Advanced Micro Devices
Where is AMD Going Now...

Format for Printing

Format for printing

Request Reprints


By eachus
October 22, 2004

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

Things at AMD are getting, well, interesting. This is the point at which I would normally expect AMD's CPU inventories to be at their highest. From now until early December, OEMs would be drawing down AMD's inventory, to put into systems for the Christmas rush. Instead we are hearing about spot shortages in various CPU models. If this was December that would indicate wonderful things about the fourth quarter. What do spot shortages now indicate instead?

First let me dispose of the 'bad news' case. If AMD had yield problems right now on either 130 nm or 90 nm CPUs, they would not have lowered prices this week. Creating additional demand would have been just silly. So what is going on?

First, there are the on-line vendors. They knew the price drop was coming, and we can assume that the higher volume vendors like Monarch and Newegg were aware of the price drop, and scheduled their orders several weeks ago so that their inventories would reach zero when the prices dropped. To the extent they raised prices and ran out early, it means that they underestimated demand. (The vendors guessed wrong, not AMD.) AMD may have had lots of orders scheduled to ship on or after October 19th, and they should have those parts in stock and shipping by Monday.

Also note that AMD has been playing their cards close to their chest with the new 90 nm parts. My guess at this point is that we won't see 90 nm parts outside of Socket 939 (or possibly Socket 940) before January. There are improvements in the 90 nm parts that showed up nicely in the TechReport review.

To explain the graph, it says that for maximum performance, your linear algebra library should divide large matrices into 64 kilobyte submatrices on an Athlon64 chip. The actual array size depends on whether you are using real or complex data, and single vs. double precision. (Libraries like ATLAS can create an optimal math library for you, which will be used by all your BLAS3 and some BLAS2 calls.) The plateau in the graph out to almost 512k tells you that for these results, the data is being cached in L2. The difference between the light blue (130 nm 3500+) and the blue line (90 nm 3500+) tells you that AMD apparently trimmed a clock cycle out of L2 latency.

So it is worth paying for the extra speed grade when buying the 90 nm Socket 939 parts. (AMD has even made it explicit by now offering the 130 nm 2.2 GHz parts with 3400+ model numbers, and trimming the max HT speed back to 800 MHz to justify it.) The real question though, and it is a huge question, is what will AMD have coming out of fab 30 in the next few months?

My guess is that AMD has enough Thoroughbred and Barton chips in stock to fill demand from now through the end of the year. Think about it, if AMD was expecting to build up inventories going into the fourth quarter, should they build inventories of completed SOI wafers (at 130 or 90 nm) or bulk silicon wafers? The answer is that the bulk silicon wafers cost the least to produce for inventory, and these products were mature, in learning curve terms, so that processing these wafers later would not be likely to improve yields or binning. This doesn't mean that AMD's bulk silicon CPU production is now zero, stopping and restarting a process is worth avoiding.

My guess is that AMD will have, say 10% of current wafer starts bulk Socket A Sempron parts. Say 50% 90 nm Winchester dice with speeds expected to bin mostly in the 1.8 to 2.2 GHz range. Perhaps 25% will be Sledgehammer cores for Opterons and high-end laptop parts, with the remaining 15% split among Paris, Dublin and Clawhammer. (Note that these are wafer starts, larger dice will result in fewer dice and fewer good dice.) Oh, and just for the record, throw in a couple 50 wafer lots of 90 nm Opterons.

If AMD has that sort of mix coming out of fab 30 right now, good for them. But I expect that they will still have some Newcastle wafers in process, and that it wasn't until sometime in September that Newcastle starts were eliminated. Also notice that AMD needs some 2.2 and 2.4 GHz dice for Socket 754 3400+ and 3700+ parts and Socket 939 3800+ and 4000+ products. (Some are Clawhammers, some are Newcastles.)

So, do I expect AMD to make money this quarter? Um, can you say lots? I don't think AMD will hit $1 per share for the year, but $0.50 for the quarter should not be out of reach. (My guess, though is around $0.40.)

The important thing to notice though, is that the switch to 90 nm wafer starts may be 70% or more completed by the end of the quarter. Those products will be shipping in late 1Q05 or early 2Q05, but they will affect the WIP inventory this quarter. However, this will show as a bulge in WIP valuation, but shouldn't affect earnings. (The work in process inventory is kept at cost. The cost basis will increase as AMD ramps SOI wafer starts. I'm just mentioning this here since it could show as an inventory bulge in the year-end statement.) My big expectation is an increase in earnings in 1Q05, from a very good 4Q04, while Intel will look to be floundering.

Become a Complete Fool
Join the best community on the web! Becoming a full member of the Fool Community is easy, takes just a minute, and is very inexpensive.