The more you look at the PowerPC 603e, running at 100MHz, the more it looks like the natural successor to the 601 at the low end of the desktop market. It is likely to be cheaper, it is very nearly as fast and it has a lot more headroom. In order to get the 601 up to 100MHz, IBM Corp needed to move to a whole new process. While it used 0.5 micron CMOS, the company introduced a new transistor geometry that enabled it to cram the components onto a smaller die size. So dramatic was this shrinkage that the 100MHz 601 became physically smaller than the 603, despite its much higher transistor count. When asked why the new 603e does not use the same process, IBM and Motorola Inc reply, quite simply, that they could achieve 100MHz using the older technolo gy so they are keeping the new stuff in reserve for the higher clock speeds. As expected, the gross structure and capabilities of the 603e are essentially the same as the older 603, but there are a few nips and tucks worth noting. Several logic paths have been re-written to achieve the 100MHz clock speed and it has not just the cache’s size that has been increased, so has its performance; the 603e cache is four-way associative, rather than two-way. The new processor’s load-store unit has also been tweaked, so that any stores that hit the cache only take one cycle where previously they took two, and mis-aligned accesses are also handled better – improvements that have incidentally been carried across to the new PowerPC 602. From Apple Computer Inc’s point of view the extra cache is the most obvious boon, since its current 68000 emulator technology needs a lot of cache elbow room. However, the PowerPC 603e also adds the ability to run the chip at fractional speeds comparative to the machine’s bus and this will give Apple the flexibility to ship a range of PowerPC PowerBooks with differential clock speeds, as well as easing its task of getting upgrade boards out for the existing Powerbooks and Duos. Our sister publication PowerPC News has questioned why the SPECint figures of the 603e were marginally better than the 100MHz 601, while the floating point figures were marginally worse. IBM Microelectronics PowerPC product line marketing manager David Ryan said the 601’s unified 32Kb cache suits the cache-hungry floating point tests better than the 603e’s 16Kb instruction cache. The integer tests, on the other hand, reflect the 603e’s cleaner design. The general shape of the low-end PowerPC market will now depend almost entirely on how the companies decide to price the new part, and the proportion of the initial production run that gets swallowed by Apple. Integer performance is still the key to fast machine performance with most applications, and the 603e beats the 601 there. Similarly, the 603 is significantly cheaper than the 601, and although the 603e is bigger than the older model, it should still be cheaper to fabricate than the 601. This is probably the beginning of the end for the 601. – Chris Rose
