Now, let's talk more specifically about those Palomino-Morgan enhancements:

• Power/heat reduction: The new core features design improvements that AMD claims reduce power consumption and heat by twenty percent from the old design, given equivalent clock speeds. Given how hot Athlons and Durons run, this is a welcome improvement.

• TLB entries: The number of Translation Lookaside Buffer entries has been increased. This buffer stores translated memory addresses so the CPU doesn't have to do a translation it has already done. Increasing the number of entries should decrease the number of buffer "misses" and thus help performance.

• Hardware prefetch: Palomino and Morgan include a prefetch algorithm in hardware that attempts to predict what data the processor will require after the data being processed. If the data is not present in the cache, the prefetch function will go to main memory and retrieve the data before the processor actually requests it. Ideally, by the time the processor did actually request the data, it would already be loaded into the cache where the processor could retrieve it much more quickly.

• 3DNow! Professional: With the inclusion of 3DNow! Professional, the Duron now supports the full SSE instruction set, putting it on equal footing with the Celeron in this area.

The result of all these changes is a processor which runs significantly cooler, consumes less power, and offers better clock-for-clock performance than its predecessor.

You want celery with that?
Since the Celeron is Intel's answer to the Duron, we'll briefly discuss how its features compare to AMD's offering. While the Duron has 128KB of L1 cache, the Celeron has only 32KB of L1, split evenly as 16KB data and 16KB instruction cache. The Celeron does have an edge in L2 cache size, however, boasting 128KB to the Duron's 64KB. Of course, there are other factors to be considered as well.

The Celeron's cache is inclusive, meaning that the 16KB of data in the L1 data cache is duplicated in L2. As a result, the effective total data cache of the Celeron is 128KB instead of 144KB. The Duron, on the other hand, features an exclusive cache, so no data is duplicated between the L1 and L2. Thus the Duron's total data cache size is also 128KB.

As we mentioned before, the Celeron has recently made the migration from a 66MHz to a 100MHz front-side bus. However, the Duron still has the edge here, because while its front-side bus is 100MHz, it tranfers data on both the rising and falling edges of the clock. This means that the Duron effectively has a 200MHz bus. Also, while the memory interface of the Celeron is 100MHz, commonly available Duron platforms allow the memory to run at 133MHz, resulting in greater memory bandwidth.

Now that we've covered the technical aspects of these processors, let's see how that technology translates into performance. Though AMD bills the Duron as a value processor, technologically it is very close to the Thunderbird Athlon. Indeed, with the core enhancements, the new Duron even has a few tricks that the Thunderbird does not. As a result, we expanded our tests to include not only the Duron and Celeron, but also a Thunderbird Athlon at the same clock speed (running on the same motherboard) as well as a 1.2GHz Athlon on a DDR platform. On the Intel side, we added a 1.4GHz Pentium 4 and a 1.2GHz Pentium III.

The motherboard we used for the 1GHz Duron and Athlon tests was the Asus A7VI-VM, a VIA KM133-based motherboard AMD supplied with the Duron. This board was the only one we used with the Duron; because the Morgan core is just now being released, none of our other Athlon motherboards would properly support it. (Our Gigabyte DDR board wouldn't even boot with the 1GHz Duron installed.) It will take a BIOS update to enable Morgan support on most Socket A boards.