Dual Core Design Considerations

AMD was emphatic that one of the keys to their dual core implementation is the integration of the northbridge and the newly introduced crossbar controller and not the actual addition of the second core. With the second core, there will be resource contention for memory and I/O bandwidth and the combination of the system request queue and the crossbar switch alleviates to minimize this as an issue. We already see very low latency numbers already with the onboard memory controller and as this improves, the reliance on cache also decreases which means transistors can be further dedicated elsewhere. In the future, the addition of further cores in the future will build upon connectivity between the cores.

We have already briefly mentioned that dual cores means a slight reduction in temperature but there are additional benefits to having dual cores. Heat output over the surface of the heatspreader is not uniform - the hotspot is directly on top of the processor. With dual cores, there are actually two hot spots, one on each die meaning heat is spread out more evenly, which in turn means better cooling.

Heat is spread out onto two cores instead of one.

There are additional issues though and one of them will be heat management. AMD understands load may not be consistent across two processors; for example a single threaded application like Prime95 can run exclusively on one processor die while the other is relatively idle. This results in a huge temperature delta between the two cores and analog thermal diodes will not be sufficient in reporting correct temperatures. The solution to this is the addition of serial interfaces to the processor that are spread across the die. The serial interface will be read by external devices and through this the processor temperatures can be better managed.

The final topic is overclocking and once again, the problems with increasing the frequency. According to the speaker, overclocking on a dual core chip means that better cooling is not a choice but will be mandatory as current draw goes up significantly and the thermal threshold will exceed current levels. The speaker also stressed that all dual core solutions, AMD or Intel will face the same issues. My interpretation on this comment is that there may not be a huge amount of headroom available in frequency scaling and that we should expect the same general trend of 200 Mhz jumps every six months or so. There was also a quick comment on quad-core processors and they were expected to debut after AMD makes the switch to 65nm. Coupled along with the comments made about Moore's Law, there seems to be little doubt that AMD's strategy is to double the number of cores and reduce frequency every time they approach the alarming voltage / heat level. While process technologies will not be able to sustain clock speed jumps indefinitely, AMD has proved with SOI and other engineering refinements that they are can manage heat and voltage requirements to a certain degree.