A recent event by the Open Compute Project highlighted work being done to create standard designs for servers. AMD showcased its Roadrunner specification, which is targeted at financial services. Low-Latency.com got the background from John Fruehe, AMD's director of product marketing for enterprise products.
Q: AMD recently announced its open 'Roadrunner' specification for servers, specifically targeted at financial services firms. Can you explain what Roadrunner is and what elements of a server it encompasses.
A: Roadrunner is a specification designed to be shared through the Open Compute project so that any customer, OEM or ODM can download the information and bring a product to market. The open source nature of Open Compute gives customers the ability to innovate off of the base design, either by improving upon or further customising for their exact needs.
Roadrunner was built/ based on specific feedback and requirements we received from the financial services community. This community has a variety of workloads so the goal was to define a platform that was flexible enough to be deployed in a variety of places throughout the data centre, allowing them to reduce management costs through commonality.
The Roadrunner specification was defined to cover the feature requirements for general IT, storage, cloud and high performance compute (HPC) applications, all areas that the financial industry is deploying servers today.
Q: How many, and what kind of, financial services firms did AMD work with in order to formulate the specification?
A: AMD worked with 12 global financial services Firms that provided feedback on the specification over two sessions of reviews. As mentioned in the AMD press release of Roadrunner, Fidelity Investments, was one of those firms.
Q: Why are financial services firms looking to standardise on their server hardware? Aren't they always seeking the bleeding edge in terms of performance?
A: The financial services community is looking for a standardised, open platform for several reasons. From a platform cost perspective, there can be many unnecessary features and/or capabilities on industry standard OEM product offerings that the financial segment does not need, yet they have to pay for. From a power perspective, these “extra” integrated features can amount to increased platform power. Designing the platform with a targeted feature set matching the needs of the financial segment can reduce power consumption and heat produced by the unused components.
Q: Can you give a thumbnail of AMD's 6000 series of microprocessors, and what kinds of application loads they have been designed to handle?
A: The latest version of our AMD Opteron 6000 family is the AMD Opteron 6200 Series processors. These are designed for general purpose computing as well as scalable computing environments like cloud clusters, virtualisation, and high-performance computing. With models spanning eight to 16 cores, high memory bandwidth and high memory capacity with four channels per processor, the AMD Opteron 6200 offers the highest core density for incredible scalability to handle demanding multi-threaded workloads.
Q: How is AMD integrating GPU technology with its traditional processors, and what does this mean for financial services applications?
A: In recent years, AMD has taken a leadership role in bringing to market APUs – a processor with CPU and GPU on a single piece of silicon – into the consumer market. AMD has been involved in GPU compute for the commercial market for some time with efforts focused around AMD’s FirePro and FireStream family of professional GPU graphics cards.
AMD’s vision is to bring APUs into servers because they bring a potent mix of low power design along with massively parallel compute in a single package. The three key things that these APUs will need to be successful for server-based compute are 1) ECC memory; 2) A software ecosystem and 3) A shared memory space between the CPU and GPU. We believe that in time, our Heterogeneous System Architecture (HSA) will fully support servers, bringing a rich development environment that lets the GPU compute engine really shine.
Q: AMD recently acquired SeaMicro, which builds high performance, low power servers based on Intel chips. What gives?
A: AMD acquired SeaMicro for its technology. SeaMicro has developed a Freedom Fabric, which allows for extremely dense, extremely power efficient server deployments. Our goal is to bring out SeaMicro servers based on AMD CPUs before the end of this year. Going forward, we believe that the SeaMicro fabric technology will become a key enabler for the next generation of highly dense mega data centre deployments.
Q: How is AMD working with financial application developers to ensure their offerings are tuned to work well on AMD platforms? Can you give any examples?
A: The key to any successful software engagement is enabling OS, hypervisors, and the standard development with optimisations that leverage our processors key features. For example, AMD works closely with the OS and hypervisor partners to help ensure that these products fully support all of the features in our AMD Opteron processors. We also are committed to optimise compilers, libraries, and other tools to help ensure that when people build their applications, they have access to processor features that help them get the most out of our products. We work closely with the major OS application tools such as Microsoft Visual Studio and GCC, as well as performance compilers like PGI and Open64 and create optimised math libraries like ACML to help ensure that when you are building financial applications, or any application on an AMD platform, that you are able to leverage all of the features.