Hybrid Computing Gets a Datacenter Boost
Even if you are IBM there is always concern that your new hardware releases will get lumped into the ""me too" category, especially when there is that flurry of hardware announcements around the latest generation of Intel server processors. But with the release of the IBM System x iDataPlex dx360 M3 IBM gets to stand up and say "me first."
The new iDataPlex system is the first offering from a top tier server vendor to be built around not just high performance server processors, but also state of the art GPUs; in this case the new NVIDIA Tesla M2050. While HP has offered add-in accelerators utilizing GPUs since late 2007, the new iDataPlex servers are designed from the ground up as hybrid computing systems with all of the energy efficiencies that the iDataPlex servers have been designed to provide in the datacenter environment.
Using GPUs in high-performance computing (HPC) environments has been a growing technology as the performance of these processors can be of significant benefit to compute intensive applications. The NVIDIA Compute Unified Device Architecture (CUDA) programming environment to take advantage of these processors is beginning to reach critical mass, according to NVIDIA, with over 350 universities offering instruction in programming and with 10 books on the topic appearing in the last two years.
The real hook for IBM here is not just in the native integration of the GPU technology into the servers, but the implementation of this technology in a server system designed for high-efficiency and high density. These HPC-focused GPU systems have been making their inroads installed in clustered computing environments and with green IT an important consideration, simple computing power won't be the only factor in the purchase decision, if an energy efficient option is possible.
The new iDataPlex also gives IBMan "entry-level" system into the world of HPC for their customers. As IBM already has a significant portfolio of HPC devices, primarily based on their own PowerXCell technology. And while HPC technology has been sold primarily into markets that require massive amounts of computing capability for vector math (such as oil and gas exploration) the availability of hybrid servers from IBM likely portends this capability becoming more common, as other vendors release their own native hybrids. Once this hardware becomes commonplace, applications that can take advantage of the GPUs and appeal to a broader market segment will undoubtedly appear.