Start Your Compute Engines

As enterprise IT evolves, it’s pretty clear we’re entering a new age of specialization around different types of application workloads. Rather than thinking in terms of general-purpose computing platforms, we’re seeing the emergence of both physical and virtual appliances that are optimized to run certain types of application workloads.

But there is a lot of nuance in terms of how IT organizations will master these various types of compute engines, especially when you consider that a virtual appliance may be running on top of either a virtual machine or on a subset of a system that has been specifically optimized for that type of application workload.

The ability to create platforms that are optimized for specific types of application workloads is leading to a convergence of graphics, transaction-oriented and general-purpose computing engines, which will all need to be brought together under a common development framework to create the next generation of enterprise applications.

In the case of Advanced Micro Devices, that convergence manifests itself in the form of an accelerated processing unit (APU) architecture, called Fusion, that integrates graphics processing alongside traditional x86 processors and a next-generation ‘Bulldozer’ Opteron processor architecture that better balances processor speed, memory and CPU interconnects to create of platform that can integrate application workloads at much faster speeds across as many as 16 processor cores.

According to Vlad Rozanovich, director of North America commercial business for AMD, this approach means that the overall application environment is going to be much faster, even if the increase in the actual clock speed of the processors is relatively minor. Rozanovich says AMD foresees more application workload convergence in the years ahead, which in turn will inform the design of future processor architectures that will increasingly need to focus on reducing the communications latency between the various elements of the overall architecture that have been optimized for specific types of application workloads.

The real challenge, however, will be training developers to think in terms of parallel rather than sequential processes, which is a topic that AMD will focus on at its upcoming Fusion Developer Summit. These new architectures are a far cry from one application per server that most developers are comfortable with today. But longer term, they should yield much richer applications where different types of application workloads are more seamlessly integrated with one another to create something that is greater than the sum of its parts.