Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: Biomaven who wrote (123959)	6/16/2004 11:11:05 PM
From: pgerassi	Read Replies (1) \| Respond to of 275872

Dear Biomaven: It depends on the binsplits. If the peak is at 2.0GHz, virtually all dies can make 1.6GHz of the Opteron 242 or A64 2700+. Remember that AMD can and has in the past downbinned (selling a more capable CPU at a lower speed). If the peak is at 2.2GHz or over, the top end needs will be met with a small fraction of output and the only true determinant to overall sales is demand. The current situation is that upper end is demand constrained. The high end is due to price (by design) and the low end to production volumes assumed 3 months prior. The top end is the 850, 250, 246HE and the 240EE versions. They all get higher ASPs than the AFX53 and are allocated first. Then comes the rest of the Opterons, then mobiles and finally, desktops. Pete

To: Biomaven who wrote (123959)	6/17/2004 5:56:56 AM
From: Plissken	Respond to of 275872

Two variables factor in: the sweet spot which yields you the most usable CPUs that run at the specified clock and the spread. The sweet spot ideally should provide you with enough parts to meet demand of your bread-and-butter model. If it is too low, you end up with what is commonly called "yield problems" here, if it moves too high you can also get slapped with the "yield problem" term, although the luxury version where you have too many high ASP parts and lack on the low end. If your spread is wide, you'll get processors that may still work but are below the minimum clock for sale as well as high speed parts that go at clock speeds you have not yet announced. The latter ones you may want to put on a shelf as they will yield a premium, the former ones you may only get rid of in some "special edition" OEM deal. If your spread is narrow you may face a draught on both the low and high end while producing a glut of midrange product. In that case you have to bin down better CPUs and sell them for less than they could yield aka yield problem #3 and have too few high end parts.

To: Biomaven who wrote (123959)	6/17/2004 8:06:27 AM
From: fyodor_	Respond to of 275872

Biomaven: So which fits the current situation better - that scenario you just gave or the one at the other end of the spectrum, where they are demand-constrained at the lower end, and so can't produce enough high-end parts without over-producing low-end processors? I think the initial step must be to determine how many different cores AMD are actually producing. Only then can one look at pricing, availability, overclockability etc. and reasonably try to reach some conclusions as to what is going on. Let's look at A64 first: AthlonFX: 128bit memory (unbuffered DIMMs), 1MB L2 cache Opteron 1xx, 2xx, 8xx: 128bit memory (registered DIMMs), 1MB L2 cache Athlon64-M: 64bit memory (unbuffered DIMMs), 1MB L2 cache Athlon64: 64bit memory (unbuffered DIMMs), 512kB L2 cache My WAG would be that there are actually at least two, and quite possibly three, different cores here: core 1) Athlon64-M, AthlonFX core 2) Athlon64 "plain" core 3) Opteron where 1 and 3 could possibly be lumped together. I seriously doubt that AMD is wasting large amounts of valuable die space by disabling half the cache on the 1MB chips, so there would certainly need to be at least two different cores: One with 512kB L2 and one with 1MB L2 cache. The width of the memory controller does not really present much of an issue, IMHO. The big question in my opinion is whether Opterons represent a different core altogether. Since they can only use registered DIMMs, it would seem that their memory controller is different somehow. Additionally, they have more HyperTransport channels (well, the 2xx and 8xx models do). The 1xx model could possibly be covered by the "normal" 1MB L2 cache core, but considering the price premium on these parts, it wouldn't really matter much. In the end, though, I really don't have the necessary knowledge to do anything but WAG on whether there are two or three different cores. Perhaps someone more technically inclined could educate us as to whether the differences between the memory controllers and number of HyperTransport channels would be enough to warrant a third die? -fyo