Of Multicores and Cloud Computing

Share it on Twitter  
Share it on Facebook  
Share it on Linked in  
Slide Show

Make the Financial Case for Virtualization and Cloud Computing

Last week's introduction of a new 100-core processor from Tilera aimed at enterprise applications drew the usual comparisons to Intel's current and future product lines, but is it possible that what we're really seeing is a fundamental shift in the goals and aspirations of the chip industry in general?

First the details. Tilera's Gx 3000 series consists of 36-, 64- and 100-core models that the company has targeted at low-latency, high-throughput Web applications. The company was quick to point out that it provides a 10-fold boost in performance-per-watt than Intel's Sandy Bridge architecture, although this doesn't hold up across all applications. Still, for the energy-conscious, the 3000 is impressive with each core drawing barely half a watt.

At the same time, the device packs a lot of punch, featuring 12 MB of cache per die and delivering 66 Tbps bandwidth between cores. It also supports up to 512 GB of memory and can achieve 48 Gbps across dual PCIe 2.0 ports. Note, though, that by forgoing a traditional x64 instruction set, you'll need to run Tilera's own CentOS-compatible Linux stack.

Intel, of course, has no intention of standing idle while its bread and butter is under attack. The company is preparing a 50-core x86 device dubbed "Knights Corner" for release early next year. The chip will feature what it calls "trigate transistors" on a 22 nm process, which the company says should eventually push performance into exaflop territory.

What's interesting about all this, though, is not the raw numbers being bandied about, but the target applications. No longer are we seeing processors laying the groundwork for the way computing is employed; rather, we are seeing more specialty devices tailored to end-user needs. In Tilera's case, according to GigaOM's Stacey Higginbotham, the target is the cloud where throughput, rather than raw processing power, is king.

To be sure, the many core designs hitting the channel provide plenty of both, but as cloud computing coupled with the rise of mobile access devices continue to drive enterprise productivity, the silicon needs of PCs and servers are no longer paramount.

That may complicate the development cycles of chip-makers across the board, but at least it shifts the focus away from what industry leaders say you can have and more toward what users want.