Slashdot Mirror
Windows Server on ARM Is Finally Happening, And It Should Worry Intel (bloomberg.com)
Mary Jo Foley, writing for ZDNet: There have been rumors for the past several years that Windows Server would be coming to ARM. Today, March 8, that rumor became an acknowledged reality. Microsoft officials said that the company is committed to use ARM chips in machines running its cloud services. Microsoft will use the ARM chips in a cloud server design that its officials will detail at the the US Open Compute Project Summit today, March 8. Microsoft has been working with both Qualcomm and Cavium on the version of Windows Server for ARM, according to company officials. From a report on Bloomberg: Intel chips have remained one of the sole big-name products widely in use. Microsoft's work with ARM, in progress for several years, could pave the way for a real challenge to Intel, which controls more than 99 percent of the market for server chips. [...] Any challenge to Intel's dominance in server chips is a threat to its most profitable business and main revenue driver as demand for PC processors continues to shrink. The company's Data Center Group turned $17.2 billion of sales into $7.5 billion of operating profit in 2016, and Intel has been running ads that say, "98 percent of the cloud runs on Intel."
I am curious on how the lean ARM processors would cope up with the Windows bloat. Windows "server" boots the GUI first!
How is to get the BSOD on arm CPUs... I am getting really curious: Intel is loosing it's crown!
Sent as ripples into the electromagnetic field. No single photon has been harmed in the process.
Intel is not worried. How is ARM any more of a threat today than AMD was?
Intel started building lower powered chips a long time ago to compete with ARM and have, in a number of areas, surpassed them. Time and again, Intel has been able to ramp up their R&D to stave off serious competition. I don't see ARM being any different.
It isn't a threat currently, but one can be sure that if ARM saw that it could make an in-road into the data center, it might just ponder making chips with greater overall horsepower for a "server line" of ARM CPUs. Just because ARM at the moment is more interested in low power chips doesn't mean they wouldn't ponder a move towards higher performance, and certainly if they could advertise both Windows and Linux as supported platforms, that makes for a pretty compelling argument.
The fact that Microsoft is going this direction clearly indicates to me that there must be some interested at the hardware end to produce server-grade chips.
The world's burning. Moped Jesus spotted on I50. Details at 11.
Last I heard, Intel still beats ARM easily on performance per watt and that is what's important in the data center, not absolute power usage. Did something change?
No, nothing changed. This is just another round of misunderstanding from the anti-Intel zealotry. The most hilarious part of their recent (relatively) attachment to ARM processors in their war against Intel is that Intel has all the paperwork done to be not just an ARM fabricator, but optimizer. If ARM ever gets any performance advantage over x86, Intel-ARM CPUs will hit the market and (based on current trends) be roughly 8% more efficient than any competitor using the same layout. If Intel takes the time to optimize an ARM layout before fabbing, probably 12-20% more efficient while remaining fully compliant to the standard.
Did something change?
ARM on 10nm.
Mass production has begun, while Intel is still at least a year away from 10nm. Its why Intel announced last August that when it finally does get 10nm to work, it will be making ARM processors not just x86 in those fabs.
Intel is rightly fucked as a vertical company. A decade from now Intel, if it hasnt been decimated and sold off by then, will be just another rent-a-fab company.
"His name was James Damore."
Got malware?
Everybody knows, "the Cloud runs on Intel."
Intel's 14nm process is in most ways better than Samsung's 10nm (Qualcomm's fab for this chip). IMHO this isn't really going to be won on feature size of the process between the two.
The advantages Cavium and Qualcomm have are cheaper chips (also more integrated, SAS, PCI lanes) with higher core counts and more memory channels than Intel. In some applications the higher memory channels count and bandwith will be very valuable (HPC). Intel will be able to discount the heck out of their high core count Xeons to match on price and they can lower the clock to match on IPS for each core. They can't add a memory channel where it doesn't exist.
While I agree that the vertical integration of Intel's past may not be the best way for the future, they still have the best process engineers in the world. They still have the best process for the foreseeable future. They also have some of the best semiconductor engineers out there. So while there is a fight coming. Intel is hardly "fucked" or headed for decimation.
