The Overclocking Question
NVIDIA's FX 5950 offers only miniscule overclocking possibilities. The strict quality standards to which the cards have to conform even in tropical climates and under the most adverse conditions prevent higher default clock speeds. At least that is NVIDIA's official answer as to why its latest processor isn't clocked at a higher speed out of the box if it indeed it has so much potential.
In anticipation of the question why we didn't conduct any overclocking tests with our review samples, let us say the following: Overclocking tests with reference cards are problematic and often have no bearing on the end-product. These cards are prototypes, and the chips found on them are the very first working samples, fresh from the factory. This doesn't mean that these chips have more overclocking potential - quite the opposite. Often, cards from later production runs are much better overclocking candidates due to refined production processes. On the other hand, these prototypes are often equipped with much faster memory specifications than that found on the retail cards. Depending on the manufacturer, some retail cards use memory that is already pushing its limit at default speeds.
The overclocking protection built into current NVIDIA chips is worth mentioning in this context. Should a chip begin to return faulty results, the performance is automatically reduced. NVIDIA didn't go into more detail on how this works in practice, but this technique has been integrated in NVIDIA's chips for several generations now. According to NVIDIA, this is also why they don't require a software feature analogous to ATi's "VPU recovery" function. In the case of the FX 5950 Ultra, the upshot of all this is that the card is much faster than at stock speeds when run at 550/1050 MHz but performs worse than at stock speeds when clocked at 570/1070 MHz.