Piledriver Microarchitecture and Vishera Processors
It is obvious that one wave of improvements wasn’t enough to eliminate all concerns related to Bulldozer. It was almost impossible to design and implement any radical changes over the year that has gone since the launch of the new microarchitecture. Besides, the production technologies available at the disposal of AMD’s manufacturing partners haven’t been improved either. Therefore there was every formal indication that Vishera processor design based on Piledriver microarchitecture should be regarded as the first and pretty superficial optimization. It won’t be until in about a year that we can finally see some radical enhancements and changes, when AMD introduced 22 nm process and finishes working on their Steamroller microarchitecture.
The formal characteristics of the semiconductor Vishera die make it very similar to Zambezi. Both of them are manufactured with 32 nm process, consist of 1.2 billion transistors, are about 315 mm2 big and have eight computing cores and an 8 MB L3 cache each. The basic principles of internal core organization also remained the same. They are paired in modules, with some functional units being shared between the two cores. These shared resources include not only the L2 cache, but also the floating-point unit as well as the instructions fetcher and instruction decoder. In other words, the maximum instruction execution rate in Piledriver processors remained the same: dual-core modules with this microarchitecture are unable to execute more than four instructions per clock cycle.
Nevertheless, AMD engineers still managed to speed up their Piledriver processors without modifying their basic structure and blowing the transistor count out of proportion. They accomplished it with minor enhancements in two directions.
First, they have seriously modified the internal core structure. If you take a closer look at the Vishera semiconductor die, you will easily notice that it is very different from Zambezi.
The restructuring enabled processors on Piledriver microarchitecture to work at higher frequencies even without any changes in the manufacturing process. Vishera are the first multi-core desktop processors, which nominal clock frequency ahs reached 4 GHz.
Second, engineers managed to increase the specific cores performance. Of course, we aren’t talking about any serious performance boost, but even minor adjustments turned out very efficient in the end. Front End now has a more precise branch predictor and a larger instruction window. The execution units acquired an enhanced scheduler and learned to process individual instructions faster, such as integer and floating-point division, for example. Moreover, the developers claim that they have increased the L1 TLB size and improved the data prefetch and arbitrating algorithms in the L2 cache.
As a result, the refreshed FX processor lineup on the Vishera design expands the Zambezi family “up” and offers similar models, but with higher clock speeds and better performance. And keeping in mind AMD’s evolutionary, rather than revolutionary modifications and optimizations, they logically decided not to introduce any new platforms, sockets or chipsets. New processors will smoothly replace their predecessors and will offer AMD fans another option for upgrading their existing systems.
At the time of launch Vishera processor family will have four members: two eight-core, one six-core and one quad-core processor. The table below sums up their specifications:
Those of you, who are already familiar with the previous AMD products, will find the new FX specifications to be quite logical. The clock speeds are higher, the TDP remains at the familiar level. However, there are a few other innovations. For example, the new quad-core FX processor will now have an L3 cache of only half the size, and the clock frequencies of the quad- and six-core processors are set at a lower level than those of the eight-core models. In the previous processor line-up it was just the contrary: the maximum clock speeds belonged to the CPUs with fewer cores. It most likely indicates that from now on AMD will focus more aggressively on their top FX offering, while previously their processors with lower core count were the ones with the most optimal combination of consumer features. In other words, it seems that the manufacturer is very optimistic that their FX-8350 Vishera processor turns out much more competitive than the FX-8150 from Zambezi generation.
And there is another good reason for that, besides the new microarchitecture and increased clock speed. It is the price. While a year ago when AMD introduced their eight-core FX processors, they positioned them as a better alternative to Core i5, today they have seriously revised their pricing vision. The top eight-core Vishera, FX-8350 processor, will cost even less than Core i5-3570K. And it means that from now on AMD considers their high-performance processor family capable of competing successfully against Core i5 CPUs, but not the faster and more expensive kind. In fact, it seems to be just as important of an advantage of the new FX family, as the new microarchitecture or increased frequency potential. Taking into account the price tag, new eight-core products will undoubtedly be of interest to a lot of users out there. Really, the top FX processors are truly unique in some way: no other desktop processor family can boast models with so many cores.