About the Author

Douglas EadlineDouglas Eadline PhD, is both a practitioner and a chronicler of the Linux Cluster HPC revolution. He has worked with parallel computers since 1988 and is a co-author of the original Beowulf How To document.  Prior to starting and editing the popular http://clustermonkey.net web site in 2005, he served as Editor-in-chief for ClusterWorld Magazine. He is currently Senior HPC Editor for Linux Magazine and a consultant to the HPC industry. Doug holds a Ph.D. in Analytical Chemistry from Lehigh University and has been building, deploying, and using Linux HPC clusters since 1995.

User Rating: / 1
PoorBest 
A blog about making HPC things (kind of) work

The recent release of the AMD "Llano" processors may not have attracted very much attention in the HPC world. It should have. The Llano processors are some of the first Fusion processors to hit the market. Simply put, the AMD Fusion processors are placing both the CPU and GPU on the same die. The archtecure is different than a typical desktop or server processor and AMD calls these new devices Accelerated Processing Units (APUs).

Let's take a look at one of the most powerful versions, the A8-3850. This processor has four x86 cores (not the new Buldozer cores, however) plus 400 Radeon cores. That should be impressive enough, but there is more. First, the A8-3850 is rated at 100W running at 2.9GHz. Each core has a 4MB L2 cache and Radeon Cores run at 600MHz. There is plenty to understand about this new design, but the two main points are:

  1. A substantial GPU and CPU are on the same die and consume 100W
  2. The GPU and CPU share the same memory
These are two game changing milestones. First, for a majority of laptops and desktops, an extra graphics card in no longer needed for good graphics. There are less things to break, less heat, less problems. The second important point is the CPU and GPU memory are fused. That is, there is not longer separate GPU and CPU memory that must be exchanged across the PCIe bus.

In the non-HPC world these are important breakthroughs, but as I said in HPC these are "game changing" developments. Consider that in HPC, the GPU is used as a parallel computing device (or SIMD unit). The history of computing shows that integration of acceleration devices is all but certain. The AMD APU is the first step in this process. Look for Nvidia (using ARM) to follow. The creation of a global memory should remove many of the bottlenecks that have plagued GPU computing since its inception. (i.e. moving data across the PCIe bus can offset any acceleration from the GPU). I'll have more on this and talk about the software issues in another post.