Intel CEO Paul Otellini has a low opinion of ARM Holdings' efforts to crack the server market, an arena in which Chipzilla's processors are the dominant force. "It ain't gonna work," Otellini told his audience at Intel's Investor Meeting 2011 in Santa Clara, California, on Tuesday. When asked what ARM would have to do to become …
It's a friggin server, for heaven's sake. James' and Microsoft's argument that "but, but... it won't run legacy applications" is almost completely irrelevant in this context, unless you're talking about IIS or some other proprietary non-mainstream environment.
If the architecture performs well with a typical LAMP stack, that alone makes it suitable for quite a large percentage of server applications, and would by itself make ARM Holdings quite happy.
and who is going to write compilers.....
I bet the Intel made compiler is used /loads/ in Linux servers. Oh sorry, I remember, they use GCC....
Well, I guess they could funnel money at GCC so it totally rules that compiler that ARM uses for compiling Linux, what was that called again?
Large enough cake
"and who is going to write compilers....."
Both Intel's ICC and ARM's RealView-CT use the Edison Design Group C frontend.
ICC will compile the Linux kernel and warn about more issues than GCC compiler
RealView armcc/tcc will compile the Linux kernel and warn about more issues and produces smaller code than GCC compiler, even the CodeSourcery ARM GCC compiler. But the ARM EABI standard means there is more freedom of choice so you can obtain a compiler from another vendor for all or part of your server's code. For eaxmple, IAR have just released an ARM EABI compliant toolchain. By working with CodeSourcery, ARM may even avoid Intel's old Unix86 iBCS' fate, swept away by Linux ABI and LSB.
With ARM for servers you have not only have a choice of silicon vendor, but also of tool supplier.
It's all about the ecosystem.
"the software environment, you know?"
You mean like HP/UX, or VMS, or Tandem NonStop?
Oh, sorry, thought you'd mentioned something unrelated to x86, e.g. something IA64-related, but obviously you too think that mentioning IA64 would be a mistake...
ARM chips may not work...
...but the Bulldozer ones will.
An evil mu-hahahahaha laugh at Intel...
That bogus "legacy" argument
I always find it most amusing when (pro-)Intel float this argument that applications would have to be "rewritten" for ARM. Surely they have heard of C and other so-called "high-level languages" - you know, the ones that allow programs to be effortlessly ported across architectures, provided the necessary tools (compiler, interpreter, etc) are available?
Of course, if we're talking about a massively parallel architecture, then most probably a rewrite will be in order - but that would be due to legacy applications often not scaling well to multi-processed environments, and have nothing to do with the underlying instruction architecture.
I can only conclude Intel must be very scared of something, to so viciously turn their FUD barrage against ARM. I wonder what that could be?
Well he would say that wouldn't he!
Those were the first shots of a very big FUD war that is about to break out.
This town (called Servers) isn't big enough for both of is.
BB (aka Intel) getting worried.
What are they afraid of?
There have been a fair few anti ARM stories this week, with all sorts of FUD: the lies about backwards compatibility, the nonsense about a new server to run 'legacy' code (as already noted)...
...sounds to me like somebody is scared and is trying to discredit the platform. That might work for politicians and games consoles, but people around these parts are generally more clued up.
The case against MIPS et al. 20 years ago was different - there weren't the toolchains around to support software that ran well on multiple platforms. Things are very different now - gcc produces acceptable results and then there is CLANG and LLVM - and the application vendors happily support several flavours of unix and windows and customers are, by and large, happy with their choices. And 20 years ago Intel was the cheap upstart breaking into the mainframe world - Intel servers were significantly cheaper than equivalent Sun, HP, SG or IBM boxes. This is exactly where ARM is now with the advantage of having multiple manufacturers accelerating and cross-licensing the production processes. As for the hardware - ARM has championed the ability for manufacturers to choose whether they want particular sections such as encryption done in silicon or software.
Asbestos underpants must be on Otellini's Christmas list!
In other words
Pleeeeeeeeease don't sell your shares.
If Oracle and IBM were to support ARM on top-end servers, who is Intel to contradict them?
Nor do today's top dogs need to lead the way. They could just be following a trend driven by the smaller players who are working on ARM/Server today. Oh, and Intel's old double-timing partner from Redmond.
Having said that, there's no denying Intel have a real lead. For example, porting Apache Trafficserver to ARM comes up against the problem of emulating 64-bit atomics, which means you wouldn't want to use the combo in production. Not yet, anyway.
speaking of MIPS
The rumour mill is persistently saying that longsoong is using MIPS CPUs.
This could get interesting.
Re: speaking of MIPS
"The rumour mill is persistently saying that longsoong is using MIPS CPUs."
Rumour mill? It's established as fact that Loongson CPUs are MIPS derivatives:
You can even buy stuff like this using that architecture:
If everyone weren't whining about the possibility of running Windows and Office on everything from a wristwatch upwards (and then not bothering to do so), we'd see a lot more of this kind of thing.
