So, first things first -- Nvidia rolled out a refresh to their GTX 280, conveniently named GTX 285. The new GPU is architecturally identical to the old one, it's just 15% smaller, having been etched using lasers with a wavelength capable of etching at 55 nanometers, as opposed to 65nm. What, at the end of the day, does this process improvement mean to us, the end users? Two things (or three, depending how you count), all of which are interrelated.

First, and more importantly, potentially higher clock speeds, both from the manufacturer, and via our own overclocking efforts. Our hypothetical commuter only has to go 55/65ths as far as it used to, which means it ought to get to the destination with its informational payload that much sooner.

So, the commute is done ~18% faster, theoretically, which means we should expect to see stock clock speeds jump about 10% in practice (due to switching latencies and all that sort of messy analog reality). And, lo and behold, the stock clock on the GTX 285 is about 8% faster than the GTX 280 (648MHz vs 602); cards that come overclocked from the factory do, in fact, approach 18%. For example, EVGA's "SSC" variant, at 702MHz, comes in at a tidy 100MHz overclock, for 16.6%.

The second part of the benefit of the die shrink comes in terms of power usage. To wit: smaller circuits need less power, and, as a result, run cooler. Lower operating temperatures mean greater longevity and more overclockability, in addition to, you know, such prosaic things as using less electricity. The savings aren't huge -- idle power consumption drops by about five watts, full load by about twice that -- but it's still less heat you need to whisk away to keep everything crunching numbers happily.

Stepping (or, as AMD calls it, "revision number"), on the other hand, is a modification or update to the silicon microarchitecture itself. Specifically, it refers to changes made to the mask through which the etching laser shines used to manufacture the chip. These changes serve several purposes: to improve clock performance, to fix errata (ie. bugs), to improve chip yields (the number of chips cut from a single silicon wafer), or to resolve electrical "hot spots" and other engineering issues.

A chip's stepping is usually designated in one of two ways. The first version of the CPU that comes to market is A-0; small changes increment the number (A-1, A-2, etc) while bigger changes warrant a new letter series (A-3 to B-0). What constitutes a big-enough change for a letter series increment is probably the subject of blood sacrifice and other black rites (actually, engineers deep in the bowels of the chipmaker's fortress citadel probably flip a coin or play rock-paper-scissors).

If you're curious about the stepping of your own chip, third party utilities like CPU-Z incorporate both Intel and AMD's identification applications, and also provide a host of other info (speed, cache, bus -- everything the hardware geek needs for a healthy breakfast, really).


Rafe Brox spends his days wielding a phone in one hand and a screwdriver in the other. When not causing friends and enemies alike to /facepalm electronically, he can be found extolling the virtues of the weird peripherals in his life, from kettlebells to the Trackman Marble. Those of you wishing to inflict or solicit hardware and gaming geekery in person can catch him volunteering at Dragon*Con. If you also share an unhealthy passion for PC hardware or know a good place he can get help for this addiction, the target coordinates are rafe.brox AT weblogsinc DOT com.