Cloud providers to stick with CPUs: Intel

Chip maker Intel has remained unconvinced of a market for GPU-powered cloud computes, despite Amazon’s launch of one such service last November.

While GPUs offered “some advantages” for some supercomputing workloads, the cost and difficulty of developing suitable software could be a barrier, Intel Australia’s enterprise technical manager Peter Kerney said.

Unlike CPUs, which operated sequentially, GPU clusters performed operations in parallel, and thus required a different approach to programming.

“The percentage of workloads that fit nicely on a GPU are in [the] single digits,” Kerney told journalists yesterday.

“We’re seeing adoption of [GPU-based computing] in research, but software engineering costs are huge.”

Kerney’s comments came at the Australian launch of the Intel Xeon Processor E7 family for “mission-critical computing”.

Intel claimed to have improved data integrity, uptime and energy efficiency with E7, as well as delivering a 40 percent performance boost over Intel’s 7500 series processors.

E7 processors also featured Intel’s Trusted Execution Technology (TXT), designed to enable cloud providers to avoid attacks on a virtual machine layer.

Sydney infrastructure-as-a-service provider SteamEngine said that it intended to upgrade its current HP-Intel stack when commercial E7 server platforms became available, following a recent proof-of-concept trial.

SteamEngine provided high-performance cloud computing services for applications such as high-energy physics research at the University of Melbourne, and rendering the Harry Potter and the Deathly Hallows movie.

Graphics card vendor NVIDIA expected its Tesla high performance computing brand to appeal to similar, data intensive applications, launching an on-demand offering in partnership with Amazon Web Services last year.

But according to SteamEngine chief operating officer Michael Chanter, the scalability and power of its “massive” CPU cluster gave the Australian provider “no compelling reason” to deploy on-demand, GPU-based supercomputing locally.

"We think there will be some workloads where that's relevant," he told iTnews, "but in the cloud space, it's an economics discussion.

“We will continue to look at it [GPU computing], because our customers are looking at it. But we don’t have a compelling reason to change from one to the other [currently].”