Overclocking

One of the more spectacular achievements of Phenom II is its headroom for overclocking. Using liquid nitrogen, the world record is approaching 6.3 GHz but even with standard air cooling, AMD is finally adding more headroom again. Intel's Penryn-based Yorkfield and Wolfdale CPUs have built a reputation of reaching 3.6-3.8 GHz on air, depending on the size of the cooler even higher frequencies. We ran the Phenom II X4 through the noverclocking paces. With the stock cooler, at ambient RT and without increasing the core voltage, we reached 3.4 GHz on a 17x multiplier at which the system was running rock solid. Swapping out the stock cooler for an OCZ Vendetta2 gave us another 100 MHz by notching up the multiplier to 15x. At 3.6 GHz, we managed to get into Windows but under heavy load, some applications like Cinebench 10 froze or simply did not utilize all cores.

Cinebench 10 x CPU benchmark started running at 3.6 GHz but only three cores showed activity and at some point, the benchmark started to get confused and just hung. The single threaded rendering worked without problem.

We increased the core voltage to 1.45V, which partially solved the issue in that crashes occurred only randomly but at RT, using the sample at hand, 3.6 GHz were just slightly out of reach if we applied stability requirements. Using some cold weather, we lowered the room temperature to 12° F (approximately -10°C), under which conditions we had no problems running at 3.8 GHz at a reported core temperature of 4-6°C. At 3.9GHz as well as 4 GHz, the system POSTed and got into the Windows Splash screen but failed to complete booting up into the OS.

We ran a series of Crysis CPU benchmark to show the scaling of the results as a factor of the core frequency:

Final Thoughts

The biggest accomplishment of the Phenom II release is the fact that they successfully mastered the transition from 65 nm dry lithography to 45 nm immersion process technology. Despite the fact that Intel has done the same geometry shrink transition about one year earlier, the migration from dry to immersion lithography is an important milestone in that it evens the way to further shrinks, with the 32 nm process now being within reach. This raises the question how long it will take Intel to implement the same transition, namely going from dry to immersion. We have no doubt that Intel will execute as always but for the time being, AMD’s process appears to be slightly less "on the edge".

Especially since Shanghai-based Opterons already stole some of the surprise moment, nobody expected any miracles from Phenom II and along the same lines, the flood of leaked articles from all parts of the world did confirm that most of the performance gain stemmed from frequency increments rather than logic innovation. With most of the buzz already gone, therefore, is there anything to still be excited about? Well, it is a new processor and it is faster than its predecessors and it does have quite a bit of headroom, so ..yes, there is!

Clock for clock, AMD’s instructions per cycle lag some 6-8% behind Intel’s Core2 technology and the truth is that Core i7 even dwarfs any of its older brethren. Accordingly, if there are no gold medals to be won in terms of raw performance, AMD needs to emphasize the price/performance ratio on a platform basis (including motherboards) instead of just keeping CPU prices at a point where they become an invitation for Intel to enter the next price war. This appears to be the strategy AMD is centering on and it is a logical choice in that it fragments the competition.

There is no doubt that the Phenom II is a good processor. AMD has caught up in a lot of benchmarks to the Core2 performance but it appears as if there are still too many software applications that are heavily favoring Intel, Nero would be one example, and there seem to be enough games that fall into the same category. One of the more interesting benchmark results we got was what happened to the Core i7 965 in DIEP Chess when HyperThreading was disabled, namely a 20some % drop in performance. Since it really doesn’t matter which setting is enabled when it comes to show the “absolute winner” maybe it is time for AMD to take another look at that technology and potentially complement it with separate L1 caches for the logical CPUs. Just food for thought…

Discuss this article in our forums