Why isn't Calxeda's ARM processor running at 2GHz or more?

Calxeda made a lot of news with multiple news articles.

HP Builds Servers With Cellphone Chips

New York Times (blog) - ‎4 hours ago‎

Hewlett-Packard announced on Tuesday a new design for some of the world's largest computer centers and says it could reduce power consumption in some cases by 90 percent. The design, called Project Moonshot, replaces the conventional...

Calxeda Stretches ARM into the Clouds

Wired News - ‎4 hours ago‎

By jonstokes On Tuesday, Austin-based startup Calxeda launched its EnergyCore ARM system-on-chip (SoC) for cloud servers. At first glance, Calxeda's looks like something you'd find inside a smartphone, but the product is essentially a complete server ...

Calxeda Introduces EnergyCore ARM Processor for Servers

PC Magazine - ‎4 hours ago‎

The EnergyCore can draw as little as 1.5 watts for a dual-core server chip, and can create a full quad-core server, including 4G of DRAM and an SSD, that draws just five watts of power. But what makes the chip stand out, ...

ARM Breaks Into One Of Intel's Strongholds (ARMH, INTC, HPQ)

San Francisco Chronicle - ‎5 hours ago‎

Chip designer ARM has long dominated the market for smartphones and tablets, leaving Intel scrambling to catch up. Now, it's getting chance to break into one of Intel's strongholds -- server hardware. This morning, Hewlett-Packard announced a plan ...

HP Plans Low-Power Servers Using Calxeda ARM Chips

InformationWeek - ‎5 hours ago‎

HP's tiny servers built on Calxeda's energy efficient Cortex chip are designed to handle large Web data streams, video processing, picture uploading, or Hadoop-style big data analysis. By Charles Babcock InformationWeek HP on Tuesday launched Project ...

But, I don't know if I am as excited as the press is.  Why?  I have been talking to ARM for over a 3 years on the opportunities for ARM servers in data centers and have been waiting for over a year for Calxeda to make a chip announcement.

So, looking at the specifications.  One of the questions i have is why does the Calxeda processor run at 1.1 - 1.4 GHz and not at 2GHz?

EnergyCore™ ECX-1000: Technical Specifications

Processor Cores Up to four ARM® Cortex™-A9 cores @ 1.1 to 1.4 GHz

Here is the spec for the A9 processor.

Speed Optimized: The speed-optimized hard macro implementation provides system designers with an industry standard ARM processor incorporating aggressive low-power techniques to further extend ARM’s performance leadership into high-margin consumer and enterprise devices within the power envelope necessary for compact, high-density and thermally constrained environments. This hard macro implementation operates in excess of 2GHz when selected from typical silicon and represents an ideal solution for high-margin performance-oriented applications.

In the HP press announcement there as quote to emphasize performance needs.  So, why not a 2 GHz clock rate?

“The volume of data processed in financial markets has increased exponentially, and traditional scale-up or scale-out architectures are struggling to keep up with demand without vastly increasing cost and power usage,” said Niall Dalton, director of High-Frequency Trading at Cantor Fitzgerald, a company that is currently evaluating the technology. “HP is taking a holistic approach to solving this problem and working to bring unprecedented energy and cost savings for tomorrow’s large-scale, data-intensive applications.”

Another question I have is what is the architecture to manage the Energy Cards.  This could be the opportunity for HP.

EnergyCard Reference Designs

EnergyCards are production-ready boards designed by Calxeda to demonstrate the full breadth of capabilities offered by the EnergyCore platform. With this as a building block, system OEMs can leverage Calxeda’s design expertise, allowing them to easily bring hyper-scale solutions to market in a fraction of the time required for ground up custom designs.

...

The HP Redstone Server Development Platform is the first in a line of HP server development platforms that feature extreme low-energy server processors. Initially incorporating Calxeda EnergyCore™ ARM® Cortex™ processors, future Redstone versions will include Intel® Atom™-based processors as well as others. HP Redstone is designed for testing and proof of concept. It incorporates more than 2,800 servers in a single rack, reducing cabling, switching and the need for peripheral devices, and delivering a 97 percent reduction in complexity.⁽¹⁾ The initial HP Redstone platform is expected to be available in limited volumes to select customers in the first half of next year.

