Google, IBM, Mellanox, Nvidia, Tyan form OpenPower Consortium | Hacker News Reader

Google, IBM, Mellanox, Nvidia, Tyan form OpenPower Consortium

ibm.com

79 pointspdknsk41 comments

I have a question. Why did a moderator lowercase OpenPOWER? It makes a difference.

To the mod:

POWER is not a typo, and it is not a emphasis, this is the trademark from IBM, the thing is really named POWER in all caps.

IMHO, that doesn't mean it's not shouting. At some point, gratuitous typography runs the risk of people refusing to respect it. For example, that company that seems to expect people to spell its name with a trailing exclamation point.

I have a bit more sympathy for IBM since they weren't exactly going for viral adoption and (seriously) their products have a long tradition of being lowercase-challenged.

>"The name is an acronym for Performance Optimization With Enhanced RISC."

http://en.wikipedia.org/wiki/IBM_POWER_Instruction_Set_Archi...

Right, not contrived at all.

It will be interesting to see if the New York Times refers to it as "I.B.M.'s P.O.W.E.R. technology."

A.F.A.I.K., N.Y.T.'s quaint style guide advises periods for initialisms (which are pronounced letter-by-letter) but no periods for acronyms (which are pronounced like words).

We also write Patriot Act not PATRIOT Act.

May I suggest: "Patriot" Act

Shouldn't we call out particularly egregious bits of language engineering when we see them?

"IMHO, that doesn't mean it's not shouting. "

So why did you write IBM, instead of Ibm? Or ARM, instead of Arm?

Because I find meaningless three letter acronyms much more palatable than five letter contrived ones.

The contrivedness of an acronym increases exponentially with its length.

EDIT: OK, I give in. I had too much fun hanging out with the password cracking guys at the awesome Passwords 13 conference last week.

Have you been anywhere near technology for the past 20 years? POWER has been its name forever.

Indeed. It was exactly 20 years ago, almost to the day, that I quit working for that IBM reseller in RTP, NC and took a career detour into lesser things.

IBM is furloughing US hardware employees this month: http://www.forbes.com/sites/alexkonrad/2013/08/05/report-ibm...

I find this move greatly entertaining. Its like a Hollywood remake, "AIM Alliance II"[1] Many of the things that make ARM successful IBM has dismissed in the past as pandering to special interests. And, like Sun with its failed attempt to make SPARC a thing, IBM has shown it can be a bit schizophrenic when it comes to sharing enough that folks can compete. ARM, not having a fab, and not having any products for sale, has very little to discourage it from agreeing to things. IBM on the other hand has other interests that are often in tension with the goal of making a processor architecture ubiquitous.

Glad they are trying again, not sure if they will be any more successful this time, but will be pleasantly surprised if we get a credible third horse in the 'mass market instruction set architecture' race.

[1] http://en.wikipedia.org/wiki/AIM_alliance

Anyone else here remember the CHRP/PReP days? Good times.

http://en.wikipedia.org/wiki/Common_Hardware_Reference_Platf...

http://en.wikipedia.org/wiki/PReP

Yes, there is a Back To The Future aspect. That said, cpus are just cheaper and easier these days. POWER, as a core, might break out.

Industry consortiums are always iffy things, but here's one thing I don't think I've seen since Lisp Machine days that's potentially very interesting:

"The consortium will offer open-source POWER firmware, the software that controls basic chip functions."

Open source firmware is a thing; check out U-Boot and Coreboot. I hear Google loves Coreboot.

Of course, but those aren't CPU firmware. With MIT and LMI's Lisp Machines (and probably the Symbolics LM-2 repackaging of MIT's, don't know about their own architectures), depending on how much microcode SRAM you had you could add your own microcoded instructions to the machine. There was even a compiler to make it easier, although I don't know how widely that was used.

ADDED: the elReg article in this item, https://news.ycombinator.com/item?id=6168144, based on talking to an IBM guy, implies it is that level of BIOS/chipset firmware that's being talking about. Ah well.

It sounds like you're going to be disappointed. Firmware != microcode, and we didn't say we're opening the microcode.

We? Are you involved in the OpenPOWER project?

Not personally.

"OpenPOWER is open to any firm that wants to innovate on the POWER platform and participate in an open, collaborative effort."

Translation: There is no way that this hardware is going to end up being affordable for small time builders, hackers, and enthusiasts. Unless someone screws up and accidentally ships something like a Playstation 3 rev 1 using it.

I'm only cynical because I'm jealous.

Too little too late? I dont get it. Intel has been iterating x86 at insane speed. In the high performance side Intel has finally caught on and eclipse Power. While Intel x86 has been winning the performance and performance / power in certain wattage scenario, they still dont have something to show for in Low Power segment. Although the next gen Intel Atom should be on par with the best in ARM, pricing would put Intel at disadvantage.

