Robert Ober, a systems and processor architect for semiconductor maker LSI, hopped into theCUBE during the recently concluded Open Compute Project (OCP) Summit to share his insider’s perspective on open-source hardware with hosts John Furrier and Dave Vellante.

Ober starts off the interview with a brief rundown of LSI’s two newly announced contributions to the initiative. The first is a board design for a 12Gb/s SAS Open Vault storage enclosure that provides greatly improved performance using existing drives, and the other is a PCIe flash accelerator card specifically designed to address the unique thermal, cooling and airflow requirements of Open Compute servers. He explains that standard SSDs “tend to fry” in OCP environments, which are built from the ground up for high-performance at massive scale and pose a different set of challenges than traditional architectures.

Vellante mentions that Amazon has the advantage of being able to tailor individual components for specific workloads and data center environmentals, an approach that enabled it to secure a dominant position in the public cloud market. Asked how hyperspecalization affects the broader market, Ober says that this paradigm  will increasingly help shape the industry going forward.

“The evolutionary direction we’re going in the data center, you can call it many things – you can call it pooling, you can call it disaggregation – but at a large scale, at a rack or multiple racks or a hyperscale data center, you wanna start pulling apart the parts,” he remarks.

Optimizing infrastructure down to the component level has many benefits, both architectural and operational, but Ober considers the improvements in thermal management to be the most notable. The reason, he details, is that processors, DRAM, flash and mechanical disk all have different temperature thresholds that have to be sub-optimally balanced in traditional configurations.

“Today, when you try to put all those things in one box and manage them in one temperature profile, everything is a compromise, whereas if you pull them apart, you can manage each thing correctly, and in the end you can get a much denser packaging,” Ober says. He sees disaggregation becoming the standard in data centers, with hardware virtualization extending beyond compute to networking and storage in order to allow resources to be “pooled and made accessible from multiple servers.”

Despite the vast potential of Open Compute, Amazon has opted to stay on the sidelines of the initiative. Ober explains that at present the project has far too few configurations to address the requirements of the web giant, which has already come up with its own solutions to the problems OCP is tackling today.

Ober predicts that the dramatic rise in unstructured data will not only serve as a catalyst for hyperscale adoption, but also accelerate the industry-wide transition from traditional block and file storage to key-value and object-centric models. He’s also looking forward to upcoming memory types such as PCM and spin-transfer torque, which he believes will push the envelope on database design and enable users to extract new value from their information.

To hear the full interview, check out the video below.