AppliedMicro confirmed that it had shipped samples of the X-Gene on April 3, after a story and photos ran on GigaOM.com. The server-on-a-chip design from AppliedMicro is interesting in part because it uses the ARM design and has hit the market before other products from other licensees.
What's more interesting is that AppliedMicro's design includes onboard networking, complete with a programmable Ethernet switch controller with an ARM Cortex A5 core at its heart (the same core used in the latest smartphones). By including very fast switch controllers and interconnects on the chip, AppliedMicro has created an ideal chip for high-density servers. By using ARM's 64-bit design as the CPU, it's likely these chips will perform on a par with Intel's Xeon chips. The opens up a new architecture for scale-out server design that to date hasn't been possible.
So far, microserver architectures have shown how a single rack could easily house thousands of lower-powered servers that could be well suited to tasks like scale-out Web services. By using ARM-based chips that consume about 6 watts of power rather than Xeon chips, which will chew up about 100 watts, manufacturers could achieve much greater chip densities within a single enclosure. HP's Moonshot is an example of new architectures that are possible with microservers (though it should be noted the Moonshot 1500, HP's first product out of the gate, is based on Intel's Atom processor).
AppliedMicro has eight cores running at 2.4 GHz, and the ARM design has the advanced features you'd find on an Intel chip, including full hardware virtualization support. As such, it's likely that it will consume a good bit more power than the existing 6 watt ARM chips, but a lot less than the 100 watt Xeon.
[ Join us at Interop Las Vegas for access to 125+ IT sessions and 300+ exhibiting companies. Register today! ]
With an onboard networking switch that can support 10-Gbps Ethernet ports now and 100-Gbps Ethernet ports in future versions, it will be a very attractive building block for Web-based enterprises with truly massive computing and I/O needs like Facebook and Google. With its built-in networking, the X-Gene architecture could knit together over 1,000 cores without the need for an external layer 2/3 switch. Take that, Cisco and Arista!
As the server-on-a-chip name implies, it also supports peripherals through PCI-E 3.0. The internal network of the second-generation chip has a total bandwidth of 160 GB/second.
Intel certainly hasn't been blind to the need for chip-level interconnects that will allow for building massive scale-out computing systems. Last April it bought the networking assets of supercomputer manufacturer Cray. In 2011 it bought Fulcrum Microsystems, which made Ethernet switch chips, and in 2012 it bought QLogic's InfiniBand business. The result is that Intel sees including interconnect logic on the chip, and probably full-blown fabric controllers, too. Some will be InfiniBand, some will be Ethernet, but so far, none have made it to market.
That AppliedMicro has been able to beat Intel to market, or at least to samples with compelling products, shows the power of ARM's licensing strategy. And if AppliedMicro has an interesting spin on the ARM server core, it's a pretty good bet that AMD, Samsung and others have some good ideas, too.
As it turns out, Intel's giving up the smartphone market to ARM was probably as a bad a stumble as large computer manufacturers giving up the PC market to Intel some three decades ago. There's nothing like owning a low-end market to allow chip makers and designers to push into higher and higher-end markets. Now it looks like ARM and its licensees are ready to battle with Intel. The result will be kickass servers.