If you manage a server farm, you have to worry about power consumption:
1. the more power you require, the more heat you generate
2. the more heat you generate, the more cooling you require
3. the more your hardware runs on higher temperatures, the lower its lifespan
Granted, #3 might not be relevant with rapidly changing technologies, but then again there's still companies out there using old servers running COBOL and FORTRAN so you never really know how long your servers will run. Could be a year, could be a decade, could be a quarter of a century.
#DeleteFacebook
Or maybe it's a chicken-egg problem and Microsoft decided to make the first step. Or maybe they're seeing more requests for Windows Server on ARM and they've made projections about future demand and it's going to happen soon according to their numbers.
#DeleteFacebook
Considering MS already ported the Windows kernel to ARM a while back, and certainly Windows NT, including the older CLI toolkit were all ported over to multiple architectures back in the day, it may not be all that hard. Obviously there's going to be some recompiling and porting over of a lot of server components, so I'm not saying that porting the entire Server ecosystem to ARM will be trivial, but it's not really a massive leap, just a lot of recompiling and testing. A lot of the admin tools have been moved over to .NET due to Powershell's increased use, so in some cases we're talking not much more than testing on the ARM versions of the CLR.
The world's burning. Moped Jesus spotted on I50. Details at 11.
Whether this should worry Intel does not depend on Intel's market share in the server world, but on Microsoft's. I suspect not all that much.
Intel has been an ARM licensee for over a decade, they can produce them whenever they desire to, and DID produce a lot of ARM chips in the pre smartphone PDA days.
luckily AMD recently pulled their ARM server Zen line turned back to X86. guess they wanted to shore up that 1%.
you have to worry about performance per watt, not just raw watts. If you get lower transactions (or whatever) per watt with ARM then you've solved nothing and perhaps made things worse. Previous generation ARM chips couldn't match Intel on performance per watt, perhaps the current and future gen ARM chips can change that? I haven't seen that they have succeeded yet though.
sorry i was incorrect - the ARM parts were the Opteron product line, and the Zen chips are x86.
Intel's 14nm process is in most ways better than Samsung's 10nm
Citation needed, and no I dont accept a citation that basically says "thats not real 10nm" because that 14nm intel isnt real 14nm. IIRC the smallest feature size on Intel 14nm is still 28nm, and thats after improvements while still at "14nm." Intel invented lying about feature size.
Also, Samsung isnt the only fab company beginning 10nm mass production well before Intel. TSMC and Toshiba will be mass produing before the end of the year (the end of the year is when Intel last said would be the absolute earliest that they could do it.)
TSMC, Samsung, and Global Foundries are also in a partnership with IBM on working out the kinks on 7nm. Intel is very fucked.
"His name was James Damore."
I started my career running NT on Alpha's, great runners, good performance and reliability. However, near zero support from the third party ISVs or add on hardware manufacturers (third party NIC and RAID cards for instance). Unless there is strong market uptake of servers using ARM, I foresee much the same path as NT on Alpha did.
Will it run in a VM? will arm systems be locked to windows boot loaders? don't want to be stuck with hyper-v.
Do ARM chips have the pci-e for storage / 10-gig-e?
Intel has enough money they could throw 2-3 teams on it if they wanted to. One of them would succeed.
Yeah but XScale started out as StrongARM from DEC. Intel basically got the design team when they bought some of the remains of DEC. They basically applied some of the chip design ideas from the Alpha for an ARM core. Eventually Intel sold that business to Broadcom I think. They still have an ARM license but I'm not aware of them using it for anything right now.
Let's assume the following :
- 25% of servers run Linux
- 50% of servers run Windows
- Linux is compatible with ARM
- 1% of server CPUs are ARM, the remaining 99% are Intel
- All ARM servers run Linux
- The situation with Windows Server will now be the same as with Linux regarding platforms.
With these generous assumptions in mind, Intel market share will drop from 99% to 97%, that's 2% less sales, big deal...
The reason servers don't run ARM is not because of incompatibilities. It is because they need more computing power than ARM chips can offer.
Enjoy the CPU overhead of emulating x86 programs on an ARM CPU!
