HP launches Project Moonshot, powered with Intel's Atom at first

HP launches Project Moonshot, powered with Intel's Atom at first

Summary: HP's bet to be a server leader revolves around hyperscale servers, targeted workloads, power savings and better density. At stake is HP's reputation as an innovator.


Hewlett-Packard launched its Moonshot server line with an Intel-based Atom system available today and more chipsets planned in the future. The promise: HP will create Moonshot hyper scale software defined servers for custom workloads and an ecosystem.


The bet for HP is that it can launch new Moonshot systems at three times the product cycle of traditional servers. For HP, Moonshot represents the company's ability to innovate, remain a server leader and keep up with cloud customers, who are increasingly building their own gear.

HP CEO Meg Whitman started a Webcast by talking "brontobytes of information," argued that the Web on its current path isn't sustainable. "The data center space needed will be 200 football fields" to meet demand, power plants will have to be built, she said.

The case for Project Moonshot, a multi-year effort, is that these systems will cut space and power in the data center. "It's the foundation for the next 20 billion devices," said Whitman.

Overall, HP is pitching a "library" of Moonshot servers and an architecture that's designed for developments like the Internet of things and machine-to-machine applications. Initial reference customers include Savvis, a data center hosting and cloud company, and Purdue University. 

HP executives likened Moonshot systems to the move from Unix to x86 servers and the innovation created from the use of blade systems. 

Dave Donatelli, general manager of HP's enterprise group, said HP has had more than 50 beta customers for its first-generation Moonshot system. The second gen of Moonshot will be geared to creating a software defined server. Donatelli quipped that the industry will look back at the Moonshot launch 10 years from now and note the event as a marker where the server industry changed. 

Key points:

  • HP Moonshot servers can support 1,800 servers a rack.
  • Each chassis shares components for cooling, power supply and management software.
  • The Moonshot 1500 uses Intel's Atom S1200 processor.
  • HP is working on an ARM-based server in partnership with Calxeda, the partner for the first-gen Moonshot server. A Calxeda spokesperson said that there will be an ARM-based Moonshot on deck. The company said it can support HP's launch schedule to have Calxeda-based Moonshot servers later this year with up to 180 quad-core in the 4U Moonshot server chassis. 
  • Pricing starts at $61,875 for the enclosure, 45 HP ProLiant Moonshot servers and an integrated switch.
  • The first Moonshot system runs on Linux, but is compatible with Windows, VMware and traditional enterprise applications since it's x86 based via Atom. 
  • Power savings data is based on comparison to HP's dl300 servers. 


According to HP, the first Moonshot system is available today in the U.S. and focused on cloud workloads. The system is powered by Intel's Atom processor. More systems will have multiple partners and target verticals ranging from genomics to cloud to video to analytics, telecom and gaming. A global and channel rollout begins next month.

Here's a look at other chip partners for Moonshot. 





Donatelli also plugged an ecosystem of software developers who will optimize applications for Moonshot hyperscale servers. Donatelli argued that enterprises and cloud companies won't be able to simply add more servers over and over. Power and space will eventually run out.

The big question for HP is how long the Moonshot architecture will take to proliferate. Blade servers didn't take off immediately. Today, Moonshot is for large installations, but over time HP expects the architecture to broaden to other use cases at smaller companies. The reality is that few enterprises would qualify for a hyperscale server treatment. As HP rolls out systems for workloads such as analytics, genomics and financial services that equation will change. 

The nuts and bolts were explained by Patrick Moorhead, principal of Moor Insights & Strategy. Moorhead said in a white paper:

The HP Moonshot 1500 System chassis is similar to a blade chassis, but on steroids. It is a 4.3U (7.5 inches tall) chassis that hosts 45 independent hot-plug ProLiant Servers, all attached to multiple fabrics. Like a blade chassis, the HP Moonshot 1500 System chassis sports shared power, cooling, and management resources for those server

Unlike blade chassis, it does not have a single, fixed, and shared interconnect backplane. It contains three independent network fabrics – an Ethernet switch fabric, a storage fabric, and a cluster fabric. Each ProLiant Server has access to all three fabrics.

In the HP Moonshot 1500 System, network access for its server cartridges is implemented as two removable Ethernet switch modules that can be configured for redundancy or maximized bandwidth. The initial switch modules implement 1Gbps links, and each server cartridge may have up to four 1GbE links to each switch, for eight links total.

Moorhead noted that Moonshot's cluster fabric isn't in blades. Groups of server cartridges are connected via bandwidth lanes. Each cartridge has storage lanes. This separation of rack-level network traffic boost throughput.

The only planning wrinkle appears when customers mix and match server cartridges. HP doesn't have that capability yet, but will over time.

