We have 17 guests online

Login Form



Remember me
Password Reminder

AMD Phenom II "Deneb" Print E-mail
Written by Michael Schuette   
Jan 05, 2009 at 04:00 PM

In the shadow of Intel throwing one megahit after the other into the CPU arena, AMD’s Phenom, based on the Agena core has led somewhat of a wallflower existence. Despite actually having some very advanced features, including clock gating for each individual core and a highly advanced memory controller there has always been a certain performance lag behind the Core2 designs that could not be overcome on an instruction per clock basis nor on the raw core frequency front. Given these issues, even the fact that AMD was the first company to release a native quadcore CPU into the market was not enough to boost the success of the Agena design. Paired with some dubious marketing decisions, as for example the release of the X3 triple core series to find a home for some defective dies and, more importantly, Intel’s process technology advancements that stole the show along with the power efficiency crown, AMD has been in a situation where the company had to show some sign of life in order to stay afloat.

Welcome to 45 nm Immersion Lithography Process Technology!

Cash for Fabs

In the interim, in a reversal of former AMD CEO Jerry Sanders’ doctrine that "only real men have fabs", AMD recently spun off its foundry business into Foundry Co. in exchange for a major financial infusion from an Abu Dhabi-based investment group, which at least for the short term warrants some financial liquidity. At the same time a modicum of control over Foundry Co. is maintained by AMD with those rights – in the broadest sense - also keeping the cross licensing with Intel regarding the x86 patent pool alive.

Server vs. Desktop - Shanghai vs Deneb Market Requirements

With the financial infusion, the immediate onslaught has been weathered, however, there is still the above outlined scenario that AMD has to show progress, particularly in the domain of process technology in order to maintain their credibility in the field of energy efficiency with the added benefit of possible frequency headroom gained through better transistor optimization. For the readers of our forums, the discussions regarding dry lithography vs. immersion at the 45 nm process node and the potential consequences for transistor optimization are nothing new but in a nutshell, given some data published about the Barcelona replacement Shanghai, AMD seems to be doing something right.

The desktop market abides by different rules than the server market. In the server market power and thermal values epitomize the total cost of ownership including the entire infrastructure of a setup. End users, even if they are energy aware, rather emphasize looks, performance and initial cost of any setup. Needless to say, that the latter market is also more prone to manipulation through buzzwords. The ground rule here, however, is that any desktop system, regardless of whether it is for the primary purpose of office, media encoding or gaming applications, constitutes a tradeoff of bottlenecks, and it is up to the end users to educate themselves which one of the many is the worst and which are the ones that can be tolerated – within the budget constraints.

The High-End CPU: A Slave to the Rest of the System

One abused example is the term "gaming CPU" which is a conundrum in itself, at least as long as any of the better offerings of either Intel or AMD are concerned. Be assured that, regardless of which processor drives the system, at any resolution that is acceptable for a hard core gamer, the graphics card is the limiting factor for system performance. A similar scenario holds true for media encoding, most processors that are out there and as long as the software is multithreaded, will exceed the typical bandwidth limitations of any single HDD-based system in which the source and target locations are on the same physical drive. The latter scenario is somewhat easier to remedy, additional hard disks are fairly inexpensive and easy to install and all current motherboards feature an abundance of SATA ports anyway sans the hassles of parallel ATA cabling. More convenient, of course would be external drives as offered primarily in the USB flavor variety but make no mistake, those drives, handicapped by the limited usable bandwidth of the USB interface will not solve the bottleneck problem.

There still exist quite a few applications that are primarily sensitive to raw CPU power and the interplay between the processor and the primary data storage areas a.k.a. system memory. 3D-rendering, modeling of dependencies (logical choices) and behavioral patterns are among the most CPU-intensive applications that, if the software is appropriately parallelized, can take advantage of multiple cores without having too much to worry about other system bottlenecks.

Otherwise, a lot simply depends on the platform architecture rather than on the actual processor itself. On the other hand, software optimizations and intelligent prefetch algorithms can alleviate even such bottlenecks as those imposed by Intel’s AGTL host bus. On the memory front, DDR3 is gaining market acceptance at an increasing pace, pushing DDR2 slowly into obsolescence. This is something where AMD is currently still lagging, at the present, there are no DDR3 motherboards available yet despite full support for the third generation of DDR memory by the controller.

Last Updated ( Feb 01, 2009 at 03:24 AM )
<Previous Article   Next Article>