Pages: 1 2
A Consistent Theme in Sight
The theme of the week here at IDF is maturity. There are a lot of technical themes: CMP (Chip MultiProcessing), power consumption, mobility, wireless broadband, and we will try and discuss each of them (to varying degrees). However, the overall picture I came away with is that Intel is truly cognizant of their place in a maturing industry. Paul Otellini’s presentation compared semiconductors and computers to some very innovative predecessor industries: automobiles and steel, and several others. He made a somewhat compelling case that computers and semiconductors will continue to grow, particularly in the developing world, where the benefits of new technology are so much greater than in North America and Europe. One question which nobody has really addressed is how companies that are used to high margins will fare in a world where many of the incremental sales are in the developing world.
One significant change is that Intel really is listening very carefully to their customers. Everyone knows that GHz is no longer the predominant religion at Intel, but for a while there really was not a clear, consistent replacement; there was multi-core, but that is not really a message or direction for a company. This week, Otellini heavily emphasized that their new metric is performance/watt; and it is likely to drive some of their product decisions to varying degree for the different lines. He also made it explicitly clear that they now have fixed TDP guidelines for mobile, desktop and server, which is quite helpful. ULV notebooks will be aimed at 5W, desktops at 65W and servers at 80W (there is also a new product line targeted at 0.5W, but it is unclear if it is x86 or ARM). Part of the problems with heat dissipation had to do with it being a second order design constraint that could be fudged around if necessary. Quite clearly, this is no longer the case, and now the primary religion at Intel is performance/watt.
To drive this point home he showed the following slide, which indicates that Conroe (the desktop version of Yonah) will have 5x the performance/watt of Northwood (as measured by SPEC_int_rate). Woodcrest (the server variant of Yonah/Conroe) will increase server performance/watt by a factor of 3x, but this is measured by the TPC-C benchmark, which is much more difficult to scale than SPEC_int. TPC-C probably reflects reality a bit better than SPEC_int_rate which tends to scale near linearly. So, I’d say that the 3x estimated increase in performance/watt is what most users will likely see (for parallelizable stuff), while 5x really represents the best case scenario. Even so, a 3x improvement is good, however, it also begs the question “What about absolute performance increases?”
One of the nice things about managers (as opposed to engineers) is that they are really careful to know where the dollars are going. While a lot of people have said that “power dissipation is an issue”, I have yet to hear anyone put a price tag on it. Thankfully, someone from Google cleared this up; the power needed to run a computer for 4 years usually costs about half of what the hardware itself does (based on Google measurements). Otellini estimated that Intel could save consumers around $10 per processor by using more efficient products. While this isn’t enough to make an end-user excited, if you look at world x86 shipments that is still around $2 billion for 2005 (based on Gartner estimates of 200 million PCs sold this year).
For now, the most important thing to really take away is that while last IDF, there was a lot of confusion about where Intel was headed, it is very clear right now where Intel plans to go for the future. There are plans, there is direction and a year or so away, there will be some great products. The real question is how everything is executed, and Intel has always been best under pressure.
Be the first to discuss this article!