Cray Launches XK6 Hybrid Supercomputer

By Douglas Perry
published

Cray announced its first hybrid supercomputer, a system that combines traditional CPUs with GPU acceleration units.

The XK6 system will be using AMD Opteron 6200-series processors as well as Nvidia Tesla 20-series GPUs. According to the vendor, the new supercomputer system is scalable to more than 50 petaflops of raw computing horsepower.

Included with the XK6 is the Cray Linux as well as a suite of unified programming tools libraries and compilers that allows developers to effectively access available computing resources. Taking advantage of GPUs in supercomputers has been especially difficult and even more so in hybrid scenarios. Cray promises that the XK6 can blend scalable hardware, software and their network.

The first organization to acquire the technology is the Swiss National Supercomputing Centre (CSCS), which will be upgrading its existing XE6m system. Thomas Schulthess, a representative for the CSCS said that he is not looking for "another GPU-based stunt to place high on the Top 500 list", but for a "general-purpose supercomputer based on GPU technology". Cray said that is XK6 system will be available starting in the second half of this year.

Douglas Perry
20 Comments Comment from the forums
  • _Cubase_
    AMD and Nvidia, together at last... So which company claims the crown on this one?
    Reply
  • TheViper
    The current world leader is 2.7 Pflops. 50 Pflops is a massive increase.
    Reply
  • zachary k
    So, where would the D-wave fall performance wise? would it even be measured in flops?
    Reply
  • speedemon
    "another GPU-based stunt to place high on the Top 500 list"

    What does that mean.. Flops are flops
    Reply
  • @speedemon: How many apps do you currently use that utilize GPU acceleration? GPUs aren't programmable enough to be useful in many scenarios, and rarely hit their theoretical throughput unless you're computing 3d graphics, hence this is a stunt.

    China did the same thing with their supercomputer, and it too was a stunt, purely to overcompensate for their collectively small p3nis.
    Reply
  • oh, so that is how you get AMD chips to catch up to an i7 processor
    Reply
  • Kraszmyl
    If i recall correctly quickpath isnt bi-directional so hypertransport im betting is the better choice for something is large. Also in theory the gpus can tap into hypertransport where as last i was aware quickpath is locked down.
    Reply
  • memadmax
    But does it play crysis?
    Reply
  • GPU_daredevil is asking exactly the right question. This is all about real applications running on GPGPUs. Cray will need to show real applications performance as it ships systems. Cray is currently working on many applications with customers like CSCS and ORNL and using OPENMP/MPI methods and seeing early results that we believe will be promising for materials science, fusion, chemistry etc. It is good to be skeptical!
    Reply
  • The Greater Good
    gpu_daredevil@speedemon: China did the same thing with their supercomputer, and it too was a stunt, purely to overcompensate for their collectively small p3nis.
    You American have-a such-a large-a penis. Our penis, so small.
    Reply