Another Richter Scale(tm)
column by Jake Richter
This column first appeared in Vol. 6-1 of the Panacea
Perspective,
circa May, 1994
The latest request we've seen on the electronic airwaves have been from users looking for 64-bit PCs to run those newfangled, fast 64-Bit graphics boards. Why? Because with the waves of "lotsa-bits is faster than not-so-many-bits" mania that periodically sweep across our industry, not many people are well informed enough to be able to distinguish between various bit types, or even understand what they all mean. And all this just when you thought all bits were just 0s and 1s!
By our count, there appear to be four distinct types of bits intermingled in articles and marketing literature:
Back to the 64-bit trauma, as you can tell, things get kind of muddy. When a graphics hardware company announces a new 64-bit part, it needs to be made clear what type of bits they are referring to. Intelligent observers will realize they aren't talking about graphics bits or CPU bits, but a distinction still needs to be made between Memory and Bus Bits. Let's take a look at the value behind each kind.
As best we can tell, all of today's "64-bit Graphics" products fall into this category.
Having a 64-bit memory bandwidth (unfortunately this is sometimes also referred to as a "64-bit memory bus") helps a device improve its memory-to-memory operations. For a Pentium, the on-chip data cache gets loaded twice as quickly than a similar external cache on 486 systems. For graphics devices, it means that memory intensive operations, such as copying or moving large blocks of data (also known as BitBLTs) will occur faster than in more traditional 16 or 32 bit wide memory architectures. This is great for on-board operations, especially because even ISA (16 bus data bits wide) boards will benefit, but by itself does little to improve operations which require memory to be moved across the system bus to or from the graphics device.
On the down side, the 64-bit support also means that the 64-bit devices will probably need to have at least 2MB of RAM allocated to them, with growth in 2MB chunks, based on prevailing memory chip architectures. This means slightly higher costs for 64-bit graphics boards.
On PCs, 64-bit wide system busses are only available in two flavors: PCI 2.0 and VL-Bus 2.0. If you don't have one (or both) of these busses in your PC, your chance of having real 64-bit bus support is nil.
However, having one or both of those busses in your system is not a guarantee of immediate performance boosts either. not all 64-bit graphics boards (i.e. with on-board 64-bit wide memory support) are 64-bit bus capable. And, even if the device is 64-bit bus capable, it doesn't necessarily buy you a lot of extra performance. Since 486 and Pentium CPUs are still basically 32-bit parts, they really have no concept of moving a chunk of 64-bits of data around at a time. The way that the 64-bit wide data bus transfers get used is via a mechanism called "burst mode", which, as the name might imply, requires a burst of data to be sent across the bus. The only time this can happen in a beneficial way is if your system software is moving large chunks of contiguous data from the PC to the graphics board or back, something that the average PC user's applications rarely do. If the data isn't contiguous, the 64-bit bus doesn't buy you anything. Additionally, if the system software (usually a display driver or low-level application) doesn't use certain types of CPU instructions, burst mode will never even kick in. From what we've seen, this lack of burst mode support in system software is quite common.
Whether 64-bit busses will make much of an impact on real users in the next couple of years is definitely debatable. However, 64-bit memory technology shows more promise - it is certainly inevitable on graphics boards, where it can provide a significant boost in raw pixel pumping power. And 128-bit memory devices are lurking just around the corner!
Richter Scale Home Page | Other Richter Scale PC Hardware Columns