Tilera stuck its low-power Linux-running 64-core processor to Intel and AMD Tuesday in a head-on challenge to the x86′s infrastructure dominance, particularly in Web 2.0 and cloud accounts, where Tilera’s many-core design is supposed to excel at executing their millions of small parallel tasks simultaneously.

Way better, it says, than the high-power legacy processors with their minimal performance improvements across generations.

Tilera’s system on a chip (SoC) is breaking out of the embedded appliance ghetto where it’s been living and going into the server business with what is supposed to be the most power-efficient and highest compute density system ever, the building blocks for squeezing 10,000 cores into an 8kW rack.

Quanta Computer, the big ODM and a recent investor in Tilera, has built a 2U box codenamed S2Q out of eight Tilera TilePro64 chips reportedly good for up to 1.3 trillion operations a second and costing only 400 Watts to run.

It built the S2Q on spec on its own nickel – an unusual thing for an ODM to do – because those 512 general-purpose cores in that little bitty space are supposed to be able to replace eight 2kW-burning high-end Xeon 5000 class dual-socket servers – and Quanta expects to make a killing on thing once it’s released in limited quantities in September followed by general available in Q4.

It won’t be priced until it starts dribbling out.

Some of the biggest OEMs, the x86′s nearest and dearest, are reportedly going to be buying the widgetry off of Quanta and those that don’t are supposedly developing their own Tilera machines in answer to the cloud’s prayer for rational power budgets and modest space demands without sacrificing performance.

SGI, for one, says it’s going to be peddling a Tilera box. (Tilera chips are rumored to be a variant of the Mips chip once associated with SGI but that’s not true.)

Tilera’s VP of marketing Troy Bailey says that if nothing else vendors are getting no real differentiation out of the legacy chips and they’re “sick of fighting over the last penny.”

Tilera quotes Facebook’s VP of technical operation Jonathan Heiliger saying at GigaOm’s Structure Conference last year that “To build servers for companies like Facebook and Amazon and other people who are operating fairly homogeneous applications, the servers have to be cheap, and they have to be super power-efficient. The latest generations of server processors from Intel and AMD don’t deliver the performance.”

Quanta’s widget was reportedly designed in collaboration with cloud data center providers, end customers and software partners and is targeted at large-scale data centers running high-performance web, database, hosting and finance applications.

Figure markets with highly parallel applications like web serving, in-memory cache, data mining, financial and scientific analytics and government surveillance.

Quanta’s promising to slash users’ TCO and deliver an 80% OPEX savings of over $2.5 million over three year in power they won’t have to pay for to run 1,000 server or $75 million across a whole data center.

Tilera claims to do on a chip what the newly up-from-under-the-radar SeaMicro start-up does on a board.

SeaMicro is aiming for the same space with the same arguments as Tilera and a Xeon-displacing system chockablock with 512 low-power Atom processors, 2,048 CPUs to a standard rack, a design that cuts out 90% of the power-hungry widgetry around the CPU – the switches, terminals servers and load balancing devices in the classic server that are actually responsible for two-thirds of the power it consumes.

Since SeaMicro is supposed to be able to accommodate any processor on its credit card-size motherboards, it’s conceivable it could someday use Tilera’s chips, which are supposed to good for 100 cores on a 40nm process next year and 200 cores on a 28nm process in 2013.

That should make for 20,000 cores to a rack next year and 40,000 cores to a rack by 2013. (Samples of the chips should be circulating by 4Q10 and 2Q11 respectively.)

But first things first. Quanta’s S2Q, which is reportedly undergoing performance tests as we speak, provides up to 16 x 10 GbE interfaces and 16 x 1 Gb interfaces without adding the power or the cost of additional chipsets and networking cards.

It’s got 176 Gbps of I/O bandwidth, up to 64 DIMM slots, and up to 24 front-mounted 2.5-inch hot-plug SAS, SATA or solid state hard drives along with IPMI 2.0 dedicated management ports and dedicated console ports.

Each server node consumes 35W-50W maximum. Its hot-pluggable power supplies are supposed to be 90% efficient or better and its shared fans and power supplies conserve space and power.

The widgetry runs SMP Linux 2.6.26-36 and supports virtualization. Tilera’s iMesh technology enables it to integrate many cores with coherent caches to deliver scalable performance.

Figure a standard software stack, stuff like Apache, MySQL, Hadoop and MemcacheD supporting PHP, Python, Ruby, Perl and Java programming with tools like the Eclipse IDE and GNU GCC and C/C++ compilers.

