: News
: Reviews
: Editorials

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




AMD Q3-2005


And Why some Analysts May Need to do Their Homework


by Josh Walrath



AMD posted a record revenue for the past quarter, and beat the street by more than a fair margin.  The consensus was that AMD would post $0.08/share earnings on $1.38 billion in sales, but AMD came out with a surprising $0.18/share on $1.52 billion in sales.  This far outstrips last quarter's performance, and is a huge improvement from this time last year.  AMD continues to execute on their server, desktop, and mobile processors and are gaining marketshare in each area.  AMD apparently gave out around $66 million in bonuses this quarter, as compared to the usual $16 to $17 million.  This has some investors and analysts spooked, but I for one believe a company should reward their employees for a job well done.  Working for a stingy or miserly company is not always a good experience, as often a good portion of an employee's time is spent trying to convince the bean counters that new equipment/tools/help is needed.  This takes away from productivity.  Also, working for a company that does not reward achievement is something that often causes a lot of employee turnover.  Rewarding employees with bonuses for a quarter that exceeded anything the street had predicted, and being able to execute on their internal roadmap is something more employers should do.  Of course, we don't know where all the bonuses went to, as it would be unfortunate if the majority of that money went to top level management.  Still, AMD was able to post a $76 million profit this quarter.  Yes, it would have been nice for the street to see earnings in the $125 million range with standard bonuses, but if the company and its employees continue to execute, then perhaps it is money well spent?

This Friday Fab 36 officially opens its doors and starts fabricating 65 nm parts.  65 nm is a complex process, and it probably takes around 10 to 12 weeks to complete a wafer.  Fab 36 can do around 5000+ wafer starts a week, and it uses 300 mm wafers.  On the current 90 nm process the Athlon 64 512 KB L2 die is about 84 mm square, the 1 MB L2 model is 112.9 mm square, the X2 with 512 KB of L2 is around 156 mm square, and the X2 with 1 MB is around 199 mm square.  Going to 65 nm with few changes to the product should result in die sizes of around 59 mm square, 79 mm square, 109 mm square, and 140 mm square respectively.  On 65 nm the 512 K L2 Athlon 64's will be tiny!  Not only that, but they will most likely be very energy efficient.  Heat production will probably be about the same as the 90 nm version, mainly due to the smaller contact surface such a chip would have with the heat spreader.  We would also expect to see a jump in overall processor speeds once 65 nm really hits its stride.

To give a quick and dirty idea of what kind of production we could see in very short order, let's take a look at what each wafer could do.  On a 300 mm wafer being used for single core 512 KB L2 Athlon 64's, around 1096 gross dies can be placed.  Now, some of those dies will be on the edge of the wafer and not be complete, and some dies will be used as test sites and alignment marks.  With a mature yield though, we could expect around 750 to 800 good dies to come off a wafer (the smaller the die size, the less chance of a defect being found in it- on average there are around 50 defects on a wafer due to dust/contaminants).  So, with 5000 wafer starts a week, and around 750 good dies per wafer, we get around 3.75 million dies per week coming off the line.  Now, AMD will not only be producing 512 KB L2 Athlon's, so their product mix will lower those overall figures to around 2 million good dies a week from both of their Fabs running processors.  Still, AMD expects to be able to provide more than 100 million processors a year.  This is enough to be able to achieve around a 30% marketshare if they sell every good die.

Now, this is assuming that they are using essentially a shrunk version of the current cores, and that is bad thinking.  The initial products off of the 65 nm line will have a great resemblance to the current cores, but it will still be somewhat different.  We can look back at what AMD did with their transition from 130 nm to 90 nm for a good idea.  The first desktop Athlon 64 core was the Clawhammer, then AMD focused on the Newcastle design for desktop (smaller die size due to eliminating 1/2 of the L2 cache, which also had a favorable effect on power consumption and heat production).  When AMD made the jump to 90 nm they introduced the Winchester core, which not only was able to improve power and heat characteristics, but was also able to show a slightly higher IPC than a Newcastle core at the same speed.  Many users also found that these cores, though branded as 3000/3200/3500 parts, could regularly exceed 2.6 to 2.7 GHz.  This was a good jump up from what most Newcastle parts could do as well.  The next generation of products after Winchester were based on the San Diego and Venice cores, and these had a very positive gain in IPC over Winchester, and were able to clock even higher and show better power and thermal characteristics.

In applying this train of thought to 65 nm, we can assume that AMD will not try anything radical out of the gate, but rather follow a plan of action that has been shown to work.  The first 65 nm parts we see will be mainly based on the current Venice level core, but they will have some tweaks that will most likely improve their IPC to a small degree.  Once AMD gets all the kinks worked out of the 65 nm process by using a well known core technology, they will go for a more aggressive design 6 to 8 months down the road.  So, by late Summer we should see AMD's 2nd generation 65 nm part.  A friend at AMD once told me when I broached this subject, "Evolution, not revolution."  AMD is taking a steady, and relatively low risk approach to process and design changes.  We will not see a massive design change until K9 reaches production, but rather a steady process of improvements that keeps costs as well as power and heat under control.


Next: More 65 nm


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


Copyright 1999-2005 PenStar Systems, LLC.