Last week it was open source, this week Sun (SUNW) wants to change the planet. Unfortunately, doing so will require a street fight with Intel.
Nicolas Carr writes about the opportunity for low-power enterprise computing and cites an interesting paper from Google about the economics of powering a server farm. He also talks about a blog posting from Sun’s president, Jonathan Schwartz, who wants to “Change the Planet”. I’m OK with that provided he can “Make some Money” while doing it. Finally, an internal Sun blog posting with links to lots of technical details on Sun’s new CPU (Niagara) can be found here.
Nicholas outlines a number of reasons why power is an increasing priority in enterprise computing and summarizes the issues well. He also thinks that this has created a window of opportunity for Sun – but this is where I just don’t see an opening.
It’s very clear to me that Intel (INTC) and AMD (AMD) are making performance per watt a priority to stay competitive in existing markets. Intel clearly states this in roadmap reviews, and has already successfully executed this strategy with their new mobile processors. This is being driven by the market reality that laptops now outsell desktops.
In addition, mobile technology is being applied to emerging applications in the home and office that require less heat and no fan. Intel already has the VIIV home media PC initiative, Apple is going to announce (using Intel processors) one as soon as Steve Jobs gets his black turtleneck back from the cleaners, and Intel wants a piece of the emerging cable/telecom/Tivo set top box market. A box with no fans and locked in your wooden TV cabinet will need to be low power. Just do a web search on XBOX 360 power and you’ll see that no one needs reminding of this problem. Sure, they may not need to have 32 threads and many cores, but the single low power design can be cut and pasted multiple times in the more powerful versions.
It’s a logical next step to adapt the same approach to enterprise servers.
At the end of the day most of the magnitude improvements in performance per watt may come from improvements in static, not dynamic power consumption. Static power consumption scales linearly with gatecount and is a function of process; dynamic power consumption is a function of clock speed and architecture. Sun has initiatives in areas such as asynchronous design to reduce dynamic power consumption but is dependent on Texas Instruments (TXN), their fab partner for improvements in process. Overall though, it just doesn’t seem like Sun can make a radical impact on power consumption when compared with Intel or AMD. They are at least trying to make a marketing impact.
Natural selection between AMD and Intel in the consumer landscape is clearly going to force the evolution of low power consumer CPUs, and you can bet that Google and other enterprises (as we said in a previous posting) are going to reuse consumer technology wherever they can. I just don’t think Sun can build a CPU specifically for the enterprise market and successfully compete with high volume consumer technology that is near parity. I would love to see Sun restored to its former glory but by differentiating themselves in a different hardware space. This will require innovation outside of the CPU with proprietary upgrades of the North/South bridge approach Intel has taken, better (and proprietary) memory/storage performance, or some other unmet need of the enterprise world.