Tilera says a tier one server OEM – and there are only a few of those – running MemcacheD on its own Tilera and Nehalem servers got 10% better performance out of its Tilera box. SPEC benchmarks aren’t any good at measuring the performance of these newfangled boxes so prototypes have apparently been running real loads in web centers for months.

Tilera, which has MIT antecedents, kicked off in late 2004 and has two processor generations in production. It closed a $25 million C round from Quanta, NTT and Broadcom in early March, all new backers, bringing total investment to $64 million. It claims 40 design-wins, soon to be 50, from a hundred paying customers. It’s not scheduled to be cash flow positive for another year. Revenues, it says, have been doubling every two years.

IBM filed a breach-of-contract/misappropriation of trade secrets suit in New York Supreme Court last week seeking to hold Joanne Olsen to her non-compete and stop her from going to Oracle for a year.

It’s gotten a temporary restraining order according to InformationWeek.

Olsen, with IBM for 31 years, was a general manager in IBM’s services unit and was hired by Oracle as senior VP of on-demand services, the SaaS versions of its software, reporting to none other than Larry Ellison, who’s aiming to do as much damage as he can to IBM.

IBM alleges that she knows too much about IBM, its operations, its growth strategies and its potential acquisitions to fall into Oracle’s hands.

IBM has gone to court a couple of times in the last couple years waving its non-competes around.

Oracle’s on-demand revenue in its fiscal fourth quarter just reported Thursday was $295 million, up from $204 million a year ago.

x86 Cloud Franchise Under Attack
Neon To Complain about IBM to the EC
IBM Demands Cloud Cover; Sues To Stop Exec from Working for Oracle
Oracle Surviving Sun Just Fine, Thank You
Azul Zings its Java Hardware – Poof, It’s Software
Next Stop Graphene Valley?
IBM Accused of Racketeering
EC2 Creators Strike Out on Their Own
Red Hat to VMware: So Who’s Cisco’s Pet Now
Intel’s Negotiating a Consent Decree with the FTC
Red Hat’s Server & Desktop Virtualization Move in Together
Apple Sells Three Million iPads in 80 Days
Red Hat Compares Itself to Microsoft
Should We Be Writing Epitaphs for the Mac?
Now for the Fun Part of the Cloud: The Money
Eucalyptus & HP New Best Friends
Swiss Post Abandons US Market To Settle RPost Suit
Courier’s Not the Only Twin-Screen Tablet in Town
Dell Offers First Guidance Since 2006
Sun’s Ex-Cloud CTO Finds Home at Cisco
Dell Stakes Growth on Data Centers
Next Up: 10 Cores
Larry Hires David Boies To Snag a Billion Off of SAP
Reorg at CA?
Google’s Tit’s in the Wringer
Canonical: OIN First Associate Member
Cloud Services Good for $149B in 2014: Gartner
Ex-CA Boss Joins Silver Lake
Russia Trolls for Investment for Its Own Silicon Valley

Starting a server company these days is like setting out to cross the cold, dark, unforgiving Atlantic in a dinky Viking long boat. It’s not an adventure for the faint of heart.

The bards only tell tales about those who made landfall, not those who got swamped and drowned. And the computer industry hasn’t produced many heroic sagas lately.

But that’s kinda SeaMicro’s point. It argues that there’s been no novelty in the volume server space in 20 years – other than blades – and that the tired old x86 box, optimized for the legacy workload, isn’t any good at what it’s being asked to do now – namely run the Internet.

It takes up way too much space and demands way too much power.

According to the Environmental Protection Agency the volume server is consuming more than 1% of all the electricity in the United States at a cost of more than $3 billion a year, and according to Google the thing costs more to run than to buy.

Power is an Internet company’s biggest opex item, claiming upwards of 30% of its opex spend. That’s why SeaMicro, a stealth start-up that broke cover Monday, is building scale-out servers around Intel’s puny Atom chip thinking that it will live to plunder the modern equivalent of Lindisfarne, establish the first Russian state and dicker with the emperor of Constantinople just like its Viking antecedents.

It claims that if you put enough of these Atom suckers together you get a 75% reduction in power and real estate, no change in software and way better output. Figure roughly three times the performance-per-watt of the classic x86 server CPU running web-tier applications.

That’s why SeaMicro got a $9.3 million grant from the US Department of Energy, which is a whopping lot of money in DOE terms, more than any other server company got, and why Khosla Ventures, the green VC fund started by Sun co-founder Vinod Khosla, put $25 million in the joint along with Draper Fisher and Crosslink Capital. They’re looking for – and hoping they’ve found – an alternative to the energy-draining volume server.