Little (micro) green servers achieve product category at Intel Developer Forum, promotes Micro Server Lab

At Intel Developer Forum I missed a session, but luckily one of my friends from ARM went to it and said it was quite good.  Yes, there is a guy from ARM walking around an Intel conference.

The presentation is here.

For over three years I have been discussing the idea of little green servers, and now they are called micro servers.

Why did Intel give a speaking slot to micro servers?  Because in 4-5 years they think it could be 10% of the current market dominated by Xeon.

The Micro Servers are composed of Supermicro, SeaMicro, Dell, Tyan, and Quanta

Why would you want a micro server?  Dell’s Brandon Draegar presented.

It is interesting that Dell included SeaMicro in its portfolio slide.

Wonder how important micro servers are to Intel?  Intel has launched a Micro Server lab.

http://www.intelcloudbuilders.com/microserverlab

A small team is forced to design small, for example original ARM designers

The success of the ARM processor vs. Intel is obvious in the mobile market.  Going back to the history of the ARM it is interesting to note the small UK design team vs. the US.

One of the reasons the ARM was designed as a small-scale processor was that the resources to design it were not sufficient to allow the creation of a large and complex device. While this is now presented as (and genuinely is) a technical plus for the ARM processor core, it began as a necessity for a processor designed by a team of talented but inexperienced designers (outside of university projects, most team members were programmers and board-level circuit designers) using new tools, some of which were far from state-of-the-art. With these restrictions on design and testing, it is hardly a surprise that a small device was developed.

While the ARM was developed as a custom device for a highly specific purpose, the team designing it felt that the best way to produce a good custom chip was to produce a chip with good all-round performance.

In the US the RISC teams at Intel, AMD, Sun and MIPS.

For example, Sun developed the SPARC RISC chip and architecture for its own computer workstations, while notable RISC processors from established chip producers include Intel's i860 graphics processor and AMD's 29000, which has mainly been used as a graphics accelerator or in printers. However, both Sun's and MIPS' efforts were based on earlier research efforts at Stanford and Berkeley universities respectively, while Acorn's project was effectively begun from scratch, although reports on the Berkeley and Stanford research were read by the Acorn team and were part of the inspiration behind designing a RISC processor.

The ARM team also was a team focused on price/performance.

The ARM processor has always differed from other commercially available RISC processors in that it is intended to meet a price/performance ratio rather than to be the most powerful processor available. Acorn's computers have always been aimed at the middle of the market, so the processor designed to power them was too. ARM processors are not the most powerful, but offer an extremely good price/performance ratio compared to other processors, at about a dollar per million instructions per second (MIPS) in the case of ARM6.

…

High performance for low power consumption

A further advantage of the small size of ARM devices is that they do not consume as much power as other, larger processors.

This has proved a critical key to the success of ARM processors. Unlike many other processor designs, the ARM was easily re-implemented in static form rather than the usual dynamic CMOS. This, along with the small die size, reduced power consumption, making ARM processors ideally suited for power consumption-critical products such as portable computers. Furthermore, it allows the clock to be stopped, a useful powersaver in portable designs.

Think about this when picking vendors.  Sometimes the biggest is not the best as they have the resources to create the most complex designs and to market it.

Is Seamicro SM1000 series the mainframe of the Web2.0 data centers?

There has been a bunch of press on Seamicro's latest SM1000 product.  The press release is here.

SeaMicro Introduces the SM10000-64HD, Setting Industry Record for Energy Efficiency and Compute Density

July 17, 2011

With 384 Intel® Atom™ Dual-core 1.66 GHz Processors; 768 64-bit Cores and 1,275 GHz in a 10 Rack Unit System

SUNNYVALE, Calif., July 18, 2011 – SeaMicro™, the Silicon Valley pioneer of low power server technology, today announced the immediate availability of the world’s most energy efficient 64-bit x86 server: the SM10000-64HD™. SeaMicro has once again defined best in class by improving its own compute density record by 150 percent and increasing its own industry leading compute per-watt metric by 20 percent. The new SM10000-64HD replaces 60 traditional servers, four top of rack switches, four terminal servers and a load balancer while using one-fourth the power and taking one-sixth space—all without requiring any changes to software.

