wbmw,
You might think that gaining 2-3% is worth millions of dollars for AMD in creating a new motherboard division. I don't think it is.
There is a lot more to it than 2-3% performance advantage. I am talking more about future than now. Right now, there is nothing AMD can do with Socket A motherboards to gain any additional performance advantage. There is still a window of opportunity for a low power Socket A motherboard, supporting PowerNow fully.
The area that AMD may need to address with Hammer, (if it looks like nobody is addressing these needs):
- A bait and switch system for Opteron, single CPU system supporting 2 DDR channels, maybe even including PCI-X on one HT link, AGP 8x on another. These will always be low volume, and there may not be any partner willing to sell these. The expense should really be booked into marketing budget, because that's the only thing that it would be used for, performing well on the benchmarks. And cost of this motherboard should not really be a consideration, it can be $300 or $500. It is rare to see any kind of reality check in the reviews as far as cost of the system is concerned.
- an OEM only single CPU Hammer motherboard supporting all power saving features of Hammer
- 4 way Hammer motherboards (optionally 2 and 8 way).
As far as expenses are concerned, lets assume that the total expense of creating motherboard division is 100%. How much do you think AMD already spends in the chipsed design, software driver development, reference motherboard design, motherboard maker support, as a percentage of this 100%? I think the number is very high, my WAG is 90% if AMD limits the sales to OEMs or 75 to 80% if these boards would be available in retail.
But these motherboards would open the door for AMD to markets where they are completely shut out from. This is really the main reason for the existence of motherboard division. SFF desktop is probably 15 to 25% of desktop CPU sales. 2-way systems for Opterons is something I am not that worried about (other than timing of 3rd parties). There should be sufficient support from existing motherboards vendors.
4 and 8-way systems will be very tough nut to crack. I am a little bit on the fense on this issue. AMD has just barely overcame the 2-way system hurdle, and normally, I would be against the expenditure of getting to > 2-way systems until 2-way systems have been fully digested. But in Hammer - Opteron - AMD has a great CPU to really outclass the competing systems due to superior system architecture, and to leave this on the table may be a huge opportunity lost. I am sure there are people at AMD doing the cost benefit analysis here, given the new, more challenging environment. But AMD already went a long way down this road, spending resources to enable Opteron to have built in (on-die) n-way capability, they have a reference design of a 4-way Opteron system at some stage of completion.
As far as going all the way and sell these > 2way motherboards, I think AMD will have to do it alone. I doubt motherboard makers will be willing to invest the necessary R&D $$$s, and OEMs seem to be having a wait and see attitude. By delivering a 4-way motherboard (let's forget 8-way for now), AMD can open the door to thousands of white box vendors who have been completely shut out of this market. AMD would be the only choice for these vendors, and each motherboard would sell up to 4 processors. If AMD does not sell the motherboards, I don't think they will ever materialize.
Even though there is a trend to clustering of less expensive replaceable servers / blades, database systems still like big iron - that 4-way Opteron would represent.
Joe |
