bit-tech.net

ARM server pioneer Calxeda all-but folds

ARM server pioneer Calxeda all-but folds

Calxeda's server-centric ARM SoCs might be clever, but a delay in the production of 64-bit versions could prove fatal for the company.

The case for ARM processors in the data centre just got a little more difficult to push with the news that industry pioneer Calxeda has all-but folded amid cashflow troubles.

For the past few years, numerous industry giants have been investigating using low-power system-on-chip (SoC) chips based around the ARM architecture in servers - particularly for supercomputing and cloud computing roles, where raw compute performance from the CPU is less important than the ability to run as many threads as possible in as small a power envelope as can be achieved. Companies like Facebook and Amazon have been trialling the technology themselves, while recent rumours suggest that Google is going a step further and looking to design and build its own custom processors for its server farms using ARM's latest architecture.

The path to putting ARM in the data centre hasn't been smooth, however. Initially, concerns were raised regarding the architecture's 32-bit exclusivity - especially for cloud computing, where massive quantities of RAM is a common sight. ARM addressed this with the launch of a 64-bit architecture variant, but for some it appears to have come too late.

Calxeda, a start-up formerly known as Smooth-Stone and which looked to dominate the many-core microserver market with its ARM-based products, is one of those companies. 'Carrying the load of industry pioneer has exceeded our ability to continue to operate as we had envisioned,' claimed chief executive Barry Evans - a former employee of both Intel and ARM licensee Marvell - in a statement to the WSJ which confirmed a rumoured 'restructure' that will see the vast majority of its 130 staff laid off at Christmas. 'We just literally ran out of cash.'

The news comes as things looked to be on the up for Calxeda: its designs were chosen by Hewlett Packard for its Project Moonshot servers, but a last-minute shift in the company's plans saw Calxeda's ARM designs replaced with Intel Atom chips. A delay in the production of 64-bit processors, too, is believed to have hurt Calxeda badly.

In a statement to press, the company has promised that this is not the end: 'Calxeda has begun a restructuring process,' it claimed of the move to lay off the overwhelming majority of its staff at Christmas. 'During this process, we remain committed to our customers' success with ECX-2000 projects that are now underway.'

17 Comments

Discuss in the forums Reply
rollo 20th December 2013, 10:36 Quote
Surprised to say the least if demand is there then surely they would be raking it in. Maybe the demand is just not there as people think it might be.
Gareth Halfacree 20th December 2013, 10:40 Quote
Quote:
Originally Posted by rollo
Surprised to say the least if demand is there then surely they would be raking it in. Maybe the demand is just not there as people think it might be.
It appears to be the Osborne Effect in action: Calxeda has been talking up its 64-bit products, due next year - so nobody's buying its 32-bit products, instead choosing to wait until the 64-bit chips are out. The result: Calxeda has no cash to actually make the 64-bit chips happen. Bye-bye, company.

It's named for Adam Osborne, who killed sales of the Osborne 1 portable computer by announcing a pair of successors that would blow it out of the water - sadly, significantly before either were ready to launch. Sales of the Osborne 1 dropped through the floor as buyers waited for the promised successors, then Osborne ran out of cash to actually build said successors and the company folded.

What's that saying about those who do not know history?
Guinevere 20th December 2013, 11:30 Quote
Quote:
a last-minute shift in the company's plans saw Calxeda's ARM designs replaced with Intel Atom chips

Who else thinks Intel gave HP a price and support package that was way too good to refuse? Suspicious ol' me thinks Intel would have known that Calxeda desperately needed this sale and would have done anything to help them fold.

I wonder if money changed hands between Intel and HP... how much was it? And in which direction?
Ergath 20th December 2013, 12:57 Quote
Watch your homophones, Gareth "a common site" X-o

Interesting article, and even more interesting comment about the Osborne Effect.
Gareth Halfacree 20th December 2013, 13:02 Quote
Quote:
Originally Posted by Ergath
Watch your homophones, Gareth "a common site" X-o
A slanderous accusation; some of my best friends are homophones! I mean, err... (Fixed, ta!)
Margo Baggins 20th December 2013, 13:03 Quote
Crikey hadn't seen that one coming, had been reading all the hype about the 64-bit arm architecture.
jb0 20th December 2013, 14:32 Quote
Quote:
Originally Posted by Guinevere
Who else thinks Intel gave HP a price and support package that was way too good to refuse? Suspicious ol' me thinks Intel would have known that Calxeda desperately needed this sale and would have done anything to help them fold.

I wonder if money changed hands between Intel and HP... how much was it? And in which direction?
Given Intel's track record, I would say it's a pretty safe bet they greased the rails. HP probably got their Atoms for pennies.
I mean, they've been pretty cut-throat just trying to keep competing x86 processors marginalized, going back to the 486 clones. Hell, they quit using model numbers for the Pentium just so they could trademark the name.

