Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
November 2, 2011

First HPC Cluster with AMD Fusion Chips Debuts at Sandia

Michael Feldman

Sandia National Labs made a bit of HPC history this week when it announced it had installed the first HPC cluster outfitted with AMD’s Accelerated Processing Units (APUs), based on the chipmakers ‘Fusion’ processor design.  The chip integrates x86 CPU cores and ATI GPU cores onto the same processor. 

This is an experimental system only, to be used as a platform to evaluate the CPU-GPU heterogenous processor model of supercomputing.  Currently, AMD’s Fusion chips are designed for personal computing platforms and are not intended for servers.  But apparently Sandia’s Scalable Computer Architectures group was eager enough to get their hands on an APU cluster to contract Penguin and AMD to be build them a system.

According to the press release, the Penguin system, known as the Altus 2A00,  was specifically designed with AMD APUs in mind.  The QDR InfiniBand-based cluster is comprised of 104 servers, which houses an unspecified number AMD A8-3850 APUs.

Designed for desktop PC duty, the A8-3850 a quad-core x86 design integrated with 400 Radeon (HD 6550D) cores. The CPU cores run at 2.9 GHz, while the GPU side runs at a more modest 600 MHz.  TDP is a respectable 100 watts.

The whole cluster delivers 59.6 peak teraflops, but the idea here is not to break any performance records. The Sandia researchers will use the machine to explore programming models for integrated CPU-GPU platforms, relying primarily on OpenCL and MPI. The draw here is to be able to access the considerable vector capabilities of the GPU within the same memory space as the CPU.

The rationale for such research is that heterogeneous processors such as these APUs will be the basis for future exascale computers.  In a roundtable discussion organized by AMD on Wednesday, AMD Technology Group CTO Chuck Moore noted that while their x86 server CPUs, like their new Interlagos Opterons, continue to make strides, those designs will likely not be the path to exaflop performance by the end of this decade.

“Instead what we need to do is to make really good use of heterogenous computing,” said Moore. “By combining the best of CPU and GPU technology, presumably on the same chip, we believe we can build computing nodes that are on the order of 10 teraflops in that timeframe.”

According to him, AMD has plans in the works for an “HPC APU that would utilize even a larger GPU and fewer x86 core” than the current desktop chips.  He thinks the company can build HPC APUs that run about 150 watts and would be powerful enough to power an exaflop computer that consumes no more than 20 MW. And, thanks to the integrated CPU-GPU memory space, these machines would be reasonably easy to program.

SC14 Virtual Booth Tours

AMD SC14 video AMD Virtual Booth Tour @ SC14
Click to Play Video
Cray SC14 video Cray Virtual Booth Tour @ SC14
Click to Play Video
Datasite SC14 video DataSite and RedLine @ SC14
Click to Play Video
HP SC14 video HP Virtual Booth Tour @ SC14
Click to Play Video
IBM DCS3860 and Elastic Storage @ SC14 video IBM DCS3860 and Elastic Storage @ SC14
Click to Play Video
IBM Flash Storage
@ SC14 video IBM Flash Storage @ SC14  
Click to Play Video
IBM Platform @ SC14 video IBM Platform @ SC14
Click to Play Video
IBM Power Big Data SC14 video IBM Power Big Data @ SC14
Click to Play Video
Intel SC14 video Intel Virtual Booth Tour @ SC14
Click to Play Video
Lenovo SC14 video Lenovo Virtual Booth Tour @ SC14
Click to Play Video
Mellanox SC14 video Mellanox Virtual Booth Tour @ SC14
Click to Play Video
Panasas SC14 video Panasas Virtual Booth Tour @ SC14
Click to Play Video
Quanta SC14 video Quanta Virtual Booth Tour @ SC14
Click to Play Video
Seagate SC14 video Seagate Virtual Booth Tour @ SC14
Click to Play Video
Supermicro SC14 video Supermicro Virtual Booth Tour @ SC14
Click to Play Video