"When information is power, privacy is freedom" - Jah-Wren Ryel
ARM has been an interesting platform of late, but a lot less useful than it could be. Proprietary bootloaders, custom hardware trees, all work against it. No ARM device that I know of can run a stock, off-the-shelf Linux distro with a fairly stock kernel. Not even the Pi. Maybe if MS starts pushing a Window ARM platform, it might provide impetus to manufacturers to standardize the boot loader and the platform so off-the-shelf OS's can run.
I have a drawer full of various ARM devices that were theoretically really neat and useful but in practice proved to be more trouble than they were worth. For example I have two sheevaplugs but the effort to try to update them from their default ancient ubuntu distro is via tftp and serial port u-boot prompt is just not worth the effort. I got more utility with a cheap Intel NUC, even though it was several times the cost of the plug.
Life is a bit better with the Pi since I can just burn a new SD card and boot on it. Still requires a custom distro and kernel. Repeat for every SBC like the Pine64.
Until things get more standardized, I'm skeptical that ARM will do any serious damage to the Intel hegemony, low power notwithstanding.
If all this was true then AMD could not sell their server processors.
Yet they are highly popular.
And they, like all AMP processors run hot.
Do not look at laser with remaining good eye.
They won't. There seems to be some belief among younger folks that ARM is somehow special and that instruction set is more important than anything else. The reality is that is simply NOT true. Instruction set is essentially meaningless. Intel chips haven't used x86 internally since the first Pentium chips. The x86 is decoded and fed to the internal processor by a decoder that uses less than 123K transistors (not even 0.1% of the die). Because x86 is nothing more than an abstraction layer at this point this allows Intel to use whatever is the new hotness inside their CPU.
Even if you believe non of that consider this. There have been at least 6 companies that have tried to produce ARM servers that dropped the effort or abandoned the project after the test hardware was produced. The simple reality was that even with brand new top of the line designs they couldn't beat Intel in either raw performance or performance per watt. Every attempt has failed and that was using Linux where the software can be fully optimized for ARM and all software recompiled for it. Expect nothing like that on Windows and just like Windows RT it will be a colossal failure.
Intel will keep winning, they are literally 2 process nodes ahead than every other silicon manufacturer, their CPU's could be two generations old and still beat top of the line processors from other manufacturers just because of the process node advantage.
Had an XScale based PDA "back in the day" (Dell Axim X5, 300MHz?), and I'll attest that the performance/usability was quite good.
There is no XUL, only WebExtensions...
Reply to myself to avoid spam, but I also had a Compaq iPaq 3630 (206MHz StrongARM), and it was more than capable of doing "useful" things such as NES emulation.
There is no XUL, only WebExtensions...
It's not just the decoder, branch prediction is more complex (more scope for bugs), the pipeline has to be more complex due to the variable length instructions that can be one byte long up to 7 bytes long. It doesn't matter a lot in chips where you have few very powerful processors (traditional servers), but where you have many many low power processors it adds up.
Oolite: Elite-like game. For Mac, Linux and Windows
Why yes ... Just the other day I was accosted by a man who warned: "I better not find out you got a mind toward that there Intel or there'll be some need to call the Sheriff! Zealots I tell you! Like Savior Faire they are everywhere!!
Guns don't kill people; Physics kills people! - John Lithgow as Dick Solomon on Third Rock From The Sun
Do ARM chips have the pci-e for storage / 10-gig-e?
Yes.
Marvell:Armada XP supports four PCI-e 2.0 ports (two x4 ports can be configured to Quad x1 – up to 16 lanes)
Calxeda: Energycore SoC supports PCI Express Four (4) integrated Gen2 PCIe controllers
nVidia Tegra 2 also supports PCI-e. The ARM and PCI-e licenses are compatible. Electrically of course, the choice of supported buses is entirely up to the chip designer.
PCIe2, wow. Obsolete junk.
and only 16 lanes.
If it is about low watts vs shear performance Intel should toss Atom processors in to the server market. Atom processors are x86 compatible so you don' t have to redo all your code to run it.
And when you want to move off the baby processor and on to a real server then its just a straight move up.
Why am I still posting at 0 when my karma is excellent?