I have been talking to a couple of people who have hands on experience with the Seamicro boxes.  One way to think about the Seamicro box is as if it is different type of mainframe.

Mainframes are designed to handle very high volume input and output (I/O) and emphasize throughput computing. Since the mid-1960s, mainframe designs have included several subsidiary computers (called channels or peripheral processors) which manage the I/O devices, leaving the CPU free to deal only with high-speed memory. It is common in mainframe shops to deal with massive databases and files. Gigabyte to terabyte-size record files are not unusual.^[5] Compared to a typical PC, mainframes commonly have hundreds to thousands of times as much data storage online, and can access it much faster.^{[citation needed]} Other server families also offload I/O processing and emphasize throughput computing.

Thanks the processor and high volume server wars, the data center in dominated by Intel Xeon based servers typically running in 2 processor configuration.  Few 4, 8, 16 processor configurations are sold.  For Big Iron for Amazon.com and eBay type of loads there are the Sun Servers for big Oracle databases, but Sun no longer has the presence in data centers it used to.

SPARC Enterprise M9000 Server

Oracle's most scalable mission-critical server for the largest, most demanding workloads

Designed for mission-critical environments, Oracle's SPARC Enterprise M9000 server delivers massive scalability, with up to 64 processors and 256 cores for the most demanding virtualization, consolidation, and multi-hosting deployments.

Now the problem with a mainframe metaphor is mainframes are considered dying, but look how IBM has extended the life of the mainframe.

The safe thing to do for a Web 2.0 company is to continue down the path of low cost dual processor servers, networked with top of rack gigabit switches.

Even though some look at Seamicro in terms of the Intel Atom processor, I pay more attention to how Seamicro is solving the IO and networking issues that are typical bottlenecks for throughput.  Consider this job posting at Seamicro.

SeaMicro is looking for an experienced Senior Hardware Design Engineer to architect and implement a flexible and scalable networking solution for its next generation data center products. This is an excellent opportunity for high-energy candidates who can take a complex networking solution from conception, through execution, to first customer shipment.

Qualifications:

• 10+ years of experience in high-performance/high-bandwidth micro-architecture
• 10+ years of experience in Verilog RTL development, with some experience in design/development of networking chips
• Experience with designs based on network processors, 10G interfaces, and DDR memory controllers desirable
• Solid understanding of L2 Ethernet switching protocols including VLAN, Broadcast/Multicast, and LACP is a plus
• Working knowledge of IPv4, IPv6, ACLs, and QOS

What’ll win for the hardware of the future? Lowest TCO, not efficiency or performance

GigaOm has a post on the Structure session with Calxeda, SeaMicro, and Tilera.

Efficiency vs. performance: What’ll win for the hardware of the future?

By Ryan Lawler Jun. 23, 2011, 6:09pm PT 2 Comments

At GigaOM’s Structure conference, hardware executives from Calxeda, SeaMicro, AMD and Tilera battled it out over how the hardware infrastructure of the future will take shape. For the most part, the debate came down to the need for highly efficient hardware that used up a fraction of the space and a fraction of the power of legacy hardware solutions, versus the desire for more powerful options from existing hardware makers.

I started blogging on the idea of the little green server with an Intel Atom 3 years ago.

Little Green Server Ideas Starting

TUESDAY, JULY 29, 2008 AT 12:48AM

ExtremeTech has an article building an Intel Atom based PC.

Intel's Atom has generated a lot of attention. Some of that attention has been positive: Intel building an x86 CPU whose primary design goal is very low power usage while maintaining good performance. On the other hand, Atom has been criticized for given up some key performance features, such as speculative, out-of-order execution.

What the author misses in making this an efficiency vs. performance battle is with many software services at large scale the cost of providing the service is the issue.  People want performance and efficiency, but want defines success is the cost of the service.

The next year will be when we see these servers hitting the market in mass.  Whoever is providing the lowest cost for performance per watt is most likely going to be winning the most amount of business.

I would watch the guys at Calxeda with ARM processors.