And that means in Low Power segment ARM realm supreme. Not likely to be dethrone in the next coming decade. As they have solid roadmap and pretty much the plan for next few years are all set and done.

So where does it left for Power? Coming from IBM they are very likely to be expensive. IBM has never been known to iterate as fast as Intel. Let alone ARM which now offers all sort of ARM chips for everything except High performance in their portfolio.

I just dont see how this could change anything.

For those that do not know, IBM processors are used in high-performance computing applications. Some of the world's fastest supercomputers are based on POWER [1]. These supercomputers can have hundreds of thousands or even millions of cores for massively parallel computing. We've seen recently from some of these systems that the POWER processors are much more effective than x86 processors. They consume lower power and generate less heat. If combined with Nvidia GPU's for CUDA and Mellanox's commodity InifiniBand interconnect hardware, we could have affordable clusters for heterogenous high-performance computing.

In the past, the supercomputers/clusters sold by IBM do not include hardware from these other companies. I think this partnership with Tyan (major manufacturer of current x86 chipsets) will also allow 3rd-party vendors to build systems with IBM POWER processors.

In other words, POWER is not in the same market as x86 or ARM. It does have its own niche in the market.

[1] http://www.top500.org/lists/2013/06/

We've seen recently from some of these systems that the POWER processors are much more effective than x86 processors. They consume lower power and generate less heat.

In other words, POWER is not in the same market as x86 or ARM. It does have its own niche in the market.

Look at that Top500 list again and compare #1 (x86) and #3 (POWER). #1 has about double the number of cores, performance and power usage. I'd say the high-end Xeons are comparable to POWER in all aspects.

The computer in the #1 spot was recently announced in China and sports Intel's Xeon Phi coprocessors, which include many cores per unit. The system in the #3 spot uses all cpu's and is completely homogenous. I worked in HPC at LLNL last year and saw the POWER processors. They are very small units and don't draw nearly as much power as Intel Xeon series.

Here's a picture of the Sequoia machine: http://imgur.com/6UrZndM

Each of those cards is a single node in the cluster.

Great perspective. Any more to share on your view how will POWER changes, i still dont see it goes anywhere apart from HPC. And being solely used in HPC just isn't a variable long term strategy.

Another worthwhile list to look at is the Green500 [1], which ranks machines based on MFLOPS/Watt. With the exception of a few GPU and Xeon Phi heavy machines at the top, the high end is dominated by IBM's Blue Gene machines.

[1] http://www.green500.org/lists/green201306&green500from=1&gre...

If something cool comes out of this I will be very happy.

I always considered POWER much more cool than x86, and was pretty sad when Apple abandoned it.

Also I was happy the current gen consoles used it, but I think Sony move to make it bizarrely complex to the point that noone could code for it, kinda stupid.

What is the status of the old T1/T2 SPARC chips? Those were similarly open-sourced, yes?

The RTL source was released under the GPLv2, GPL and hardware is a bit iffy since synthesis isn't the same as compilation, but depending on who owns the IP (in this case Oracle) they could make the case you need to open source your RTL source as well. I personally wouldn't invest the high hundreds of thousands to millions of dollars needed to produce a chip without getting a license agreement first with the ambiguity of the license, and which point I would probably consider the POWER architectures (like many FPGAs vendors used to) or ARM. In the case of pushing it to FPGAs, the OpenSPARC would use a ton of resources to the point where I'd rather have a dedicated off chip processor, or a FPGA with a built in processor.

I know people have put the OpenSPARC code on FPGAs, but I don't know of any use in industry in FPGAs or ASICS. I'd be interested if anybody knows any.

It's a shame too... so many beautiful threads.

Relevant, for anybody thinking of jumping aboard that sinking ship: http://www.forbes.com/sites/alexkonrad/2013/08/05/report-ibm....

Hmmm. Google having significant influence on the lowest level of my hardware...

You mean the bits that are already made in china?

any chance to have a parallella sort of SOC, a cheap HPC contender out of this consortiums?

imagine a PowerPC HPC SOC for less than $99 :)

POWER != PowerPC, although they are related.

There are/were plenty of PowerPC-based microcontrollers. Freescale, among other vendors, had a few families. It's all ARM now though as far as I know.

PPC is still incredibly common in industrial embedded systems.

So should I not throwout my powerbook G4?

Probably a good idea to deprecate your POWER infrastructure. This can only mean bad things for the customer support. Getting your APIs twisted in this so-called "open" software on proprietary, niche hardware mess with no single POC will mean you're dependent on them for a long time to come and your software will only become more and more complex and integrated as time goes on, until nobody on earth can untangle it.

Reply on news.ycombinator.com