Overclocking

Another issue with slocket design is stability. Motherboards today are fragile creatures, not in sturdiness, but rather design. It's not just a matter of placing all of the components onto a PCB and making sure the wires aren't crossed. With today's incredibly speedy CPUs, timing issues have become of paramount importance. Place the CPU too far from the DIMM sockets, and you introduce potential problems with memory and bus timing.

Any company that makes a Socket-370 derivative of a Slot-1 design knows that it's not just a matter of fitting the socket where the slot was designed to be. The design of the board must be optimized to ensure that trace and lead distances remain within tolerances. We saw that with the Supermicro 370SBA; while it resembled its equivalent Slot-1 brother, the P6SBA, a number of motherboard components were either spaced differently, or placed on different section of the PCB.

This brings to question slocket design - does incorporating the CPU on a board introduce excess trace distance between the CPU and the motherboard? While it may not affect normal performance at 366 or 400Mhz, it is possible that slightly tighter timings, such as those required for overclocking operation at 450 or 500Mhz may be affected by the slocket.

Using today's standard components for high-reliability overclocking, including a Celeron 400 capable of running at 500Mhz, Abit's BH6 motherboard, and CAS-2 Corsair memory, we ran a few stress tests to determine any instabilities in our converters. Our Celeron 400 PPGA runs at a warm but stable 504Mhz (6.0 x 83.3Mhz). Knowing this, we figured that pushing the CPU to this speed in some intense conditions should be enough to stress the tolerances and timing of the additional slocket circuitry.

For each slocket, we covered the test system with a plastic bag, and ran it overclocked for 2 hours in a room at 24 degrees C, looping through Winstone 99 and the Quake II crusher demo. The CPU temperature on each converter card rose to 48.5 degrees C at 2.10v, but there were no freezes, errors, or other such anomalies. From our testing, it seems as if the socket-to-slot converters are the real deal - no performance hit, no added instabilities, and no detriment to overclocking.