AMD Phenom II X4 940 Review - Not the Second Coming
The Phenom II Processor
So what exactly differentiates the Phenom II from the original? There are a host of changes that AMD applied to the design, and in so doing increased the transistor count from the 463 million mark with the original Phenom, to a much more rotund 758 million. Because of the shift from the 65 nm processot 45 nm, AMD was able to actually reduce the die size from 283 mm square to 258 mm square. The primary recipient of all those extra transistors is the enlarged L3 cache. The original was 2 MB, and the Phenom II now sports a much larger 6 MB L3. Considering how well the Phenom architecture liked L3 cache when comparing per core performance between the dual core 7750 and the true quad cores, the addition should give a nice boost.
Some of the new features of the Phenom II.
AMD also put the fine toothed comb through each of the processor cores and tweaked and twiddled to their hearts’ content. While they did not radically change the design, they did improve IPC in ways in which they were unwilling to share with us mere mortals (reporters). One area which received a lot of attention was that of the memory controller. AMD improved the efficiency of their DDR-2 controller, and also added in DDR-3 support. While the current Phenom II X4 940 and 920 editions do not support DDR-3, the AM3 variants which are soon to follow will. AMD also improved how the memory and the triple cache structure communicate which other.
This affects not only performance, but how the processor can turn off individual cores by transferring all of the L1 and L2 cache data to be accessed by other active processors in the large L3 cache. This supposedly will help idle power to a much greater degree than the previous Phenoms, but we have yet to see how this will work in the real world due to limitations in how current operating systems handle this behavior.
Some more marketing points for the Phenom II.
Perhaps the most interesting aspect of this release is that of AMD’s new 45 nm process. While the architectural changes to the Phenom are welcome, and are certainly paying off in terms of performance, the real story here is AMD’s process. AMD worked hand in hand with IBM to develop their 45 nm process, and the maturity in which it appears to function at is quite impressive. While the Phenom II is hardly a radical redesign, it is not a simple die shrink. Have a complex, new design on a complex, new process has historically been an invitation for disaster. In this case it appears as though AMD has pulled it off. AMD has almost doubled the transistor count, reduced the die size, and kept thermal and power considerations in check all the while increasing the clockspeed of the product significantly higher than the Phenom II’s predecessor.
The 45 nm process is still SOI based, but it includes a new immersion lithography step which improves the accuracy of the lithography stage. This improved accuracy allows for tighter tolerances between structures without increasing leakage. There are also other additions to this new process which are not talked about, but are not nearly as important as the use of SOI and immersion litho. When we look back and compare this release to the other previous releases which included a process node transition, none of them were as impressive as this one. Going from 130 nm to 90 nm did not improve IPC in the new design, and the thermal and power consumption was not dramatically improved. The same goes for the jump from 90 to 65. Until fairly recently the fastest Athlon X2s were still based on the older 90 nm Windsor core.
The Phenom II will certainly be able to pull more tricks once it comes in AM3 form. While AM2 and AM2+ chips will not work in AM3 sockets, the AM3 processors should work fine in both AM2+ and AM3. It is likely, however, that AM3 chips will not work in older AM2 motherboards.
Last month AMD invited several publications for a sneak peak at the Phenom II and showed off some air, water, and liquid nitrogen cooling results. The results on these first production quality silicon parts in such experiments were impressive to say the least. AMD showed off 4 GHz air cooling results, near 5 GHz water cooling results, and an astounding 6 GHz+ results on LN2. A 3 GHz overclock is a far cry from the initial 400 MHz max overclock on the first Phenom processors at 65 nm. AMD has shown that not only does its design have the legs to reach higher clocks, but the initial 45 nm process could allow AMD to reach much faster speeds in the very near future.
The Phenom II certainly does seem like a tighter release than what we were served last November with the original Phenom 9600 (and unreleased 9900). The product performs nearly as advertised, and the overclocking potential is significant. While it may not convert the masses back to the AMD side like the original Athlon 64 did, it certainly may help stem the tide of defections from Team Green.