The most refreshing part of the show was the vast number of enthusiastic engineers. Engineers are typically an observant lot. They can show you things that you would have missed. Like the men with exhibitor badges and stopwatches quietly timing the speeds of their competitor's machines. This is an interesting game of cat and mouse industrial espionage to follow, especially when the engineer you are talking to is controlling the machine being timed through a terminal window. To the short red headed man: the machine that took you three tries to time, well, you were being played with. That machine is an order of magnitude faster than you were allowed to see.
Even the suits were more informative this time around. Another manufacturer's machines make laminated motherboards. Using it, you can bring the price per motherboard down to about $2000 each.
For a 20 layer design.
Incorporating several thousand gold traces.
In a limited production run with no economies of volume.
The game is up.
At some point in time, dual channel RDRAM held the memory bandwidth crown. The arrival of DDR brought a bit of disappointment to AMD fans, because single channel DDR was not even close to matching RDRAM's bandwidth characteristics. The layman's answer at the time was that there aught to be a dual channel DDR solution. The layman's response was that dual channel DDR would require far to many traces, and therefore motherboard layers, to implement in a cost effective manner. Fast forward a year, and Nvidia introduced the Nforce 1, the first dual channel DDR chipset. A few months later, Asus introduced the A7N266-VM. It retailed for about $80. Layers are cheap.
There are over 50 "major" motherboard manufacturers. Asus alone has over 100 different Pentium 4 motherboard designs currently in production. Several years have passed with pundits predicting a manufacturer shakeout. The market is far beyond saturated.
Intel keeps their customers jumping. New processor? You need a new chipset to support that or the same chipset but a new socket. You need a new motherboard. Intel motherboards set motherboard prices. Intel motherboards carry the same price premiums as their processors do.
At one time, incorporating ATA raid on a motherboard carried a $50 price premium. Soon, every good motherboard had RAID, audio, LAN, Firewire, USB 2.0 S/PDIF in and out, dual channel, 4 phase power supplies, tomatoes, lettuce and pickles on a poppy seed bun. Raid now cost $3, based on the price difference between Intel's ICH5R, and ICH5 non-raid.
Cheap dual channel. No shake out. Socket/chipset churn. Vast feature lists.
How can this be? The profit margins are good. The profit margins are incredibly good. How many optimistic analysts have predicted the chip industry's turn around based on motherboard manufacturer's impressive financial results? The same manufacturing advancements that have driven down the prices of everything else have been at work in motherboard manufacturing. Yet motherboard prices have remained stagnant because Intel's motherboard prices have remained high.
What kind of power does this give Intel? Intel has its fingers in many pies. Intel may not control motherboard manufacturers using favorable chipset prices. Intel has licensed the Pentium 4 bus to every chipset manufacturer except Nvidia. This would seem to dilute their power over the motherboard manufacturers. Intel would not do that unless it had another way of leveraging the market, for instance, by lowering its motherboard prices.
Canterwood boards were expected to sell for $200+. Intel's Canterwood board is currently selling for $150. This may be because of lack of demand due to most Canterwood boards not supporting future Prescott processors. This may also be Intel's retaliation for motherboard manufacturers supporting Opteron, and perhaps for motherboard manufacturers enabling PAT in their Springdale motherboards. Intel gets to play hero to the consumer this time around, exercising its power to reduce motherboard prices across the. board. For the motherboard manufacturers, there are still good margins in Opteron boards, as long as the volume picks up a bit.
Perhaps the guys selling the equipment to make motherboards were severely misrepresenting their products.
I would be writing a different article, an article about how dozens of engineers lied to me at Semicon. Engineers are typically an observant lot. They can show you things that you would have missed. Like exactly how fast a pick and place machine can operate. An order of magnitude faster, in fact. µ
Sign up for INQbot – a weekly roundup of the best from the INQ