We'd like to welcome you fine citizens to the second edition of 'Sharky's Private Eye'. A couple of our more 'risqué' readers suggested we rename the column to 'Private Parts'. Alas, Howard Stern got there first, therefor the name stays. Just to recap, the purpose of this column is to sum up the latest/greatest happenings in the hardware/gaming world. We converse with a copious amount of engineers, game developers and other personalities in the industry and usually end up with a barrel full (sometimes two) of information. We do our best to distill this down into intelligible trends (without breaking any NDAs).

With all the current raucous over the upcoming exciting Athlon part, it's sometimes easy to forget what's happening 'right about now' (yes Mako is beavering away with an Athlon). AMD's K6-III part got off to a somewhat late start in the spring of 1999 and although not quite as fast under gaming conditions as a Pentium III or II (at the same frequency), the K6-III is still a very viable 3D gaming platform. The K6-III is a marked improvement (especially for business applications) over its K6-II predecessor with 256K of fully-integrated on-die cache, which now runs at full clock speed (as opposed to the K6-II's cache, which ran at 100MHz) and clearly has a significant install base. Although any graphics accelerators that are particularly CPU dependent will, in reality, only get an approximate boost of 10% by the increased L2 cache frequency of the K6-III in comparison to a comparable K6-II. With 3DNow! optimizations (AMD's own equivalent offering of streaming SIMD extensions) in place, AMD is able to combat their x86 architecture's major short coming- the poor performance of the Floating Point Unit (FPU or numeric coprocessor). By improving the K6-III's (and K6-II') floating point number crunching, the 3Dnow! instruction set 'relieves' the graphics pipeline somewhat by eliminating the bottleneck between the CPU and the graphics accelerator. In very simple terms, 3Dnow! and its 21 instructions can 'accelerate' floating point intensive calculations such as game physics and geometry transform, lighting and clipping. The only problem is that you're still half a penny short of two bob (a Brit way of saying short-changed) unless you have a decent enough performing 3D accelerator card to pump triangles and render pixels. A 3D accelerator takes any particular scene's geometric data that has been processed by the CPU and then transforms this data into the resultant grid of pixels that you ultimately end up seeing. Here lies the 'problem'. When the K6-II came out in 1998, 3D chip manufacturers provided rather poor support. Game developers weren't exactly jumping on the bandwagon either by supporting 3DNow! With the advent of the Athlon, it looks as though their attention has shifted somewhat as more and more ISVs and game developers alike are rushing to rectify the matter by releasing 3Dnow! Optimized drivers and 3Dnow! patches respectively. (The K6-III will probably be a very interesting choice of platform with the next generation NVIDIA NV10 and its hardware transform & lighting.)