Moonshot's importance to HP

Aside from the obvious potential boost to HP's reputation as an innovator, Moonshot servers also have to pick up the slack for other businesses. According to Barclays analyst Ben Reitzes, Moonshot servers are needed to boost HP's server revenue, which will be more critical should PC sales drop off over time. The problem for HP is that cloud players are using Quanta and Wistron to make their own servers over branded systems. Reitzes noted:

Industry Standard servers represent 43% of total ESSN revenue, an area where HP remains the market share leader but has seen its share slip the last few quarters. We believe growth is being held back by share losses to “self-made” servers for the Cloud players and concerns around HP’s future. We are increasingly seeing Cloud players use ODM partners like Quanta and Wistron to make their own servers vs. using products from HP, IBM and Dell. Quanta’s servers are in competition with HP’s Gemini platform released last June. According to IDC, HP’s shipments last quarter were -6% y/y (vs industry growth of -4% y/y) and HP’s share was down to 32.0% from 32.6% last year. We believe HP must have a strong x86 server business to drive sales of storage and software. It may also be increasingly hard to sustain a leadership position in servers if HP loses ground in corporate PCs.

Reitzes last point is worth noting. HP needs servers to do well so it can sell more gear elsewhere.

HP is betting that Moonshot servers can represent 15 percent of the market by 2015. For now, Moonshot servers aren't being highlighted in HP's Austin data center, which is used to showcase the company's latest systems and software, but do run HP.com.

Topics: Servers, Data Centers, Hewlett-Packard

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


Log in or register to join the discussion
  • ARM Offering

    If they were smart they would have launched an ARM offering at the same time. HP has been burned by intel i nthe past looks like they just don't learn. They have a gigantic patent library you can't tell me they can't design a new processor platform that would smoke the competetion; hell have a company like mips do it for them.
    • Remember Itanium?

      They tried your idea and wasted billions on it. I personally think that the only reason that project lasted as long as it did was because some of the highly paid people that made those decisions needed to save face.

      Just because you can do something doesn't mean it is a good idea.
    • ARM

      The ARM processor is only now starting to implement the features and capabilities that will allow it to be used on servers and workstations.

      In short, ARM has only recently defined its 64-bit support for ARMv8 processors: http://en.wikipedia.org/wiki/ARMv8#ARMv8_and_64-bit

      The first ARMx8 processors are only now starting to sample and won't be available in volume until later in 2013.

      Interestingly, by the time ARMv8 chips actually start shipping, Intel's next-gen "Silvermont" Atom chips will also start shipping in greater volume and will be followed in 2014 by the "Airmont" 14nm Atom chips.

      This means that while ARM's processors will start to increase in perf (and power consumption), Intel is going to quickly be approaching the perf/watt levels of next-gen ARM chips whilst offering the considerable benefit of the venerable x86/x64 ISA.

      It's going to be an interesting few years in the ARM vs. Intel space ;)
  • They should have used the ARM processor

    I thought that this microserver angle was supposed to be covered by using Virtual Machines, if your workload is light enough that you would consider one of these units then a perhaps using a VM is what you should be doing. I'd be willing to bet that 3 of these units is more expensive than a normal server running 3 VMs.

    ARM processors still use less power than Atoms. ARM's probably cost less and would scale well in this application. The only reason that I can think of to use an Atom is to support Windows servers which in the cloud is not so important.
    • Why? Peer pressure?


      The "let's all jump in the other bandwagon" cant isn't so simple...

      And it depends on a target market - whether perceived, actual, or one that needs to be conjured up (with extra marketing and other aimed psychological manipulation)...
  • Not one but two hilarious 'Dilbert' Jokes Part II

    Scene: staff meeting. Pointy-Haired Boss listens in awe as CEO explains in front of canvas.

    CEO: Our systems will cut space and power in the data center
    CEO: It's the foundation for the next 20 billion devices
    Dilbert: You should mention, that a billion is a one with 12 zeros
    PHB: No need to. This message is targeted at other CEO's. They know that without us telling!
  • I bet this is under the skin of anyone that knows about Transmeta

    The Atom steals so much of their technology, except it does not have an integrated Northbridge.

    Some of the folks from Transmeta tried to pitch these ideas well over a decade ago. Including to HP. Had HP listened, they would be ahead, not trying to catch up. I guess the same thing sort of happened when Woz tried to pitch a personal computer to HP management in the 70's.
    • Ermmmm ... no.

      Transmeta's processors were more alike Itanium than Atom in that they dynamically re-compile several x86 instructions (atoms) into single (128-bit "Crusoe", 256-bit "Efficeon") VLIW instructions (molecules).