ARM isn't even COMPATIBLE with anything they make. If they let it catch on in this market, they'll have an uphill fight to reclaim it.
They will do anything and everything in their power to keep ARM out of the server market... right until the very second they have their own ARM parts to sell, when it will suddenly be the next big innovation from the industry leader.
azazel1024 20th December 2013, 15:02 Quote
Its possible. It could also be that HP decided that the Atoms were good enough performance per watt and over all tdp combined with allowing x86 code execution and higher RAM amounts.

At least from what I can see, Silvermont arch is better than Cortex A15 in most cases as it stands currently in performance per watt and A57 is still a bit off from seeing a server room.


Price is certainly a consideration, but in the general server market, the CPU is a relatively small piece of the overall price, especially in machines that run 24x7. A z3770 is $37 in 1,000 unit lots according to Intel ARK, I'd assume any server varient probably isn't going to run significantly more than that. Maybe $50 a pop.

8GB of ECC RDIMMs are going to cost more than the CPU is for many of these machines (or at least possibly/likely).

Its also possible that the overall architecture of the server was easier/cheaper to construct with the other things that Atom was bringing to the table, or maybe not.

Intel could have subsidized the heck out of the price to cut out ARM. I wouldn't put it past Intel, but that doesn't mean that is what actually happened. The new silvermont Atoms are actually very good, in a lot of ways better than almost anything ARM designed on the actual market today.
faugusztin 20th December 2013, 18:06 Quote
To be honest - just look at how good Bay Trail Atoms actually are. 8" tablets with them are passive and i didn't even catch the glimpse of any heat yet on my Miix 2. 2W typical power consumption, 4W TDP for a quadcore x86 CPU - ARM will have hard task fighting against that. TDP difference is too low, performance is similar or better, Atom has x86 compatibility...
rollo 22nd December 2013, 11:34 Quote
The last bit is the big one that people refuse to think is the + point. But for most big companies its a huge +.

Arm servers requires someone to actually buy them to be successful. Google is gonna make its own is the rumours theres 1 huge demand just gone, Same with Facebook. They are 2 biggest potential customers and they are both thinking of making there own chips.

Does not really leave alot of target areas.
fluxtatic 24th December 2013, 10:35 Quote
Hmm...Google wants ARM servers, Calxeda is broke...and Google's pile of money is starting to get too big for Sergey and Larry to sleep on comfortably. What to do, what to do?
ch424 24th December 2013, 11:55 Quote
Quote:
Originally Posted by fluxtatic
Hmm...Google wants ARM servers, Calxeda is broke...and Google's pile of money is starting to get too big for Sergey and Larry to sleep on comfortably. What to do, what to do?

Facebook might get first dibs :p http://www.theregister.co.uk/2013/10/18/facebook_frankovsky_calxeda_board/
Quote:
Originally Posted by rollo
Google is gonna make its own is the rumours theres 1 huge demand just gone, Same with Facebook. They are 2 biggest potential customers and they are both thinking of making there own chips.

Google and Facebook isn't a "huge demand". ARM sells ~10B CPUs a year, so even 10M servers is low volume by ARM's standards.

ARM getting into servers is also about ruining Intel's margins, not entirely about selling more chips. If Intel introduce subsidized Atoms and reduce the margins on their Xeons to compete with cheap-as-chips ARM parts, then ARM still wins even if they don't sell a single socket.
rollo 25th December 2013, 20:23 Quote
ARM all combined might sell 10bil I have my douts personally, But the 2 - 3 major companies ( apple , Samsung, Qualcomm) who produce and use them in smartphone tablets and servers between them are not even close to 1bil chips shipped let alone sold.

If ARM was licencing chips for 10bil as it would it would also be a lot richer than it currently is.

Licence estimates put it at around 700million chips sold btw.
Harlequin 25th December 2013, 21:08 Quote
Quote:
Originally Posted by rollo
ARM all combined might sell 10bil I have my douts personally, But the 2 - 3 major companies ( apple , Samsung, Qualcomm) who produce and use them in smartphone tablets and servers between them are not even close to 1bil chips shipped let alone sold.

If ARM was licencing chips for 10bil as it would it would also be a lot richer than it currently is.

Licence estimates put it at around 700million chips sold btw.

http://www.arm.com/products/processors/index.php


would seem ARM somewhat disagree with you by a about 29 BILLION of units sold
ch424 25th December 2013, 21:15 Quote
Where did you get that 700m number? I got 10B from a press release - "Q3 revenue came from the sales of about 2.5 billion ARM-based chips" and did a very simple 2.5B * 4 = 10B
rollo 26th December 2013, 18:24 Quote
You kinda just showed why between you arm servers will never be profitable. 30bil chips sold or licenced net £188million for 2012. In sept 2013 they only received 118mil total revenue for the quater. (all from there published financials)

That's worse revenue per chip than I have ever seen.
Harlequin 26th December 2013, 19:55 Quote
Not bad for a company that doesnt actually make anything
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums