Understanding AMD
By Nils Dahl
Date: April 25, 2002
AMD is behaving like a partner in its industry. As a full partner, AMD supports the other companies in several ways.
AMD makes the processors and, for the most part, leaves the other components to other firms. More importantly, AMD deliberately avoids introducing changes to its processors that would force major redesigns of chipsets, motherboards, and memory systems. All this is aimed at bringing affordable products to market at reasonable prices during a business slowdown. It always helps if all the partners can minimize their own design and production costs.
Such a strategy is unknown to Intel. It might help to think of Intel as Godzilla. AMD is one of the X-Men, a team player. Or something like that.
Look at the recent past for obvious examples of behavioral differences. Intel introduces rdram, new chipsets, new processors that use new sockets, and proceeds to market its own chips, motherboards, and boxes. Intel pours money into marketing. AMD works through design upgrades of the Athlon but finally settles on socket 7/EV6 as a fine long term solution for low end and mid-range systems. AMD then provides a schedule that promises continuation of the Athlon as a product for Years. There is certainly enough processing power in the Athlon XP to serve the needs of most people.
Sad to say, many people have been conditioned to look at specific features of things and to buy 'improved' products over and over. This is the whole thrust of the disposable product market. Things are not only designed to be quickly obsolete but are forced to be obsolete by discontinuing key components - processors and chipsets in the Intel example.
AMD is introducing new technologies - in its Hammer lines of processors. There is really no point to making major changes in the Athlon just to appear to be competing with Intel's P4. And there are major disadvantages in doing such things - such drastic changes would badly hurt all the firms that are now supporting AMD.
Partnering with UMC to continue making Athlon processors guarantees that AMD can turn its own fabs to making the new, more profitable high performance processor line. This will not hurt any of AMD's partners. In fact, having product flow from UMC to other firms in Taiwan will reduce costs in several ways. Everyone, including the end users, will benefit. And Intel will have to try harder to compete, particularly with the moving of the Athlon into the low end slot that Duron now occupies - positioned against the much weaker Celeron product.
AMD is building good will by behaving as a team member. In parts of the world where Intel lacks major marketing arms, AMD is advancing rapidly. Remember that nForce, a brilliant example of a partnership effort, was launched in volume in Germany first. Anyone see any test results of those systems that began selling late last year? Gee, it almost seems as if someone was inspiring a news blackout. That launch did prove the value of nForce and did rally other supporters. Eventually, all the players will grow weary of having to buy parts from Intel and then try to compete at every market level with Intel. Riding on Intel's coat-tails can be a precarious marketing policy.
Now there is nothing to stop a major firm from buying Athlon XP processors, designing a cartridge with advanced parallel buses and huge L3 caches, and then using that product in a new commercial design - as Compaq and Sun routinely do these days. Yes, it would be nice if AMD designed a 4-way smp chipset that worked with Athlons, but Hammer has smp as a part of its overall design - and the hammer designs will perform better due to hypertransport interconnections.
Frankly, Hypertransport includes so many ways of interconnecting processor nodes that it doesn't make sense to try copying Intel. Intel's simple smp design is awful anyway. I mean really crude and inefficient. Just as an aside, Hypertransport provides for 'tunneling' - basically sending the data through a channel that has multiple drop points. If you want a superior video system design for something like 3D animation, just send the drawing instructions through the tunnel to multiple independent video drawing subsystems and merge the results into a single composite image. This is the stuff of high end networked SGI realtime simulators - many of them so secret that even I have little information about them. Yeah, right. So Hammer plus Hypertransport is aimed at all the commercial niche markets and will include support for types of work that Intel's technologies are unsuitable for. That is the future. Wait for it and enjoy the results.
nils
p.s. A new concept in sandwich fillings. Finely chop roasted white meat chicken. Do the same with onions, hot peppers, and other vegetables. In a bowl, mix all dry ingredients with shredded cheddar cheese and then add mayonnaise. Mix until uniformly coated. This combination tends to stay together, thanks to the shreds of cheese forming a binding matrix. No more slices of stuff with bits that slide right out during each bite. the finer the cheese is shredded, the better it holds the mixture together.
===================================
Pssst! We've updated our Shopping Page.
===================================