AMD Open 3.0 Highlights

• A new type of server – differnt from generic machines and those self-built by public Cloud ‘mega datacenters’
• An open motherboard design incorporating 2 Opteron 6300 processors
• Initially manufactured by Tyan and Quanta Computer
• Broadcom and Mellanox will supply cards and components
• Servers supplied by Avnet and Penguin Computing
• Pre-production units placed with HPC customers and in the Finance sector (where Fidelity and Goldman Sachs are current references)
• Many proposed use cases and targeted solutions for Cloud Infrastructure, HPC and Storage
• Potential advantages are lower acquisition and running costs, as well as multivendor supply

AMD’s Roadrunner project comes of age today as it launches the Open 3.0 motherboard. It is designed for end-users with large single workloads and offers the advantages of a single design, to be purchased from multiple Systems Integrators. x86-based servers are already relatively standardised, but each vendor has its own management software, making it impossible to manage a farm which combines products from multiple sources. Public Cloud ‘mega datacenter’ suppliers get round the problem by employing technical experts which build their own systems to support their relatively simple workloads at massive scale, enabling them to leave out components such as fans and power supplies which are coupled to each server in commercial offerings.

AMD is similar to IBM in seeing servers from a workload perspective and believes that there is sufficient requirement from large users, ISVs and service providers to help provide a radically different design.

The Features Of The Open 3.0 Motherboard

It has joined forces with component suppliers such as Broadcom and Mellanox to place its Opteron 6300 processors in the Open 3.0 motherboard (see below)

The board features 2 Opteron 6300 series processors, 24 DIMM slots, 2 x8 PCIe and 2 x16 PCIe slots, an x8 10GbE mini-Mezz slot, 3 dual DCM 5720 GbE and 1 BCM 5725 OOB management slots and 2 USB slots.

The boards are available from multiple contract manufacturers (Tyan and Quanta initially) and will be supplied by multiple SIs (Avnet and Penguin Computing). The boards themselves will be built into a variety rack-mounted servers form factors, including 1U, 1.5U, 2U and 3U height.

AMD is working with Intel through the Open Compute initiative and believes that this is the first of other boards, including those with other AMD chips and perhaps products from Intel and ARM in future.

Targeting Server Workloads

AMD has placed pre-production units with a number of customers including Fidelity and Goldman Sachs which was a public reference at the launch. It expects production systems to be available by the end of Q1 2013.

AMD sees 3 targeted solutions for the initial products. In particular:

• Cloud Infrastructure – 2U servers incorporating 12 DIMMs per CPU, 25 2.5 SATA/SAS disk drives
• HPC – 1U servers incorporating 4 DIMMs per CPU, 10 2.5 SATA/SAS drives via the add-in card and 6 via the Southbridge
• Storage – 3U form factor using 12 DIMM per CPU, 35 SATA/SAS drives, up to 4 full height short PCIe card

Each of these example configurations comes with 1GbE on board and the ability to increase bandwidth to 10Gb via the inbuilt mezzanine card.

AMD lists 27 possible infrastructure use cases, having worked initially with the Finance and Scientific computing sectors. It says its definition of ‘workload’ comes exclusively from its customers.

Some Conclusions – End-Users, ISVs and MSPs With Large Single Workloads Should Take Advantage

This is a significant breakthrough in server design, running counter to the way systems have been packaged up to now: most recently with large suppliers pushing blades as part of their converged infrastructure and integrated systems strategies. These of course are appropriate for customers who need generic servers and are happy to manage each supplier’s kit independently, while the public Cloud ‘mega datacenters’ build their own servers. A significant proportion of server shipments (14.3 million in the last year – see Figure 1) fall somewhere in the middle: new form factors such as Open 3.0 will appeal to end-users, ISVs and MSPs with large single workloads. They should be able to take advantage of multivendor supply, lower acquisition prices, easier manageability and lower running costs from the ability to share components such as power supplies and fans. They will need to have the confidence to buy their servers from relatively unknown suppliers (or SIs in AMD’s parlance) and a deep enough knowledge of the chosen workload to take advantage. Its early days, but we predict that AMD and other suppliers joining these developments should do well.