If all had gone according to plan, AMD would have already launched its dual-core Phenom processors (codenamed Kuma), but Phenom hasn't gone according to plan since day one. Tidbits on Kuma are slowly making their way to the press, however, and while AMD won't specifically comment on unannounced parts or launches, sources at the company have quietly admitted to Ars that AMD will be launching new products between now and Shanghai's expected debut in December of this year.
New pictures—and SuperPi results—are now available, courtesy of Chinese site Expreview.com. According to them, the Athlon 6500+ will run at 2.3GHz and feature the same 2MB L3 cache as Phenom. Expreview provides CPUID information for both the Athlon 64 X2 5000+ Black Edition (underclocked to 2.3GHz) and the 6500+. Both processors carry 512K of L2 cache per core; the Kuma carries 2MB of additional L3 as previously mentioned.
Unfortunately, the author chose to run SuperPi—possibly one of the most limited benchmarks ever—but we do, at least, see a notable difference between the two chips. The older Athlon X2 core takes 39.37s to finish calculating one million digits of Pi, while the dual-core Phenom finishes the exercise in just 33.43s. That's a single-core speedup of 15 percent, which, as I recall, is roughly what reviewers found when Phenom launched almost a year ago.
We can, however, glean a bit more information from CPUID. Voltage on the 2.3GHz part (at least on this sample) is 1.24v, which falls within the 1.0v-1.25v range AMD has set for its X3 8650, which also runs at 2.3GHz. Power consumption on these parts will obviously be lower than what we've seen from Phenom X4 or X3 chips. When we reviewed Toliman earlier this year, we found that whacking off a core (and 100MHz) cut max load power consumption from 265W to 198W, and while some of that is obviously due to the clockspeed drop, there's plenty of evidence that a dual-core Kuma will draw significantly less power than a triple-core or quad-core part.
If we switch gears a bit (and are willing to examine some early, possibly dubious, Deneb benchmarks at 3.4GHz), we can compare Kuma and Deneb (45nm quad-core) in SuperPi. Remember, the core count difference between the two is meaningless; SuperPi is a single-threaded test. Also remember to take the following with a mountain of salt.
So, if a 2.3GHz Kuma finishes SuperPi in 33.43s, and a 3.4GHz Deneb finishes in 20.51s, how much of Deneb's faster performance is a function of its faster clock? Deneb is clocked 48 percent faster than Kuma and finishes the test in 61.4 percent of the time it took the older 65nm processor. Scaling, therefore, isn't quite linear, but it's close. IF all of these results are actually accurate, SuperPi doesn't pick up any additional performance from whatever special sauce AMD built into the Shanghai core. Then again, only about six people in the universe actually care about SuperPi, so this is no great loss.
As for Kuma, don't expect AMD to push clockspeeds on these parts much, even if the parts themselves are capable of it. Cores count for a lot in AMD's pricing structure, and the company won't want its dual-core parts stealing sales away from higher-margin triple-core or quad-core chips. As for actual overclocking potential, if any, we'll have to see what happens when the parts hit market. At the very least, Kuma should provide an upgrade step for lower-end AM2 users who don't really want to pay a premium for a quad-core part they know they won't make much use of.Posted on