Cray launches XK6 “hybrid” supercomputer, 50 petaflops on tap
Seattle supercomputer maker Cray is joining the hybrid craze with an amazing new machine that combines AMD x86 processors with Nvidia graphics processors. — The Cray XK6 scales up to more than 50 petaflops of processing capability.
Context & Ripple Effects
The XK6 arrives at a delicate moment for AMD. The company's chief executive exited in January 2011 after months of internal tension, reportedly because board members felt he wasn't moving fast enough into mobile, even as AMD pointed to shipments of more than a million accelerated processing units and over 35 million DirectX 11 graphics chips. In Cray's hybrid design, AMD supplies the x86 processors but Nvidia supplies the accelerators — so AMD's silicon anchors the machine while its own GPU line is bypassed for the parallel workload.
First-order effects
- Government labs and research centers buying top-tier systems get an upgrade path from Cray that scales past 50 petaflops, making hybrid CPU-plus-GPU machines a first-class procurement option rather than an experiment.
- Nvidia gains a marquee high-performance-computing design win inside a Cray system, validating its accelerator business at the very top of the market where Cray competes.
Second-order effects
- Rival system vendors selling conventional CPU-only supercomputers now face pressure to field their own hybrid architectures or concede the largest deals to Cray-and-Nvidia pairings.
- AMD faces an awkward channel dynamic: every Cray hybrid sold strengthens Nvidia's accelerator franchise even as AMD's server CPUs ride along, deepening the strategic tension that already cost it a chief executive in 2011.
Third-order effects
- If buyers follow Cray's lead, extreme-scale computing consolidates around the handful of vendors who can integrate both processor families, and software ecosystems for GPU acceleration — not raw processor counts — become the deciding factor in supercomputer procurements.
The trend: Top-end supercomputer design is shifting from homogeneous CPU clusters to heterogeneous CPU-plus-GPU hybrids, with accelerator vendors like Nvidia claiming the highest-value slot in the stack.