AMD Announces New Open Compute Platform Support; Possible New Customers
Nearly two years ago, Facebook launched the Open Compute Project. The initiative was intended to rethink all aspects of data server design, including cooling, racks, and server configuration. The coalition has gathered steam and support from a number of companies; AMD announced its own Open 3.0 Platform at today's OCP Summit. AMD's hardware is designed to meet the OCP's Open Rack standard and to be much more configurable than typical servers.
To understand the significance of that, you need to understand that servers are typically fairly specialized. If you buy your servers from Dell, you buy Dell's management tools and interface capabilities. Buy from HP, you buy into HP's ecosystem. Windows and Linux won't care if you have two systems from two separate vendors, but you can't take Dell's software and run it on an HP box. One of the hallmarks of the Open Compute Project is to create common tool sets that can run on platforms from multiple vendors. The long-term goal is to create management frameworks that can handle servers from Intel, ARM, and AMD.
At first glance, the base architecture looks a bit hum-drum. Two G34 sockets supporting up to 32 CPU cores, 32 DIMMs, and three secondary function blocks. What makes the Open 3.0 platform shine is that it's designed to scale from a 1U HPC configuration with low memory latency up to 3U configuration for storage servers. Normally, that sort of upgradeability is fairly restricted -- you can buy a server that's designed for a particular role, but most aren't designed to take off-the-shelf add-in cards for other tasks. AMD's goal is to provide a flexible server that can be repurposed or reconfigured post purchase to optimize TCO over the entire life of the product.
These new server parts are compatible with Open Rack and chassis that comply with the Open Compute 1.0 standard, but isn't compatible with earlier chassis built on the draft version of the OCP. That shouldn't be a problem, given that server OEMs would tend to deploy their own custom solutions in any case, and the OCP is still in relative infancy.
According to AMD, motherboards and systems will be available from Tyan, Quanta, and Penguin Computing before the end of Q1. The new equipment is designed to leverage AMD's 6300 Opteron processors but will retain compatibility with older Magny-Cours CPUs as well. "We have eagerly awaited the AMD Open 3.0 platform as it brings the benefits and spirit of the Open Compute Project to a much wider set of customers," says Charles Wuischpard, CEO, Penguin Computing. “As we deliver a new line of Penguin servers based on AMD Open 3.0 and AMD Opteron 6300 processors, our high performance computing, cloud, and enterprise customers can now deploy application specific systems using the same core building blocks that are cost, performance, and energy optimized and perhaps most important, consistent. We think this initiative eliminates unnecessary complexity and provides new levels of supportability and reliability to the modern data center."
To understand the significance of that, you need to understand that servers are typically fairly specialized. If you buy your servers from Dell, you buy Dell's management tools and interface capabilities. Buy from HP, you buy into HP's ecosystem. Windows and Linux won't care if you have two systems from two separate vendors, but you can't take Dell's software and run it on an HP box. One of the hallmarks of the Open Compute Project is to create common tool sets that can run on platforms from multiple vendors. The long-term goal is to create management frameworks that can handle servers from Intel, ARM, and AMD.
At first glance, the base architecture looks a bit hum-drum. Two G34 sockets supporting up to 32 CPU cores, 32 DIMMs, and three secondary function blocks. What makes the Open 3.0 platform shine is that it's designed to scale from a 1U HPC configuration with low memory latency up to 3U configuration for storage servers. Normally, that sort of upgradeability is fairly restricted -- you can buy a server that's designed for a particular role, but most aren't designed to take off-the-shelf add-in cards for other tasks. AMD's goal is to provide a flexible server that can be repurposed or reconfigured post purchase to optimize TCO over the entire life of the product.
These new server parts are compatible with Open Rack and chassis that comply with the Open Compute 1.0 standard, but isn't compatible with earlier chassis built on the draft version of the OCP. That shouldn't be a problem, given that server OEMs would tend to deploy their own custom solutions in any case, and the OCP is still in relative infancy.
According to AMD, motherboards and systems will be available from Tyan, Quanta, and Penguin Computing before the end of Q1. The new equipment is designed to leverage AMD's 6300 Opteron processors but will retain compatibility with older Magny-Cours CPUs as well. "We have eagerly awaited the AMD Open 3.0 platform as it brings the benefits and spirit of the Open Compute Project to a much wider set of customers," says Charles Wuischpard, CEO, Penguin Computing. “As we deliver a new line of Penguin servers based on AMD Open 3.0 and AMD Opteron 6300 processors, our high performance computing, cloud, and enterprise customers can now deploy application specific systems using the same core building blocks that are cost, performance, and energy optimized and perhaps most important, consistent. We think this initiative eliminates unnecessary complexity and provides new levels of supportability and reliability to the modern data center."