The SM10000 machine SeaMicro’s been working on for the last three years uses 512 1.6GHz Intel Atoms – 2,048 CPUs to a standard rack – a quarter of the space at a quarter of the power running the same off-the-shelf operating systems and applications as the best-in-class volume server.

SeaMicro’s box is supposed to be able handle the millions of small, simple, independent jobs involving search, social networking, web page views and e-mail checks with far more aplomb than the classic server, which was really invented to resolve a relatively small number of large, complex, basically very hard problems. The classic server is also idle, poorly utilized and inefficient most of the time because of the Internet’s bursty traffic patterns.

Through the wonders of CPU I/O virtualization, SeaMicro gets most of its patented power reduction from cutting out 90% of the power-hungry widgetry around the CPU – the switches, terminals servers and load balancing devices in the classic server that are actually responsible for two-thirds of the power it consumes.

Once it got rid of that stuff the start-up was able to shrink the server’s typical pizza-box motherboard to the size of a credit card, leaving it with just the Atom CPU and chipset, its own all-important ASIC and 2GB DRAM, making possible eight servers in 5 x 11 inches.

Then it developed its own supercomputer-style interconnect fabric to link its 512 mini-motherboards into a low-power single-box cluster computer with 1.28 terabits-a-second throughput and created its own compute, storage, networking, server management and load balancing rack.

It says that once it wrung power cuts out of the system level, it could leverage the skimpy power demands of the Atom, the chip that up until now has been used in netbooks and – Intel hopes – some mobile phones.

The company also came up with Dynamic Compute Allocation Technology (DCAT), a combination of CPU management and load balancing so the machine can dynamically allocate workloads to specific CPUs on the basis of power-use metrics, create compute pools for any given application and ensure that energy use is optimal. The widgetry, which also promises cloud-optimized security, is fully redundant with hot serviceability.

The 10U SM10000 systems include 1 terabyte of DRAM; up to 64 SATA solid state or hard disk drives; and eight to 64 gigabit Ethernet uplinks or two to 16 10gigE uplinks. A base configuration will list for $139,000 and should be available in the US, Japan, Germany, France and Hong Kong to start on July 30.

SeaMicro CEO and co-founder Andrew Feldman, a networking veteran who was an entrepreneur-in-residence at Crosslink Capital, claims Intel, which didn’t think Atoms would be used this way, is adjusting to the notion that its precious Xeon chip might get displaced and consoling itself with the thought that it’ll sell a lot more Atoms at a lot better margin.

However, the SeaMicro design isn’t restricted to the Atom. The architecture is supposed to be able to support any CPU. It could as easily support ARM, Mips or the PowerPC, but Atom is reportedly the most efficient choice so far.

As it is now, Feldman said companies could consolidate 40 Xeon servers like a Dell R610 with one SM10000.

One dual-socket, quad-core Westmere box uses 250W-270W; the SeaMicro box with its 512 Atoms comes in under 2kW. Stacking four of the machines on top of each other in an 8kW/44U rack equals 2,048 single-core Atoms, which Feldman says is the highest CPU density-to-power ratio achieved so far.

With the CPU utilization the 57-man SeaMicro is promising, server farms of 1,000-4,000 machines could save somewhere between $1.1 million and $6.1 million in power costs and between $2.6 million and $21.9 million in power and space combined.

Feldman says the company is already backlogged 45 days and that “largest Internet brands in the world” are running with his machine. He won’t say which. He could be looking at breakeven in a year.

SeaMicro’s other two founders are its CTO, erstwhile AMD Fellow Gary Lauterbach, the chief architect of Sun’s UltraSparc III and IV, and its VP of product management and product marketing Anil Rao out of Juniper Networks and Force10.

Forrester Research thinks netbooks have only two years left before tablets pass them by and that by 2015 tablets will represent 23% of all US PC sales, some 20.4 million units, with notebooks good for 42%, desktops 18% and netbooks 17%.

“Consumers didn’t ask for tablets,” it said. “In fact, Forrester’s data shows that the top features consumers say they want in a PC are a complete mismatch with the features of the iPad. But Apple is successfully teaching consumers to want this new device.” The proverbial Steve Jobs reality distortion field apparently.

Forrester figures tablets should be good for close to 60 million units in 2015, up from 3.5 million give or take this year, it said, although with two million iPads sold in two months those projections look like only a shadow of what they might be.

Its projections have some obvious implications for Google and its Chrome-based netbook effort as well as its Android slates.

By the way, it looks like Dell is good for one of those Chrome widgets along with HP and Acer.

Forrester also thinks the consumer PC market in the US is getting bigger and in the next five years PC unit sales across all form factors will increase by 52% to close to a half-billion.

