Phenom II: New Microarchitecture or New Production Process?
AMD Company uses Phenom II name for their new desktop processors formerly known as Deneb. And although the “II” part following the well-established trademark name may be regarded as an implication of a new microarchitecture spin, the actual differences between Phenom and Phenom II are not that dramatic at all. Even AMD themselves do not deny that Phenom II use the same Stars (K10) microarchitecture as the previous generation Phenom processors.
Therefore, the main distinguishing feature of the new Phenom II CPUs is their transition to new 45nm production process and the use of immersion lithography. Unfortunately, AMD are an entire year behind their main competitor with the production process upgrade. Nevertheless, they use a number of unique techniques, because following in the footsteps of their primary technological partner – IBM – they started using immersion lithography equipment. The specifics of this equipment imply that it uses liquid instead of gas between the projection lenses of the lithography system and the semiconductor wafer. This enhances the system projecting precision and improves the characteristics of the ready semiconductor dies increasing the production yields at the same time. This is a truly progressive approach, because immersion lithography equipment is not just an interesting improvement of contemporary production technologies, but is also a possible alternative to EUV-lithography for future production technologies. This way, 45nm manufacturing process in AMD’s implementation may resolve a number of production and engineering issues all at the same time.
The use of new technologies for processor dies manufacturing affects the frequency potential in the first place. And AMD is in desperate need of higher frequency potential, because their 65nm Phenom processors could only go as high as 2.6GHz. It is low clock frequencies that do not let Phenom X4 processors to successfully compete against Intel Core 2 Quad CPUs. Luckily, the new production process proves up to all expectations. The new Phenom II X4 processors manufactured today use 45nm cores and can go as high as 3GHz clock frequency.
The second important advantage contemporary semiconductor technologies have to offer is the ability to fit more transistors onto the semiconductor die of the same size. And the more transistors there are, the better can certain processor features be improved. Namely, new Phenom II X4 CPUs use a 258sq.mm die and contain 758 million transistors, while previous generation Phenom X4 manufactured with 65nm process had only 450 million transistors on a 285sq.mm die. They managed to increase the number of transistors in the new Phenom II X4 die thanks to microarchitectural improvements aimed at increasing the performance independent of the clock speed.
Overall, AMD promises a 20% performance boost that the new generation CPUs should be able to deliver compared with the previous generation solutions.
Of course, higher clock frequencies of the new Phenom II X4 ensure most of the acceleration. The models announced today seem to be continuing the Phenom X4 lineup, their frequencies are 200MHz and 400MHz higher than those of Phenom X4 9950 – the top product in the 65nm family. However, a 15% frequency increase cannot possibly provide a 20% performance boost.
There are also other factors in the picture. Most added transistors serve to increase the L3 cache memory capacity. The L3 cache of the new Phenom II X4 processors tripled: 6MB vs. 2MB by the previous generation Phenom X4 CPUs. By the way, new processors L3 cache is not only larger; it is also faster than that of the predecessors. Its latency is 2 clocks lower, although that pushed the associativity 1.5 times higher. Phenom II X4 L3 cache has 48-way set associativity, while Phenom X4 has only 32-way set.
As a result, the performance of the L3 cache memory changed diversely, at least according to the Everest Ultimate 4.60 test.
Deneb (45nm) 3.0GHz; 1.8GHz on-die North Bridge
Agena (65nm) 3.0GHz; 1.8GHz on-die North Bridge
I have to add that unfortunately, transition to new production process didn’t let AMD increase the frequency of the on-die North Bridge and it still works at 1.8GHz in the new Phenom II X4 processors. It is really sad, because had this frequency been increased, the L3 cache and memory subsystem performance would have improved much more than after a slight reduction of the L3 cache latency.
These seem to be all the evident advantages of the new Phenom II X4. However, there have also been made a few minor enhancements that also affected the performance. The new 45nm CPUs have improved branch prediction algorithms: now AMD processors can predict indirect branching, just like their competitors. Moreover, the engineers managed to increase the capacity of internal data load and store buffers, as well as FPU buffers. Among other improvements we should definitely mention accelerated floating-point register-to-register instruction moving, LOCK pipelining, accelerated cache coherency during inter-core data transfer.
You can see the general effect from these minor improvements in the Sisoftware Sandra 2009 test results employing simple algorithms not affected by the cache-memory speed and capacity.
The demonstrated gain between 0.5 and 4% is the fruit of these particular “minor improvements”. In other words, the changes AMD engineers made to the computational cores can be described as insignificant. And it means that Phenom II X4 processors will yield seriously to quad-core Intel CPUs working at the same frequencies.
The illustration above shows that they also promise to boost the performance by using DDR3-1333 SDRAM. However, it is not the case for Phenom II X4 processors launching today. Today’s CPUs are designed for Socket AM2+ mainboards, which means they can only work with DDR2 memory, just like their predecessors. Nevertheless, AMD is going to expand the Phenom II lineup in February already by adding Socket AM3 CPUs to the existing models. These processors will be compatible with the existing Socket AM2+ mainboards, as well as with the upcoming platforms with the new CPU socket. Socket AM3 mainboards will be designed to support DDR3 SDRAM, which is also expected to additionally speed up systems with AMD processors.
Although improving performance was one of the most important goals for AMD engineers, the transition to new manufacturing technology helped resolve a few other issues that were typical of the previous generation Phenom CPUs. Here I am talking about heat dissipation and power consumption that have never been the strong sides of Phenom X4 processors built with 65nm process. And even though Phenom II X4 models launched today have the maximum thermal envelope of 125W, just like their predecessors, this number will be revised shortly. The Phenom II X4 models designed for Socket AM3 form-factor that are scheduled to come out in February will have 95W TDP, the same as Intel Core 2 Quad processors.
However, we don’t have to wait until February in order to see the progress made in lowering the power consumption levels. The Socket AM2+ Phenom II X4 CPUs launching today consume about 40% less power in idle mode. AMD has once again revised Cool’n’Quiet technology having provided the version 3.0 implemented in Deneb processors with a new cache flush on halt feature: core's L1 and L2 flush into shared L3 after a core enters a halt state allowing the core to drop to a lower speed and save power. This way you can cut the power to idling cores completely in case there is no workload to deal with.
It is fairly easy to check the efficiency of Cool’n’Quiet 3.0 technology in practical tests. We measured the power consumption in the processor voltage regulator circuitry and thus calculated how much power a CPU needs in case different number of processor cores is actually utilized. Note that the provided numbers do not take into account the efficiency coefficient of the voltage regulator, i.e. they reflect the power consumption of the CPU and VRM together. Despite that, these numbers can suit perfectly for our comparison.
New Phenom II X4 CPUs have really become more economical than their predecessors. You can see it not only in idle mode, when the power consumption dropped 60%, but also during different CPU utilization. Nevertheless, new Phenom II X4 processors are still less power-efficient than 45nm competitor solutions.
Summing up the brief discussion of the Phenom II X4 peculiarities, let’s compare them side by side with the specs of the previous generation quad-core processors from AMD:
Phenom II X4
2.2 – 2.6 GHz
2.8 – 3.0 GHz
4 x [64 KB + 64 KB]
4 x [64 KB + 64 KB]
4 x 512 KB
4 x 512 KB
HyperTransport bus frequency
Built-in North Bridge frequency
C1E, Cool'n'Quiet 2.0
C1E, Cool'n'Quiet 3.0