PAT in ASUS P4P800

The remarkable thing about ASUS P4P800 mainboard is the fact that ASUS engineers allegedly managed to enable PAT technology in it, which is a prerogative of a more expensive i875P chipset. However, ASUS had to give up these attractive statements under Intel’s inevitable pressure later on, there is just a short mention of some HyperPath technology on their web-site right now. Let’s take a look at the description of this mysterious HyperPath technology:

HyperPath provides a short cut for data transfer between the CPU and memory by bypassing redundant mechanisms included in conservative chipset designs. The ASUS innovation significantly shortens latency time and enhances performance without sacrificing stability. Before implementing the advanced technology, ASUS' engineering team conducted thorough and stringent tests to strike a perfect balance between performance and reliability.

Doesn’t it look familiar? It is almost the same thing Intel says about its PAT. However, the name of the technology is not so important. The most important thing is that ASUS P4P800 performs very close to i875P based mainboards. This is the major argument for the support of PAT technology in ASUS P4P800, even though the board is based on i865PE, which doesn’t officially support PAT, according to Intel.

Anyway, there is nothing strange about ASUS P4P800 being as fast as i875P based mainboards. Having taken a really in-depth look at PAT technology, we arrived at a very curious conclusion: there is no PAT at all. Intel’s marketing department very skillfully misled the manufacturers and end-users by telling stories about PAT being the distinguishing feature of i875P, which makes it faster and more expensive. Here is how Intel’s marketing department explains the idea behind PAT:

According to the picture, when the memory is addressed in synchronous mode in a system with DDR400 SDRAM and 200MHz FSB, shorter pipeline is involved. They claim that they use special techniques and select the dies for their i875P chips very carefully, that is why they managed to give up additional synchronization buffers inside the chipset and thus save two clocks on addressing the memory. Intel tells the users that in i865PE it is impossible: there is no optimized way for memory addressing.

In reality the truth is completely different. In order to show you the real PAT, we changed Intel’s marketing picture according to reality.

The thing is that in reality Intel’s “optimized way” is not optimized at all: it is the usual algorithm for work with memory, which is applied in i875P and i865PE when the bus frequency is 533MHz. It is a totally different story ho the chipsets behave when the bus frequency is equal to 800MHz. The faster i875P continues using the regular (optimized, according to Intel) memory algorithm, while in i865PE an alternative “bypass” algorithm (or non-optimized algorithm) is enabled. This alternative algorithm uses additional buffers and hence slows down the performance. We can’t figure out the exact goal behind it. Maybe all this is only about marketing and Intel slowed down i865PE on purpose by pushing it to use the longer “bypass” algorithm for work with the system memory. Maybe it was done to increase the chipset stability, and its functioning with 800MHz bus really does require additional buffering. In this case, Intel should be really selecting more stable and better-quality chips for i875P, while for i865PE they should be using longer and more reliable algorithm for work with the memory subsystem.