ARM Servers - Calxeda Opens The Kimono For A Tantalizing Tease

Calxeda, one of the most visible stealth mode startups in the industry, has finally given us an initial peek at the first iteration of its server plans, and they both meet our inflated expectations from this ARM server startup and validate some of the initial claims of ARM proponents.

While still holding their actual delivery dates and details of specifications close to their vest, Calxeda did reveal the following cards from their hand:

  • The first reference design, which will be provided to OEM partners as well as delivered directly to selected end users and developers, will be based on an ARM Cortex A9 quad-core SOC design.
  • The SOC, as Calxeda will demonstrate with one of its reference designs, will enable OEMs to design servers as dense as 120 ARM quad-core nodes (480 cores) in a 2U enclosure, with an average consumption of about 5 watts per node (1.25 watts per core) including DRAM.
  • While not forthcoming with details about the performance, topology or protocols, the SOC will contain an embedded fabric for the individual quad-core SOC servers to communicate with each other.
  • Most significantly for prospective users, Calxeda is claiming, and has some convincing models to back up these claims, that they will provide a performance advantage of 5X to 10X the performance/watt and (even higher when price is factored in for a metric of performance/watt/$) of any products they expect to see when they bring the product to market.

Claims of future performance against current competition are always subject to some level of skepticism, but Calxeda’s performance models that they shared with us appeared to be conservative and to have factored in a generous margin for competitors Intel and AMD to improve their performance per watt metrics. Actual benchmark performance will probably be all over the map, considering that x86 vendors have immense accumulated experience tuning benchmarks for their platforms and the ARM community is still in early days on performance tuning. When these systems are finally available, I&O groups would be well advised to emphasize benchmarking their own application stacks.

As far as I can tell, this initial peek at Calxeda’s offering reinforces my contention that I&O professionals should include ARM servers in their strategic technology plans.

See ARM-Based Servers – Looming Tsunami Or Just A Ripple In The Industry Pond? and Put ARM-Based Servers On Your Server Planning Horizon for additional background.

Comments

how about 64 bit, Ecc, virtualization, memory size?

in talking to top 10 web datacenters, 64 bit, and memory size are essential for mass deployment in their main clusters. There are still smaller clusters that may use 32-bits processors, but the TAM for those servers is minimal; may be 100K server in total. That's about 1% of the server market. In my opinion ARM will not compete until 64-bit cores are available.

64 bit, virtualization, etc.

The First commenter totally misses the point of these low-cost, high density ARM based servers and tries to logically stuff them into the same data center topology so common with today's expensive, low-density '86-based architectures. When "servers" are as cheap as a decent lunch meal, and are the size of a pack of Marlboros, exactly what is the need for software virtualization? Hmmm? VMs were created to fix an under utility problem with big, fat, hairy and expensive Big Iron servers. As for 64-bit, that's also an artifact of the trend to software virtualization. Whack away the VM overhead, trim back the OS load, and all of a sudden 64-bit offers no real-world value advantage over 32-bit.

I do not know if Calxeda is going the right direction in the price vs. performance factor insofar as what is possible with ARM based server systems. That best case direction for maximum sales in a shifting market would be toward the cheapest, lowest density approach (blades) possible. Sound to me like these guys are thinking more up-market than this, and that's probably a bad move. Early adopters/experimenters will be put off by an ARM solution as pricey as the existing solutions.

Just rambling...