus is squeezed down to 66MHz and because, of course, using system RAM is significantly slower than using on-board RAM.

With the 810 chipset, Intel offered a new low-cost chipset with a new Accelerated Hub Architecture, as shown in the picture below.

Accelerated Hub Architecture:

The 810’s architecture changes the Northbridge chip to GMCH (Graphics and Memory Controller Hub) that connects the CPU, the memory subsystem, the graphics controller and AGP bus, and Power Management.

The Southbridge is now called the ICH (I/O Controller Hub) and connects the IDE bus, USB and integrated LAN controller.

These two hubs are connected by a dedicated interlink bus that is 64-bits wide and, in 2X mode, runs at 133MHz...resulting in a bandwidth of 1064 MB/sec – twice that of the PCI bus.

Intel’s current line of 800-series chipsets (810, 815, 820, 840 and 850) all use this Accelerated Hub Architecture. Although the diagrams above shows RDRAM, some use SDRAM. They also allow use of the new CNR specs, allowing computer makers to saddle you with junk...um, I mean…sell you a low-cost system.

Click on picture for clear view of the call outs...

AMR, ACR and CNR:

AMR (Audio Modem Riser), ACR (Advanced Communication Riser) and CNR (Communication and Networking Riser) are all methods of selling you software-based sound cards, modems and Ethernet cards. Instead of just having WinModems and their lousy performance, you’ll be able to have WinSound and WinLAN too. Oh joy.

To quote from an article at Tom’s Hardware:

quote:

---

AMR and CNR are targeted to OEMs, who always try to find the least expensive way of adding features to their systems. People who build their own systems should not bother with this stuff and motherboards that are targeted to the retail market can do just fine without CNR or AMR slots.

---

Watch out folks...they’re trying to sell this to you.

For those interested in a comparison of the various riser specs, Intel has a whitepaper here in .pdf format.