I read at +2. If your post doesn't reach that level I will not see or respond to it.
https://www.bing.com/search?q=... The story is from April 2016.
https://www.bing.com/search?q=... Articles are from August 2016, but dont quite seem to be saying what the other poster was saying.
Obsolete? Perhaps you should look up a definition of the word. Junk it isn't, 8GBps* should be more than enough for any reasonable use case.
(* 0.5GBps per lane x 16 lanes)
Anon here, this article seems to break it down:
Lets check that "break down"
..."
..and the "+" in Intels new "14nm+" is Intel actually INCREASING a feature size so that they can increase voltages and thus clock higher. The last time Intel did that was Pentium 4 / Netburst. Better hope they dont do a "14nm++" generation.
"Intel’s processes use the same backbone as the advertised node (a 14nm process will use a 14nm backbone)
See right here? Thats a lie. You are reading a lie and not questioning it. Intel is the semiconductor manufacturer that invented lying about node size. There is no feature in Intel chips that are 14nm be they their "14nm" chips or their new kaby lake "14nm+" chips, yet your article says Intel is the only one doing true 14nm. Its fucking comical how big a lie your article is spewing.
Intel invented lying about node size. Even your article shows no feature size smaller than 42nm when calling Intels latest lie "true 14nm"
"His name was James Damore."
Really?
About 1 1/2 years ago, AMD was on its way out in the server market because the Opterons were not competitive enough anymore: http://www.eweek.com/servers/amd-aims-to-reinvigorate-x86-server-business.html
Now they are going to give it another try with Zen, and I think it is a promising try. But then again, Zen is not more power hungry than comparable Xeons. Maybe less so.
C - the footgun of programming languages
Sometimes maximum watts matters more than performance per watt and sometime Intel doesn't have anything decent to fit the niche. Nothing has changed really.
Intel has enough money they could throw 2-3 teams on it if they wanted to. One of them would succeed.
AMD's problem is they thought they could throw away 2-3 teams and did.
Not exactly. You have to worry about ENOUGH performance per watt. If you can get something to get the job done in a reasonable time frame without burning a lot of power that's great. That's what low power consumption systems are for.
If that's nowhere near enough performance you get whatever does the job instead no matter how many watts it burns. Trying to chase both generally just annoys everyone - pick one.
Could be Intel's biggest mistake to let that happen!
>/dev/null 2>&1
Past experiences show that Microsoft hates to develop for 2 different platforms. They couldn't even support the differing Intel Pentium and Intel Itanium architectures let alone 2 different brands. (The AMD chips are trapped within the Intel architecture.) Whether M$ will drop ARM or Intel remains to be seen thought Intel has won all the previous battles.
NetBSD: the cathedral vs the bizzare.
You might be surprised with some of the new server oriented ARM chips.
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
Yes.
Though why just 10-gig-e? The ARM chips I'm working with support multiple integrated 40-gig-e ports and multiple PCIe gen 3 buses.
I have the data sheet for the chip being discussed in the article in front of me. While I can't go into details, it is no slouch and has a massive amount of memory and I/O bandwidth. 10G? That's nothing. I regularly deal with 80G (dual 40G XLAUI) with this chip. The ARM chips are a newer generation than this so the cores are faster.
Some of the ARM chips I work with have built-in RAID engines for offloading all of the RAID calculations as well as engines for a number of other things. Here is a link that shows the I/O oriented chips.
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
The Cavium chips handle multiple PCI-e 3.0 ports. Here's a brief on the I/O capabilities (though it doesn't go into much detail):
http://cavium.com/ThunderX2_AR...
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
The Cavium chips handle multiple PCI-e 3.0 ports.
Why does that page read like the TimeCube guy? It sounds significantly dubious, as if the chip doesn't actually exist. It sorta sounds like it could exist, but the company is mostly just tossing out a proposed spec, in hopes that someone will fund development of it. Too many superlatives, too many uses of the word "hundreds" in contexts that are exceedingly unlikely. Sounds bogus.
The chip is the one referred to in the article, and yes, it does have hundreds of gigabits of bandwidth to it as it says, especially when there are many 25GBps serdes.
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.