ARM server pioneer Calxeda plans restructuring after running out of cash | Hacker News Reader

ARM server pioneer Calxeda plans restructuring after running out of cash

gigaom.com

25 pointsjustincormack14 comments

Its now looking more like Google and Facebook and the big players are interested in building there own ARM stuff, and setting yourself up as a vendor is perhaps the wrong model.

But mainly, its been impossible to buy an ARM server from any of these people, or rent time on way for validation.

I would like a 1U ARM server for home and perhaps a few for my colo if they were 64-bit and price competitive to the Xeon E3... even if significantly slower. Unfortunately, they are basically unobtanium and an E3 Xeon build can be done for ~$1k.

Do you really need 64 bit? Could you manage with more processes and more servers?

some performance comparisons by Anand:

http://www.anandtech.com/show/6757/calxedas-arm-server-teste...

while not bad, it doesn't seem to have enough advantage to beat Xeon in real market (especially while being more expensive than Xeon systems). Like with Sun's Niagara processors, being better only in a narrow niche may be not enough for a financial success in pretty general server market.

I don't have data but I have a feeling they made the mistake of acting like a corporation while they were a startup.

When you do that, you spend $100M.

Raising $100M is OK but you must save it and wait until you are sure the market is picking up.

You should act very nimble. Possibly a very thin management layer, %90 engineers. Near zero marketing and attendance to marketing events. Just silent focus on IP.

Like I said I have no data and probably some substantial amount had to be spent in HW manufacturing, but it feels like they overspent the money, and acted without being sure the time has come for the market.

For $55M they designed two fairly sophisticated processors and were working on the third generation. It's possible they could have been leaner, but I don't know by how much.

"designed two processors". No. We had working A0 production runs of two chips, with a the A0 of the third generation in the post when we closed. Leaner? Not really. Sales team was three people more the executives, everyone else was engineering. Meaner? Who on earth pulls off two working A0 runs?

They were too early, and with that didn't target a viable niche before going for the big things when 64 bit came along. Something like a 32-bit ARM tied to hardware that accelerates running memcached could have been viable for quite some time, for example. nVidia probably have a long term vision not dissimilar to that, with the ARM being a mere general purpose front end for an attached GPU, and looking at the new game consoles that appears to be the way many things are headed there too.

This will predictably cause the Intel brigade to cheer up, but it will be temporary. In the time since Calxeda started ARM has just consolidated and grown, especially into home routers where MIPS is looking increasingly like it's gone for good. Ultimately when ARM do finally crack the data centre the impact is going to be profit margins for almost all chip players going down very fast. Going into this market without a USP that isn't simply cheap general purpose SoC is destined to end in trouble. If you look at the players that survived on mobile they aren't just integration outfits for ARM/Imagination, but actually add value through other custom pieces too, either GPUs (say nVidia) or re-implementing the core themselves, like Qualcomm.

This is at the point where I really hope Imagination do something good with MIPS and it isn't consigned to history as ARM being too dominant could prove to be a real mess.

The allthingsd article submitted here: https://news.ycombinator.com/item?id=6938138 has more info, including verification from company execs.

They were way too early to the market. Maybe it would've been ok to be there since Cortex A15, but being there at the Cortex A9 generation always seemed premature - at least for big customers like HP, Facebook, and so on. Maybe they should've tried selling them to the same people who like the idea of a much weaker Raspberry Pi server. Instead they tried to sell only to a few big guys, and none of them were really that interested.

I'm not too worried about them, though. They'll probably get acquired by someone like Samsung, or Qualcomm, or Nvidia.

Writing this as a now ex-employee, I'm sure that our IP will live on. We had some really neat stuff in the pipeline, and I look forwards to seeing it hit the big time in the next few years. Whether Calxeda proper will be acquired or whether the IP will be sold piecemeal only time will tell. Personally I'm not convinced that the company in it's "entirety" will be bough out. At this point it's the IP not the assets or people that's valuable.

I also totally agree with your assessment that we had a cool idea too early. That was in fact what Barry said at today's closing down meeting. ARM in the datacenter will happen, but that we didn't have a 64 bit chip really limited our hardware offerings to say the least.

Sorry for your lost job. In all honesty though, I don't think ARM in the data centre will ever be anything more than a niche product. Considering Intel is 2 manufacturing processes ahead, the microarchitecture and any potential advantages there won't even really come into play, and Intel's not exactly any slouch either. If 64bit ARM came out a year ago, there might have been a fighting chance, but any chances of a major uptake pretty much died with the release of Avoton. That may very well have been why the next round of funding fell through. Intel's next two generations, Broadwell and Skylake, are both heavily focused on reducing power consumption, and achieving parity with their peformance/watt would be enough of a challenge, let alone creating a significant enough advantage to take away a real amount of market share.

Best of luck on your next venture.

I thought AMD was entering the ARM space.

Besides, someone will probably end up buying the IP from Calxeda if they do shutdown. Maybe HP?

AMD may be the last ARM server processor vendor left standing (I don't consider Marvell a serious contender and AMCC is iffy). Perhaps Calxeda did all the work to create the market which AMD will be able to come in and take.

Reply on news.ycombinator.com