Intel not shipping ARMs...?
What about all those ex-Infineon chips......?
Stating the obvious
> "It ain't gonna work," Otellini told his audience at Intel's Investor
> Meeting 2011 in Santa Clara, California, on Tuesday.
As if there were anything else he could have said to his shareholders!
> "This is so reminiscent of the RISC-CISC arguments of, gosh,
> twenty-five years ago," he said. "But then the argument was the
> other way around, it was that Intel can never possibly take on RISC
> architecture in the server space."
> He then allowed himself a bit of gloating. "Well, we kind of proved
> that wrong," he said.
Some upstart dislodged the mighty incumbent? That must be the kind of story your shareholders love to hear! I wonder if we're going to see Hot Patent Action against ARM licensees, a la Android licensees?
ARM design only, Intel design, manufacture and manufacture development
ARM's business model is simple,
Design, design, design oh and lisence
Lots of large players licensing the design.
Intel, design, manufacture and keep up with bleeding edge manufacture development i.e. 22 and 20 nanometer.
If ARM can get its A into G, I can see that their business model is much more agile than Intel's.
Think of all the mfgt overhead that ARM doesn't have, instead they can focus on design.
Just my thoughts.
Of and one other thing i.e. scenario:
Linux and LAMP (all easy to recompile to run on ARM) stack, web based apps, web servers etc can all run on an ARM or any other CPU for that matter.
We don't need high performance graphics etc.
However I do agree that ARM must adopt 64 bit soon, and (if they have not done already) allow for virtualization support etc.
Just been to a cloud tech conference today, there are two types of cloud models:
1. High performance computing clouds.
2. Low power, with enough CPU power, but high density blades etc.
Model 1, clearly Intel and AMD
Model 2, this is clearly a market for ARM
Just my two pennies
well it might *look* the same as the CISC/RISC argument of 20-25 years ago.
But it's not.
ARM have steadily been growing over *roughly* that period.
Tool chains did not really *exist* 25 years ago to handle the whole development/build process
A *significant* body of developers have grown up on the ARM architecture
Quite a lot of people got burned with MS Koolaid of "NT is Windows, but *better* and development is *just* the same." Except it wasn't.
And of course large scale e-commerce via the web did not exist.
Intel want you to run your apps on *their* hardware executing *their* instruction set at *their* prices.
ARM is an option for *anyone* whose workload is not hopelessly UI based and welded into Windows.
There are quite a lot of people who are in that position.
I repeat that Intel will make significant efforts to discourage *any* Microsoft efforts to get Windows server running on ARM.
A new Compiler...
...for ARMs migh have an advantage... compiling to machine code instead of pseudocode.
For sure someone will build a server, and then the benchmarks will reveal the story.
It is JUST like the situation 25 years ago...
The manufacturer of the cheaper processors, shipped in greater bulk and used by more developers will win out. The only difference is that Intel's on the opposite side of this battle. RIP.
Also there's a new landscape in that it is one against many: there are more independent fabs out there and ARM licensees can each produce their own designs from the chips upwards.
Intel have the lead on fabrication techniques and (as someone else said on this thread) virtualization. Neither are small things.
This sounds like a good environment for innovation.
Uh oh... Now They've Done It.
The most effective way to motivate someone to succeed is to tell them that they can't.
I'm just awaiting the first Acorn Archimedes 1U rack server....
"I'm just awaiting the first Acorn Archimedes 1U rack server...."
Another challenge. I'm just finishing the 1.8Ghz Dual Core BBC Micro PC. I have some other Acorns around. A 1U server wouldn't be difficult and I've already had Linux running on them.
"If Oracle ... were to support ARM on top-end servers"
"If Oracle ... were to support ARM on top-end servers, who is Intel to contradict them?
Hmmm. I wouldn't bet on that one for quite a while yet.
After all, Intel and Oracle are already contradicting each other on the future (or lack thereof) of IA64, presumably in no small part because Larry wants Sun to be worth something to his company (even though SPARC is no longer owned by Sun but by SPARC Intelnational)?
Hey El Reg, has anybody mentioned IA64 at the Investor Conference yet?
"...has anybody mentioned IA64 at the Investor Conference yet?"
What you mean as in; "And now could you all please stand as we hold a two-minute silence for IA64."?
Never mind error-correction or hyperthreading. Those, I think, would be simple enough to add to the ARM architecture. If Otellini really wanted to scare the crowd, shouldn't he have mentioned virtualization as one of the big things that the ARM architecture would need to have to be ready for use in servers?
"We've been doing server chips now since 1990."
Oh yes, I can vividly recall those racks and racks of shiny 486SX based gear...
Oh wait. Didn't we use proper computers in those days which ran, like, UNIX?
I'm desperately waiting for a few options to come along. Apparently 2011 is the year ARM servers will start arriving.
I've looked at the Taiwanese ASR-7110 but it is way too expensive for what you get and has some epic fails like having a 100baseT LAN port and a single one at that.
