Bundkort, AMD d. 09. februar. 2009, skrevet af The Boss 1 Kommentarer. Vist: 33334 gange.
The Phenom II is AMD's very first production of 45nm chips. Before, the 65nm Phenom-series, has been the best performing architecture from AMD - but it is far from the only improvement and update the new series of CPU's has had. That's why we've chosen to split this section into 5 main headlines, which are going to cover all the news that Phenom II brings. That way, we hope our readers will find it a bit easier to understand, exactly what makes the Phenom II faster and broader.
In AMD's press material, we found this overview, which, in a very good way, illustrates what has been done in order to enhance the performance.
Size
We'll begin with looking at what exactly it means to switch from 65nm to 45nm, when you're looking at the physical perspective.
CPU | Athlon X2 | Phenom | Phenom II |
Size | 230 mm2 | 285 mm2 | 258 mm2 |
Transistors | 227 Mill. | 450 Mill. | 758 Mill. |
As the above table shows, there has been made room for a total of 758 million transistors, against the mere 450 million transistors of the old Phenom-core, and at the same time, the physical size has in fact shrunk a bit. In comparison, we can see that an old dual core Athlon X2 manufactured using the 65nm technology, is hopelessly behind in amount of transistors, and therefor also in pure, raw compute-power. This leads us to the next point, namely the increased number of instructions.
More instructions pr. clock cycle
With more transistors it is possible to squeeze out more instructions pr. clock cycle, and AMD has improved and optimized this proces even further compared to the old, so as to achieve more computing power, without the necessity of adding more outright instruction-sets.
Higher clock frequency
Once again we can thank the switch from 65nm to 45nm process-technology for the additional performance - this time in the more obvious shape of Megahertz. At the same time, the necessary voltage-requirement pr. MHz has been lowered considerably. This, along with the improved microarchitecture, has made it possible for AMD to achieve higher clock-frequencies. Where the old Phenom-series offered up to 2.6GHz, the new Phenom II has reached up to 3.0GHz as a standard, but new models with up to 3.5GHz has been announced.
Increase in size of L3 cache.
Using 65nm production-technology, it wasn't possible to make room for more than 2MB L3 cache, as it would have simply made the power consumption go through the roof. The jump to 45nm has again put down its footprint, and has made up to 6 MB L3 cache possible. The L1 and L2 cache of 64 KB and 512 KB, respectively, has not been tampered with.
DDR3-support
A further step up the performance ladder comes from the DDR3-support. Officially there is support for up to 1333 MHz DDR3 in dual channel, but it is a number one shouldn't weigh too heavy, since there will be possibility for overclocing to "unofficial" speeds, or there will be support for higher speeds along the way, when the mainboard manufacturers dare to up the speeds by use of new BIOS-revisions. AMD has - very cleverly - chosen to name all CPU's with a built-in DDR3-memorycontroller AM3.
The chipset for Dragon, is still the same as that of the Spider-platform, that came along with the first generation of Phenom, namely the 790FX model. While the video cards have snuck themselves up to the HD4800-class. And naturally, the star of it all, the Phenom II-processor. Lets just take a look at what has changed since version I.
The original Phenom is, despite the lower performance, in fact some of the most modern we can reach when it comes to architecture. Technically speaking, the Phenom is lightyears ahead of Intels Core2-technique, which still relies on a chipset architecture from when Shakespear was a child. In fact, one could say that Intels Core i7 is really an attempt to catch up with the innovative ideas from the Phenom-architecture, as the integrated memory controller and the Hypertransport bus. Evildoers and nay-sayers might even say that Intel is copying from AMD, but... there are a few but quite crucial differences between the Phenom and the Core i7.
The most important difference is the way the internal cache is handled. The Core i7 uses a so-called inclusive cache, where all the data from the Level 1 and 2 cache can also be found in the Level 3 cache. This makes it possible for any of the four cores to find all the data at any given time at one place. This provides for a faster execution of data, as there's no real seek-time.
Phenom uses an exclusive cache, that doesn't share its data with the L3 cache, but keeps the data by the core executing the instructions. This means that in the event that a core needs help from another in order to run a program, the second core needs to sniff data from the L2 cache of the other core, before it is able to get to work. In return, all of the L3 cache is free, since no duplicated code is required to float about. Another disadvantage by this system is that you can't shut off a core that holds data that can be used by another core. It has to be active even if it isn't being used. One of the new things Phenom II has to offer, is that a core can do a dump of the data held in its Level 1 and Level 2 cache to the much larger Level 3 cache, if the core isn't being used. This should enhance the efficiency quite markedly. And, it is of course and advantage that the Phenom II has 6MB of Level 3 cache, whereas the Phenom only has 2MB. Of course there's still the integrated memory controller.
Thanks to Polarfar for the part about Dragon, and the more in-depth analysis of the Phenom II architecture.
Specifications and more on the next page