Well, maybe not desktops. Fewer of them will be sold in 2015 than in 2010, but in 2015, they’ll still be used by more American consumers than any other type of PC, it said. Figure 15.7 million units.

For its part, DisplaySearch, which also thinks netbooks will be overshadowed, figures iPad will have ~30% of the total netbook and slate sales of 9.7 million this quarter.

Gartner says 32 million netbooks were sold worldwide last year.

Green Start-up Rethinks the Volume Server
Meteoric Rise Predicted for Tablets
Neon Chips Away at IBM’s Mainframe Monopoly
Managed Runtimes Suck; Azul Organizing Rescue Party
FTC To Investigate Apple’s Business Practices
Let the Games Begin
Xenocode Morphs into Spoon
SEC May Slap Dell’s Hand 100 Million Times
IBM Buys Coremetrics, Adding to its Cloudware
Eucalyptus Private Cloud Software Now Supports Windows VMs
RainStor Revs its ‘Big Data’ Mojo
Norway Post To Pit its IT Arm Against IBM
Google To ‘Chromote’ Windows Apps
DOJ Sues Oracle
IBM To Buy Storwize: Report
AT&T Must Believe There is a God
Microsoft Does Deal with Opera
Microsoft Returns To Building Data Centers
Microsoft Arms for Cloud Fight

It’s an utter rout.

Linux is apparently saved whether it deserves to be or not.

SCO Thursday lost its bid to get the jury verdict awarding Novell the Unix copyrights overturned along with its bid to get the copyrights despite the jury decision. It also lost its right to sue IBM for copying Unix code into Linux.

Unless it decides to appeal, one of its lawyers said, it’s all over for SCO. And it’s not clear if SCO has the financial staying power to last through an appeal. The legal bill for another appeal is already paid; it’s SCO basic viability that’s in question. Of course, it doesn’t have to be an operating company to get an appellate decision.

Deciding what to do next ultimately rests with the bankruptcy trustee, the retired district court judge currently in charge of SCO, and he was out-of-pocket Thursday. SCO’s general counsel Ryan Tibbets said the company’s lead outside lawyer was also out-of-pocket and anyway “you can’t come up with a strategy in three hours.”

SCO has 30 days to make a move.

After a federal jury found in March that the Unix copyrights never transferred to the Santa Cruz Operation and so never transferred to its descendent SCO, SCO filed two post-trial motions with the Utah district court that heard the case. One – and it was a long shot – asked the judge who presided over the case to overturn the jury’s verdict as a matter of law. The second – a little less of a long shot – asked for a new trial on the grounds that it wasn’t clear Novell should have won.

The judge denied both motions. He simply found Novell’s rendering of events, its evidence and its witnesses – which he details – overwhelmingly more credible and dismissed out of hand SCO’s contention that the “verdict cannot be squared with the overwhelming evidence and the law.”

It was also agreed by SCO and Novell before the trial began that in the event SCO lost the judge would decide whether to award SCO the copyrights anyway because it should have gotten them under the muffled intent of the infamous amended Asset Purchase Agreement (APA).

SCO kinda had its hopes pinned on this move but the judge also denied this maneuver, which is called specific performance in legal circles, because SCO didn’t need them to exercise its limited rights and because the intent was absent- “Novell purposefully retained those copyrights,” he said.

More importantly, along with it he decided that Novell had every right to waive SCO’s contract claims against IBM so barring a miracle IBM will never stand trial.

It’s just what IBM had in mind when it told Novell back in 2003 that Novell had the discretion under the APA to do so although the judge finds that Novell acted “to protect its own interests and those of the open source community and…not…because of influence by IBM or any ill-will toward SCO.”

In his decision the judge wrote that “The court finds that Amendment No 2 was not intended to confirm that the Unix and UnixWare copyrights were transferred to SCO under the APA, as argued by SCO. Rather, the court finds that Novell made a conscious decision to retain the copyrights in the APA and that intent was reflected throughout the negotiating and drafting of Amendment No 2. The court finds that Amendment No 2 was only meant to confirm that SCO had the right to use the Unix technology. The court finds the testimony of Novell’s witnesses, especially Ms Amadia and Mr Tolonen, to be credible. The court finds SCO’s witnesses to be less credible for a number of reasons, including the fact that many were not directly involved in the negotiation and drafting of Amendment No 2. Additionally, as previously stated, many have a financial interest in this litigation.”

Meanwhile, Novell has filed a notice of appeal in its 2004 WordPerfect antitrust case against Microsoft after Microsoft got the thing dismissed. Any claims went along with the software to Caldera.

