(don't know what "video card overclocking" is? Click here for an explanation!)
Many users have emailed me with the question "Can 'x' chipset
can be overclocked?" I have always answered them with "No,
MCLK does not support 'x' chipset." In recent weeks, the
posters in the usenet newsgroup comp.sys.ibm.pc.hardware.video
have revisited the subject of video card overclocking : many chipsets
can be overclocked through the product's Win95 device driver!
Overclocking information is available for the following chipsets:
And of course, check out Powerstrip! Powerstrip Win95/NT desktop utility that offers enhanced refresh-rate support and even overclocking for certain chipsets! Thanks go to Mikko Mäkelä for bringing this little-known program to my attention.
I'm always looking for more information. If you know of a way to overclock a chipset which isn't already listed above, please email me!
"Overclocking" is a broad term which amounts to "speed margining." It is the act of squeezing the last ounce of performance from your PC, whether it's the CPU, the memory, or whatever. Video card overclocking, which is the reason you're reading this short explanation, involves the increase of your board's video-controller clock-rate. Video cards, like CPUs, operate at a distinct frequency. Whereas your PC's microprocessor relies on the motherboard to provide this clock-signal, contemporary video chipsets provide an on-board (integrated) programmable PLL circuit to generate both the CRT-clock (sometimes called the VCLK or video-clock) and the system-clock. For instance, increasing the graphics-clock (GRXCLK) on a 3dfx card is a good example of overclocking. That's all you need to know! (You can continue reading for some more background if you want.)
…
In all ASIC (application specific integrated circuit) architectures, a clock-signal (usually designated "system-clock") drives the chip's core-logic. In the case of a graphics accelerator, the 'core-logic' consists of the chipset's 2D-engine and 3D-engine datapath(s), and possibly the video memory subsystem. (Some graphics ICs, like the Matrox Mystique and Rendition Verite, operate the core-datapath and memory-subsystem at different speeds. In these chips, the PLL produces a "master clock", which feeds separate clock-dividers for each core-block.)
Video card overclocking depends on the programmability of the video-controller's PLL. Old EGA/VGA adapters did not feature PLLs, but numerous fixed-frequency clock oscillators chips (primarily to support the variety of different display resolutions, but one of those oscillators provided the board's system-clock.) Since the whole board was powered by a fixed-frequency oscillator, there was no way to change the chip's clock rate, short of desoldering the oscillator and installing a different one. Furthermore, it isn't just enough to have a PLL on the video card; it's location is of equal importance. Although modern graphics cards have only programmable PLLs, high-end graphics cards tend to incorporate discrete RAMDACs with PLL functionality while consumer graphics cards rely on a single-chip solution housing everything. Since the PLL is programmable in both cases, what's the difference? As I explained to many S3-964/968 owners, if the graphics-controller does not house the PLL, then I, the application developer, must come up with code for every S3-964 card on the market : Different S3-964 graphics boards use different RAMDACs, making general support of the S3-964 impractical. Returning to consumer graphics cards with single-IC solutions, having an on-board PLLs guarantees a single, common register-programming interface will work for ALL products based on that particular IC. Wow if you're still reading, you must be pretty bored!
The high-level of integration (RAMDAC, PLL, host interface, video-input, DDC, etc.) in today's video controllers has several other implications. First, a product built with an single-chip integrated solution will have a lower manufacturing cost than a similar product built with multiple ICs. Second, barring unforeseen stupidity on the part of the video board designer, ALL products built with the single-chip video controller will possess the same baseline functinoality and hardware interface. The semiconductor vendor provides board OEMs with a single reference driver-set implementing core device functions (video BIOS, OS device drivers, etc.) This lets the board designer focus on the specialized features of his product, i.e. product differentiation, as opposed to making the product just work.
< Go back a level > | < Return to Royce's HomePage >
© 1997 [email protected] (last updated 12/28/97)