What I want is a SOHO router/server in several form factors ranging up from the Acer Revo to a small mini-tower capable of holding four or more 2.5/ 3.5" disks.
Must have Gigabit Ethernet x 2, SATA ports and USB2 (at least). It must take standard DDR RAM sticks (or standard laptop RAM for the smaller units)
Bonus points for PCI slots (space permitting), WiFi and USB3.
And finally, the smallest "Revo" sized unit should be less than $200. The Minitower with say six SATA + Drive bays no more than $400.
They want $320 for the ASR-7110 which is more than I paid for my Revo!
Most apps don't *need* 64bit
It's funny watching Otellini wilfully miss the point over and over again. Maximum Fail for you Otellini. Gee, do you think he's worried that the Atom isn't going to cut it against the ARM ? For my money he's missing a trick, he should be banging on about ARM's relatively poor memory bandwidth compared to the x86 chips out there. I suspect that what you pay in x86 Watts you get back in memory bandwidth.
"ARM's relatively poor memory bandwidth"
Surely memory bandwidth is a feature of an implementation not a feature of an architecture?
And as such, if there is a relevant memory bandwidth issue vs x86 (references welcome), an ARM licensee is free to address it?
"we proved that wrong"
I don't think Intel proved the cisc/risc thing wrong, I think Micro$oft did. They chose to drag Chipzilla along with them.
Java and Oracle
Most important s/w for contemporary servers are Java and Oracle. While I hate both, I fail to see why the Register readers don't have a slightest clue here (contrary to Intel execs).
Maybe to you
But a lot of people (myself included) couldn't give a damn about Java and Oracle at the server end. Oracle just seem to be very good at bumping their gums at everybody.
The big strength that ARM has for a prospective buyer is power consumption (really, nobody buying servers cares if their CPU can be made by one company or a dozen so long as they work and they can replace them when they don't). If I put my gear into a datacentre, I'd probably only get a rack half-full before being levied with extra power charges. That's on (mostly 2U) Xeons. If ARM can get the work done on half the power, that's a rack filled all of a sudden. And that's money saved every month on power. Once that saving starts to approach the cost of moving your legacy app from Windows / Oracle / Java / whatever, migrations become feasible.
Full disclosure - I was one of the guys with an Archimedes at home. I had hours of fun hand-coding ARM assembly, and learning to tweak individual clock cycles out of my inner-loops. I would *love* to see ARM succeeding in the datacentre. But realistically it's not going to happen overnight. My next server hardware cycle will be x64 - no question. The one after that? That's a tough one to call right now. And I'm in the Wintel camp - a migration isn't just a recompile for me, but a whole knot of licensing.
If I was an Intel shareholder I'd sell now
Ottelini is obviously a deranged idiot - LAMP works fine on ARM and so does .NET - though I'm not sure how well SqlServer would function on ARM - but a well written piece of code can be re-pointed at a different dbase engine with ease.
Me - I'm waiting for that $15 usb arm stick computer thingy to come out with 16 cores.....
MS will try real hard to get it's current lover to do what it wants. Upon failing to get this it will start two-timing in an attempt to make them jealous (force them to do what's required) or as a last resort start a new romance: SteveB has put his nads on line with Tablets market and doesn't mind who he snogs to get the job done (and if it just happens to do Jobs all the better).
Other points to note:-
-Intel have dropped HT in the newest chips, so all this BS about ARM growing new limbs like HT is contradictory.
-Legacy software.... what exactly is legacy software? We talking back to Win7, WinXP, Win2k3, Win2K, NT, Win3.1, DOS? Would it be so bad if we lost some of backwardness, might force companies to r-evaluate if they really need XYZ application they've been using for 15 years? Plenty of legacy stuff DOESN'T work on Win7\ Win2K8 so in some cases we are already biting that bullet. Maybe a new shiny compiled app on a new platform would be better? Don't know, WinTel has a huge legacy and the bagage that comes with it.
And sorry, I've been reading the cisc\ risc argument for 25 years- I don't give a monkeys any more. It's about what does the job best for the budget, we are way past the days when the Engineering picked the platform based on technical merit.
@Ramazan: I only have 1x oracle running on a RS\6000 and 1x Tomcat\-Apache, so it isn't the most important thing in my life. Keeping my hot servers cool is, especially as I'm encouraged to run them hard 24\7. My point being different people will have different priorities that affect a platform choice. Now if you'll excuse me I'm off the find out why another "legacy app" won't run under Win7 O_o
"Intel have the lead on ... virtualization.
Excuse me? Is there anything other than Intel PR to back up that statement?
Depending on exactly what folk mean by virtualization, the leaders are things like VMware, QEMU, SIMICS.
Intel make chips that happen to have some hooks that make virtualisation of x86 systems less tiresome than it would be without those hooks.
Afaik INTC have nothing indispensable, nothing even market-leading, either for virtualisation or for anything else, as far as product technology is concerned. They're still a leader at PR and deals though.