Looks Like Sculley’s Finally Got His Newton
Novell Wipes the Floor with SCO
VMware To OEM SLES
iPad Privacy Hacked
Meg & Carly Win California Primaries
HP Launches New Line of Web & Cloud Printers
More Heads To Roll at Sun
Terascala Brings in New CEO; Cuts Dell Deal
Amazon Lights Fire under Large-Scale Data Transfers
Ingram Micro Seeks Piece of Cloud Action
Altor Testing Cloud Security
Pentaho’s Open Source BI Goes On-Demand
Amazon CloudFront Gets Security
CA Puts Money in Medical SaaS Firm Chaired by John Sculley
Office Web Apps Inch Out
Microsoft Revs Azure Widgetry
Windows 7 SP1 Beta Due
Microsoft’s Communications Server 14 Set for Later This Year
Imagine Dell Private, Michael Has
Oracle Delays Sun Rebates
SAP Claims Interest in More Big Acquisitions
Apple Poaches Second Cisco Brand
Palm Leaking Talent
IBM Opens Research Center in Brazil
35 Million iPad Apps & Counting
Greenplum Does Europe
Bing Cash Backs Run Out
HTC Buys Abaxia
Gizmodo Barred from Apple Fest
High-Tech Chiefs Want Screws Tightened on China
HP Accused of Stealing Printer Cartridge Secrets
Apple Blocks Google Ads

Intel has started laying the groundwork for what it says will eventually be at least a 50-core x86 co-processor called Knights Corner based on a newfangled Many Integrated Core (MIC) architecture.

It says the widget, the first of a family of Knights, will create HPC platforms running at trillions of calculations a second without sacrificing the benefits of standard x86 processors, a hard-won lesson learned from its Itanium adventure.

It means to create supercomputers that run most of their workloads on Xeon chips but accelerate specific highly parallel applications complements of the MIC architecture.

The anticipated 500 gigaflops systems will be aimed at uses such as exploration, scientific research and financial simulation.

According to none other than DreamWorks Animation CEO Jeffrey Katzenberg speaking at the chatty D8 conference the other day, Intel also has workstation intentions for the thing and the reason he called the thing Larrabee despite the fact that Intel canceled the prospective GPU last week is because the thing is Larrabee redirected.

In the initial 32nm developer prototype version that’s floating around out there Knights is a 32-core Larrabee. The first commercial product will be 22nm 50-core device that derives not only from the canceled Larrabee but also from the research-y ultra-low-power 48-core Single-Chip Cloud Computer that Intel disclosed in December and TeraScale.

Larrabee, as Intel puts it, will “use Moore’s Law to scale to more than 50 processing cores on a single chip.” Each of the cores will reportedly run at 1.2GHz, have 8MB of shared coherent cache and four threads apiece for a total of 128 threads. It’ll also support 1GB-2GB of GDDR5 memory.

The company’s already got design and development kits codenamed Knights Ferry shipping to hand-picked developers, and beginning in the second half it means to start delivering MIC developer tools. The trick is that they will share a lot of in common with Xeon tools as well as common software algorithms and programming techniques. The same languages, compilers and libraries can be used. It eliminates the need for a dual-programming architecture.

Apparently the CERN openlab team was able to migrate a complex C++ parallel benchmark to the MIC software development platform in just a few days. Its CTO Sverre Jarp said that “the familiar hardware programming model allowed us to get the software running much faster than expected.”

Since the dingus is 22nm it can’t go commercial until next year.

Google, in its attempt to challenge Microsoft’s PC dominance, has moved up the market entrance of its cloud-based Chrome operating system from 2011 to late fall, less than six months from now according to reports out of the Computex show in Taiwan quoting Sundar Pichai, the Google VP in charge of the Chrome project.

He said Google expects the OS to be “widely adopted across the netbook sector” and “reach millions of users on day one.”

Reuters said Microsoft last Thursday compared Chrome to the early days of the PC when applications had to be written to specific boxes because netbook vendors are going tinker with the open source Chrome code.

Pichai denied the allegation claiming similarities in the base core would allow applications to run on all Chrome machines. He also told Reuters that “Chrome OS is one of the few future operating systems for which there are already millions of applications that work.”

Interfaces are also expected to vary.

Penetration of the Chrome browser on which the operating system is based hit 7.05% worldwide in May according to Net Applications. It was down a half-a-percent in the US to 4.51% from 4.97% in April. Google has claimed 70 million people are using the browser.

© 2012 LinuxGram Suffusion theme by Sayontan Sinha