HP has the largest range of servers and server types of any vendor, or at least they will when HP’s Project Moonshot ships into the market.
At that time their line will start with tiny ARM and Intel Atom-based technologies and scale up to Itanium-powered large-scale enterprise systems, which will also give them the biggest range of x86 servers as well. This should provide an unprecedented level of scalability, though that is somewhat offset by the different x86 architectures used throughout the line.
I think we are seeing a massive server change coming into the industry. If you look hard at these new mobile processor powered servers you may be seeing the birth of what will become the prevailing architecture, once these servers are in market and embrace the massive move to multi-threaded applications.
If this happens, Project Moonshot could give HP a huge initial advantage in what could be as big an event as client/server was, tied to the future of the Internet and Cloud Services.
Let’s explore the powerful potential future that HP’s Project Moonshot aggressively anticipates and how the company is hedging its bets.
About five years ago Intel started talking about massive multicore systems and Microsoft started talking about data centers as computers. Both were anticipating a similar world and yet approaching that world in very different ways.
Intel, the chip maker, thought up unique processors, which would eventually contain thousands of cores, with the “eventually” part years if not decades off into the future. Microsoft was talking about purposefully building datacenters from server components so that the end result was one massive multi-core server. But to keep costs down they wanted to build using low cost processors. Kind of like RAID, yet rather than a redundant array of independent disks, though applied to processors, and they initially started experimenting with Intel’s Atom processors. But Intel wasn’t happy with that move, opening the door to an ARM based alternative.
At the core of the move to a massive multicore model was the fact that processors were becoming increasingly inefficient at high frequencies and that, if this continued, the heat in systems would reach the point where all metals lost integrity and melted.
Massive cryonic cooling methods, outside of labs, proved impractically expensive and would only defer – not remove – the problem, so dividing up the work became the way to approach it.
Currently even cell phones are moving to this approach to improve battery life and reduce their effectiveness as pocket/purse heaters. Finally, neural networks, or the efforts in place to use computing architectures to create truly intelligent systems, are by nature a multi-processing environment because the decisions reached require a high degree of parallelism. Given most systems are expected to be intelligent by mid-century we were likely to go massively parallel anyway, eventually.
HP’s initial announcement of a massively multicore system was based on ARM hardware, which was also likely because Intel was slow to embrace the concept of low cost processors in servers. This put Intel at risk at both ends of the spectrum.
This was made worse when Intel’s lead partner in this space, SeaMicro, was acquired by AMD. AMD was clearly going to take SeaMicro off of Intel and try to move into the gap. On paper Intel was suddenly out of the running for what could well be the next generation of server.
HP recognized this would be a horse race and then partnered with Intel to create a competitive offering. Recognizing that Intel is the best resourced of all of the players – and now very motivated to spend massively to assure they wouldn’t lose the server business —– HP created a second platform based on Intel. So no matter what horse won, AMD or x86, HP would be in the winner’s circle.
Each platform has its own unique advantages and AMD, also a close HP partner, offers an interesting and potential hybrid of the two platforms, bringing in GPU computing, which is already massively parallel by nature, as a potential third offering.
In the end what HP appears to be trying to do is own this new generation of massively parallel servers, which will likely become the next standard platform regardless of which core technology wins. Moonshot is an appropriate name because HP is shooting for the Moon here – and they may just make it.