Open Compute unveils AMD Open 3.0, Intel photonics

January 17, 2013 // 11:22 a.m.

Tags: #amd-open-30 #amd-roadrunner #appliedmicro #arm #calxeda #data-centre #facebook #intel #open-compute #quanta #server #silicon-photonics #tyan

The Open Compute Project - founded by Facebook to bring new technologies to the server market in the name of improving efficiency - has made a raft of announcements this week, including Intel's first open-source silicon photonics designs and a new modular server system from AMD.

First, AMD's offering. Dubbed Open 3.0, previously known under the codename Roadrunner, the server platform uses a modular design to improve the efficiency, flexibility and cost of server systems. Bemoaning the current 'one size fits most' approach used by server vendors, Open 3.0 uses a custom AMD G34 motherboard measuring 16in x 16.5in and designed for installation in 1U, 1.5U, 2U or 3U 19in rack chassis along with a pair of Opteron 6300 processors to build a central platform which can be customised for almost any workload.

As an example, AMD explains that the system can be configured for use as a high-performance computing (HPC) cluster node with 1 DIMM per memory channel, support for six SATA drives straight off the southbridge, a single SR5670 tunnel and an on-board gigabit Ethernet connector with support for 10Gb Ethernet via an add-in mezzanine module board in a 1U chassis. For general-purpose use, the system can be shoved into a 2U chassis which adds support for up to 25 2.5in SATA or SAS drives, one full-height and one half-height PCI Express slots and the same 10Gb Ethernet connectivity via a mezzanine board. For storage use, the Open 3.0 platform can be placed into a 3U chassis to provide room for four full-height short-length PCI Express cards and a whopping 35 2.5in drives.

With room for 12 memory sockets on all versions - with just the number of sockets per channel varying between workload configurations - the Open 3.0 platform is certainly not lacking in expandability options, although the earlier-promised support for all AMD server processors has yet to materialise.

'We became involved with the Open Compute Project very early as we saw a pervasive demand for simplified, energy efficient servers, claimed Suresh Gopalakrishnan, corporate vice president and general manager of AMD's server division. 'Our goal is to reduce data centre power consumption and cost yet increase performance and flexibility – we believe that AMD Open 3.0 achieves this.'

Currently, the AMD Open 3.0 platform isn't ready for a retail release, but the company has partnered with Tyan and Quanta to build the production version with a view to releasing hardware into the market by the end of this quarter.

Intel, a fellow member of the Open Computer Project, has also added a fancy new offering to the group's technology stable: an open-source implementation of its silicon photonics interconnect technology. According to Intel's internal measurements, provided by chief technology officer Justin Rattner at the Open Computer Summit earlier this week, engineering samples of the provided technology allow for up to 100Gb/s data rates using a fraction of the power required by traditional copper-based electrical interconnects.

'We're excited about the flexibility that these technologies can bring to hardware and how silicon photonics will enable us to interconnect these resources with less concern about their physical placement,' claimed Frank Frankovsky, chairman of the Open Compute Foundation and vice president of hardware design at supply chain at Facebook, of the company's announcement. 'We're confident that developing these technologies in the open and contributing them back to the Open Compute Project will yield an unprecedented pace of innovation, ultimately enabling the entire industry to close the utilization gap that exists with today's systems designs.'

To Intel's chargrin, however, it is being joined in the Open Computer Project by ARM licensees including Calxeda and AppliedMicro, who are positioning the Cambridge-based chip giant's low-power designs as the perfect way to improve server efficiency - especially in the high-performance computing market, where systems are increasingly receiving the bulk of their compute performance from add-in highly-parallel accelerator boards like Nvidia's Tesla family and Intel's Xeon Phi.

The most interesting aspect of the Open Compute Project is, as its name suggests, its openness: everything that forms part of the Open Compute Project, from Intel's new silicon photonics interconnect technology to AMD's modular motherboard design, is publicly available through Facebook's GitHub account - although some of the latest information has yet to filter through to the repository, which hasn't been updated for the last four months. With the Open Compute Summit coming to a close, however, that is likely to change.

Initially, the technologies forming part of the Open Compute Project will be aimed at server markets - but what starts in the data centre will inevitably trickle down to the desktop. In a few years - or, depending on how successful Facebook's pet project proves, decades - the same highly-modular and customisable designs, high-speed interconnects and highly efficient CPU designs will likely be in desktops throughout the world.