ARM's Race: An Attack Plan For Servers and Mobile
HP's Project Moonshot
HP was on-hand to discuss Project Moonshot, its initiative to build a dense server product around ARM and x86 cores. While originally discussed explicitly as an ARM platform, Moonshot has evolved since its inception and now features an array of x86 cores from Intel and AMD with ARM core options from Texas Instruments and Applied Micro ramping now. No word on whether Project Moonshot will feature AMD's upcoming Seattle core, though that seems a fairly safe bet.
Again, much of this discussion is tromping over familiar ground, with an exploration of how ARM's dense servers can be used in "good enough" configurations and stacked at scales that x86 has historically struggled to match. HP implied that the interconnect fabric offered by CCN-508 is a bonus because it allows the company to hang FPGA or other custom accelerators directly alongside 64-bit ARM CPUs at higher densities than a conventional x86 SoC can offer. Exactly how well these specialized applications scale to the general market is an open question, but ARM servers are clearly moving towards prime time.
HP also hinted at the deployment of a next-gen optical interconnect system for cross-node communication, as well as a future non-volatile memory interface possibly based on memristors (memory-resistor non-volatile storage competitive to Flash). The company has previously hinted at these capabilities and we know that next-gen memristor technology is in the works, but wasn't ready to disclose or demonstrate anything definitive in either field. As always, we remain hopeful for the kind of breakthroughs HP is discussing here, but dubious about the near-term commercial applications -- even in cases where the basic technology is sound, ramp and scale almost always take longer than initially forecast.
Again, much of this discussion is tromping over familiar ground, with an exploration of how ARM's dense servers can be used in "good enough" configurations and stacked at scales that x86 has historically struggled to match. HP implied that the interconnect fabric offered by CCN-508 is a bonus because it allows the company to hang FPGA or other custom accelerators directly alongside 64-bit ARM CPUs at higher densities than a conventional x86 SoC can offer. Exactly how well these specialized applications scale to the general market is an open question, but ARM servers are clearly moving towards prime time.
HP also hinted at the deployment of a next-gen optical interconnect system for cross-node communication, as well as a future non-volatile memory interface possibly based on memristors (memory-resistor non-volatile storage competitive to Flash). The company has previously hinted at these capabilities and we know that next-gen memristor technology is in the works, but wasn't ready to disclose or demonstrate anything definitive in either field. As always, we remain hopeful for the kind of breakthroughs HP is discussing here, but dubious about the near-term commercial applications -- even in cases where the basic technology is sound, ramp and scale almost always take longer than initially forecast.
