: News
: Reviews
: Editorials

: Archives
: About
: Contact
: Advertising
: Privacy
: Links
: Forums




Athlon X2 5000+ Black Box


Once you go…


by Josh Walrath


            The promise of AMD’s 65 nm SOI process was, well… promising.  Considering what we had seen done with both the previous 130 nm and 90 nm processes, the tech community was expecting great things.  Really great things.  Things didn’t work out.

            There were two issues with AMD’s 65 nm process when it was first released.  The first was that it was not anywhere up to speed as people were expecting.  The second problem was the primary product coming off of the line.  The 65 nm version of the highly successful X2 processor was not the foil against Intel’s Core 2 line that many were expecting.  It was slightly slower than the older 90 nm Rev. F series, and it was unable to clock much higher than the older chips.  Those marketing points go right out the window.

            All was not wrong with the 65 nm X2 though.  While the design was slightly slower than the older version, it was significantly more power efficient.  AMD rates the 65 nm X2 as a 65 watt product across the board, except in the case of the X2-BE products which are rated at 45 watts.  The older X2s have seemingly been sitting pretty at 89 watts for ages.  This little marketing point became the mantra for the company, as energy efficiency trumps all these days rather than performance at all costs.  Just ask Al Gore.

            Before diving into the Black Box, we should probably explore the reasons behind AMD’s “not so hot” introduction of their 65 nm products. 

65 nm + SOI = Really Kinda Tough and Personnel Issues

            SOI worked really well for AMD’s 130 nm and 90 nm processes, and it just seemed a natural fit considering the path that AMD was pursuing.  A lot of work was done by AMD and its fabrication partners (IBM, Chartered, etc.), and these larger processes worked very well, and were brought up in very short order.  65 nm was another beast altogether.

            The shrink to 65 nm involved a lot of new technology, and this did not fit hand in glove with SOI.  At the larger process nodes, the electrical characteristics of SOI were a boon to production.  With 65 nm that was not so much the case.  It did have advantages, but not nearly as large as they were in the bigger processes.  Most of this has to do with the mix of materials that goes into the 65 nm process, and getting the kinks out and having it work as advertised is still a challenge to this day.

            AMD follows a course it describes as “continuous improvements” to its fabrication lines.  This simply means that at the end of the lifespan of a certain process it is much improved in terms of power and transistor performance than when the process was brand new.  Intel does things a bit differently.  When Intel introduces a process, after a period of working out the bugs they do not change many of the attributes of that process.  Instead they get their engineers to work on the next big process jump and not spend the time making all these smaller jumps on the older process.  If Intel finds bugs in the process, they fix them.  But as to continually improving the performance of their process as well as implementing new technologies piece by piece, that is something Intel just does not do.  This is a gross simplification, so don’t knitpick too much.

            Process philosophy aside, AMD has shown that their constant improvements have allowed their older process nodes to adequately compete against the competition.  Well… this is true when we are not talking about the Core 2 Duo and its release last year.  Think Prescott, Northwood, Tualatin, and Willamette.  If you do not understand that last sentence, Google is your friend.

            The second part of this rant is AMD’s personnel issues.  It is not like there is anything wrong with their personnel, but until recently there really were not enough of them.  Before the past two years, AMD only had design teams in Austin and Santa Clara.  Each team worked on an architecture, and the multiple specific products under that architecture.  This meant that if one of them screwed up, it was going to be a bad year (or three) once the processors were released.  Well, after AMD finished its last spins of 90 nm Athlon X2s, they created a small team that would port over the design to 65 nm.  The rest of the engineers were shoehorned into Barcelona-land, as well as the distant lands of Fusion and Bulldozer.

            This small group had very few options to increase the performance of the X2.  In fact, one of the options that they were assigned was to make the L2 cache more flexible to allow for larger L2 sizes.  This was the reason for the slower per clock performance exhibited by the 65 nm X2.  To achieve their goal of allowing the addition of much larger caches on the X2, they had to add several cycles of latency to the L2 cache.  This little “feature” took around 2% to 5% performance away from the X2 design.  Because they were a small team, they had limited chances to cover up that latency by implementing other optimizations throughout the design.  In hindsight this was a poor plan, as AMD has never released a larger L2 cache X2 at 65 nm.  Way to go management.

            Bashing aside, AMD is finally hitting its stride with 65 nm production.  Right in time for the nightmare of complexity that is Phenom apparently.  We are finally seeing 65 nm X2 parts clocking as high as the top end 90 nm X2s, and doing it cooler and with less power draw.  3.5 GHz overclocks with current 65 nm X2s are no longer uncommon, nor do they require liquid nitrogen to achieve.  4.0 GHz overclocks have been made with dry ice.


Next:  The Black Box


If you have found this article interesting or a great help, please donate to this site.


Copyright 1999-2007 PenStar Systems, LLC.