Datacenter Power Is Different

Although a lot of attention is paid to power reduction for handhelds and IoT apps, does it reflect what’s happening at the other end of things, in the datacenter?

popularity

With much focus on super low power devices for handhelds and IoT applications, I’m also interested in what’s happening at the seeming other end of the spectrum, in the datacenter.

Atrenta CTO Bernard Murphy rightly pointed out that when it comes to power reduction techniques, datacenter power is a different story. He reminded that at a unit-level, a lot has already been done or is underway to reduce blade power and power associated with storage, and that it may be difficult to improve much beyond this, architecturally, because demand is so unpredictable.

Significant improvement here will depend much more on process, he thinks, and this may be an area where continuing process node shrinks continue to be justified.

Companies like Intel are certainly trying to do their share to address exactly this point. This week in fact, Intel rolled out its Xeon processor E5-2600/1600 v3 product families to “address the requirements of diverse workloads and the rapidly evolving needs of data centers.” Hmmm, sounds familiar.

Even ARM now plays in this space.

While it’s not clear what the next big advanced will be that addresses power reduction in the datacenter, it’s one I will be watching with interest along with the datacenter managers. Will we continue to rely on process shrinks and new processors, or will another architectural/other advancement allow for a big jump in efficiency?

Readers: What is your experience here? We’d love to hear your thoughts on this.



  • My money’s on architectural changes. Stacking the memory on-chip (in 3D) eliminates a TON of power.