      Intel's Atom is MUCH simpler: Based on the "Bonnell" micro-arch they just map each x86 instruction into one or more 32-bit RISC instructions using hardware-based decoder.

      Sadly, like Itanium, VLIW processors (and many RISC processors) have failed to deliver on their predicted performance per clock promises. This is partly because x86/x64 code is extremely dense due to the variable length of it's instructions. This means that more x86/x64 code can fit into cache and instruction pipeline resulting in considerable performance benefits.
      • It isn't tremendously different

        Whether you call it "dynamically re-compile" or "map each x86 instruction into one or more 32-bit RISC instructions" is the same thing.

        "dynamically re-compile" is shorter though.
        • No, you're right, it's not tremendously different ...

          ... it's ENTIRELY different.

          Oh, and FWIW, Transmits didn't invent dynamic recompilation: Intel's P4 architecture already recompiled X86 into RISC internally and DEC built a dynamic recompiled that converted X86 into Alpha 64-bit instructions. Both long before Transmits were created.
      • Re: Intel's Atom is MUCH simpler

        Is that because "map" is shorter than "dynamically recompile"?
    • a third the promised performance

      HP was about the only sucker to ship Transmeta - in the TC1000 tablet - because it promised the best battery life with decent performance. It delivered the battery life but performance was, the system designer told me later, only a third of what Transmeta promised.
  • Not one but two hilarious 'Dilbert' Jokes Part I

    Meg's phrases inspired me to some ideas I should pitch to Scott Adams:

    Scene: Executive meeting. Pointy-Haired Boss explains in front of canvas.

    PHB: Information from Brontobytes magazine tells us, the data center space needed will be 200 football fields
    PHB: Our CEO has launched a new, brilliant strategy to make as the market leader
    PHB: Send Dilbert to Elbonia to purchase cheap football fields
    Cue in football field in Elbonia with a sign "For Sale"
  • A very big bet for HP

    Just hope it at least breaks even for them.
  • Project Moon What?

    "At stake is HP's reputation as an innovator." When has that ever been at stake? What have they innovated besides competing on price with low quality product made by indentured servants in China and Mexico? Seriously... Name me a "must have" product HP last produced? Ok, just name a cool product that HP last produced? Ok, name me a product that doesn't suck that HP last produced? Ok, let me name a few exciting innovations for you: desktops, laptops, servers, printers, mediocre handhelds, rank/horrible thin clients and tablets. Oh yes and overpriced workstations that you can get on eBay after they go EOL for 1/10th of the price. Moonshot, they should call it Alpha Centauri shot....
    • rubbish

      HP is still the only company doing the full gammit of IT from PC, Server, Networking & storage hardware though software, cloud & services. You could argue that they could do it better, but the new executive regime looks well engaged and making a difference (more now that the Board has been shuffled). Innovation? This "moonshot" (dumb name) is innovative. Their SDN story is innovative. Their security story (Tipping Point, Fortify) is innovative. HP Cloud is mixed. Printing? And so it goes. Put IBM, or Dell in the crosshairs, and what do you get? Exactly.
      • "gamut" - stop proving to corporations Americans are uneducated

        It's innovation but it's being given more hype than actual praise.

        Let's see this stuff stand the test of time as well. With cost and corner cutting being the new normal everywhere, there's bound to be a hidden cost compelling customers to upgrade more often - for one reason and/or another. These companies really do not put their customers' limited budgets in mind, or the workers whose labor that makes the company rich either...
      • Not rubbish at all

        The full gamit? When have they ever provided a complete solution? They missed mobile 100%. Only now, 10 years after it began, are they starting to understand that mobile technology and BYOD is a major trend. IBM and Dell suck too: the whole "let's use Microsoft and Intel" as our R&D team and run all of our marketing dollars thru them might not be panning out as they'd hoped. Let's face it, HP has RUN OUT OF STEAM as has the rest of this poorly articulated, short term market. The real problem with HP is that the market they have ridden as far as it can go is done and that the hedge fund managers that run the company don't understand what real technology development (and risk) is supposed to look like. HP is a bank, a financial services company (think HPFS). The only difference between them and JPMC or Citi is that they hedge on technology product and markets. They've been talking about Moonshot for almost 4 years! When's the next innovation? Oh right, an Android tablet in June (or July or August or 2014). Good riddance.
        • the first Pocket PC phone

          HP has squandered more innovations than any other two companies; it had both iPaq and Jornada and a huge share of the PDA mobile market. It would have had the first PDA phone except it built it and didn't ship it.
      • PS - One company that doesn't suck....

        One company that doesn't suck that you forgot to mention was Lenovo. My Thinkpad 420 works great 100% of the time. So does every other Lenovo product I have. They are starting to produce servers as well. How long before they start to eat Moonshot share?