Conclusion

AMD introduced its socket AM2 platform almost two years ago and it took the company more than six months to follow up with the first 65-nm processors. At that time, we found that cache latencies of the shrunken Athlon 64 X2 dual cores increased, but that power consumption decreased. The test results for this article confirm those results.

Both the 90-nm and the 65-nm devices have gotten face lifts, and they were developed almost in parallel. AMD still uses its 90-nm process for its high-end devices, including the Athlon 64 X2 6000+ and 6400+ Black Edition. These processors require the modified F3 stepping, while F2 processors are already available in clock speeds of up to 3.2 GHz. At the same time, the 65-nm processors were upgraded from G1 to the G2 stepping, which resulted in the overclockable 5000 Black Edition.

Although we haven’t looked at overclocking capabilities of the four processors in this review, we can draw some conclusions after testing all four versions:

The performance of all Athlon 64 X2 5000 versions is the same. The 65-nm models are marginally slower in some benchmarks due to increased L2 cache latency. However, such a small difference is negligible and ultimately unimportant. The 65-nm processors require less idle power than the 90-nm models, but they did not help to reduce the power consumption under load. AMD made the most significant step when switching from the 90-nm F2 stepping to F3. Knowing this, it is reasonable to say that AMD’s DSL SOI process is already excellent when looking at leakage power, meaning that a die shrink offers more business value (more silicon real estate) than potential design improvements. This could explain why AMD hasn’t offered high-end dual cores based on the 65-nm process and has instead used its 90-nm production for its flagship models since the 65-nm dies are not much better for high clock speeds. It makes more business sense to ship the smaller 65-nm devices in high volumes for the mainstream market than fine-tuning them for a smaller market segment. The progress between the two 65-nm processors G1 and G2 from an efficiency or performance standpoint seems to be smaller than the difference between the F2 and F3 devices. G2 generates better results in SiSoftware Sandra’s CPU arithmetic benchmark than the other processors, but this result isn’t reflected in the real-life benchmarks.

We will follow up on this topic with a detailed analysis of different Intel Core 2 steppings as well.