AMD Phenom II X4 940: Compared to Phenom X4 9950 BE and Intel Core 2 Q9550

In last few years it looked like AMD lost its bearings in CPU market. No one expected that company, that is in constant transition, financial problems and without complete product portfolio, can produce interesting, equivalent and concurrent product. Phenom CPUs with K10 micro architecture didn’t claim speed crown from Intel, but they were significant step forward compared to “worn out” K8 cores.
We must be aware of the fact that every processor is result of compromise between investments and goals. Probably K10 would be better product if it had larger L2 and L3 cache memory, with better and larger Branch Target Buffers, with different cache policies that are favoring intensive operations with cache memory, but in that case Phenom, that was manufactured in 65nm process, wouldn’t give adequate results. We must remind you that Intel only recently, with 45nm manufacturing process that is fully operational, introduced monolith quad-core design. Unfortunately for Intel, and on the other side, fortunately for AMD, Nehalem is totally different price group (at this moment) and thanks to this fact AMD has maneuvering space in mainstream CPU market. Thanks to new 45nm manufacturing process, developed in collaboration with “big blue” (IBM), new cores with code name “Deneb” are much cooler than older 65nm cores, can achieve much higher working frequencies and can be overclocked much better and easier. Last year AMD introduced very solid Spider platform based on AMD 790FX chipset, ATI Radeon HD 3870 graphics cards and 65nm Phenom X4 CPUs. New platform, called “Dragon”, very similar to Spider platform, is based on 790GX chipset and Radeon HD 4870 graphics cards along with new 45nm Phenom II X4 CPUs. Deneb core is compatible with older 790FX chipset, so with update of BIOS new CPUs can be used on older motherboards. This allowed us to test new Phenom II CPU on DFI LanParty 790FX motherboard with 4GB OCZ DDR2 1066 memory. All tests were done on 790FX motherboard because there were very small differences in performances compared to test results on 790GX motherboard. When Phenom II was operating at default clock we used Thermalright True cooler without fan and it worked stable. When we overclocked it, two fans (mounted in push-pull configuration) were needed.Biggest innovation of new AMD CPUs is new and very advanced 45nm lithographic manufacturing process that introduces significant advances in pFET transistor performances compared to 65nm manufacturing process. New pFET transistors have higher drive current (specific current intensity) per micrometer: 660µA/µm compared to 510µA/µm for 65nm transistors. This allows higher performances: faster “switching speed” (speed of change from logical zero to one). Higher specific current intensity is one of the Intel’s “secrets” in achieving higher CPU frequencies because their HKMG (High-K Metal Gates) are one of the fastest dielectrics on the market when it comes to “switching speeds”. On the other side, on the same frequency AMD’s 45nm CPU has lower heat dissipation compared to Intel’s 45nm CPU. AMD claims that Shanghai (45nm Opteron) has for 15% lower power consumption under full load and 30% in idle compared to Intel 45nm Xeon CPU. This can make huge difference for servers but what it brings to usual desktop user?
One of goals that AMD was hoping to achieve was smaller dimensions of chip with increase in performances, because smaller dimensions of chip allow more L3 cache to be implemented in CPU. Thanks to 45nm manufacturing process, minimal size of gate on Shanghai and Deneb cores is 38nm: 7% smaller then on 65nm CPUs. Performances are for 19% higher for nFET and 23% higher for pFET compared to 65nm process. These results are achieved not solely thanks to better manufacturing process but also thanks to advanced optimizations. One of those optimizations is SOI (Silicon On Insulator – used also on some earlier models) that resulted in lower amount of electron “leakage” and “parasite” current. Another optimization of nFET is “stress memorization” that resulted in wider n channel. Thanks to this optimization scaling to 45nm allows better mobility of electrons in “narrow” cavities that resulted in higher specific current intensity per micrometer. This increased “drive current” for 19% on nFET and for 23% on pFET transistors compared to 65nm CPUs. This is great achievement because Ion/Ioff ratio on AMD pFET transistors is about ten times lower compared to Intel’s pFET transistors. Interesting fact is that electron “leakage” is 3 times lower compared to CPUs made in older 65nm manufacturing process. Since power consumption depends on amount of “leakage”, lower electron “leakage” results in lower power consumption of entire CPU.
Next important novelty is manufacturing process itself. AMD uses so called “Immersion Lithography”. This technique improves resolution of lithographic process: it uses liquid medium instead of air that is between lithographic lens and wafer which resulted in better refractive index. Achieved resolution is proportional to refractive index of fluid. Since fluid refracts beams on wafer, resulted resolution is higher and it can achieve sizes lower then 37nm. This resulted in better CPUs, higher frequencies and great overclock possibilities.