Short Intel and AMD! ARM servers coming in!!!

[ocre]

People don't want crappy atom powered netbooks. What makes you think they want an even more under powered arm powered laptop with a phone version of windows on it? Really ARM is great for phones. It still under powered for anything else.

have you heard of the ipad? what about the i pad2? Where have you been? There are some awesome devices called tablets out there which are pretty impressive. Especially considering the low power draw and battery life,

there are some amassing games for "underpowered" arm!

http://www.slashgear.com/shadowgun-released-for-android-today-we-go-hands-on-again-video-26191183/


plenty more too. They may be underpowered for your taste, but tp say they are too underpowered for anything but cell phones is insane. Tablets, ds, 3ds, the list goes on. ARM already exist in devices other than cell pones and they are outselling x86 chips by far. Apple is filthy rich and most of their income is from ARM devices.

 

[ocre]

I don't think it will be long before 2000w PSUs become "normal". The thing is, what's going to happen to people's houses electrically when we get there? Perhaps we can hook our computer up to the dryer's socket. :thumbsup::thumbsup:

OMG!

Lets hope not. Thats a joke i hope

funny though

 

[SickBeast]

Something's going to have to happen. 2000w isn't far off from what people are using now. It won't take long to double the power requirements for a high end PC.

 

[cbn]

The need for bigger power supplies seems to keep growing and growing in recent years, without massive jumps in performance to go along with it.

Well yeah.....and the fact Consoles are still using 2005 technology in 2011 doesn't help either.

Furthermore, I think these times we are living in are different than the past (with respect to PCs) for a few more reasons:

1. Cloud is permeating more areas. Heck even the school my mom teaches at is using Cloud for more and more things this year.

2. The smartphone race reminds me of the 90's CPU race, but today's PCs are well beyond the bottlenecks most people used to notice back then. Quad Core (let alone Hex or Octo-core) for the average person using Cloud? Or even Windows for that Matter?

 

[cbn]

Something's going to have to happen. 2000w isn't far off from what people are using now. It won't take long to double the power requirements for a high end PC.

Who is buying 2000 watt PCs and what are they using them for?

 

[-Slacker-]

Something's going to have to happen. 2000w isn't far off from what people are using now. It won't take long to double the power requirements for a high end PC.

That huge power consumption is due to a lot of extra hardware, NOT more power hungry hardware.

Running 4 gtx 580s in your PC is merely an option, you don't have to do it if you don't want to, nor do you need to. Having more options in general is always a good thing.

 

[quest55720]

have you heard of the ipad? what about the i pad2? Where have you been? There are some awesome devices called tablets out there which are pretty impressive. Especially considering the low power draw and battery life,

there are some amassing games for "underpowered" arm!

http://www.slashgear.com/shadowgun-released-for-android-today-we-go-hands-on-again-video-26191183/


plenty more too. They may be underpowered for your taste, but tp say they are too underpowered for anything but cell phones is insane. Tablets, ds, 3ds, the list goes on. ARM already exist in devices other than cell pones and they are outselling x86 chips by far. Apple is filthy rich and most of their income is from ARM devices.

Yet again people expect those kind of devices to be limited and under powered. People will not buy a laptop that is as limited as a tablet. When I am on a laptop I expect a full OS with real programs not apps.

 

[Idontcare]

Something's going to have to happen. 2000w isn't far off from what people are using now. It won't take long to double the power requirements for a high end PC.

Unless people start wiring dedicated lines in their houses for their PC's (like we already do for our clothes dryers and stoves), you are going to continue to see the absolute max practical power limit to be around 1725W (15 Amp * 115V) for your standard 15A household NEMA 5-15 plugs and circuits.

Same reason you don't see plug-in electric heaters or hair-dryers above 1800W (typical, unless they are designed for a dedicated NEMA 5-20 or 5-30 line).

 

[aigomorla]

HP announcing ARM servers!

Quick! Short Intel and AMD! They're done for!

lulz...

i'll worry that when apple buys AMD.

Meaning probably never...

 

[blackened23]

I actually want to see these things made so they can publicly benchmark it. Everything I know is word of mouth and I want to actually see the numbers.

Anyone saying ARM will take over is living in a fantasy land and probably doesn't understand it very well.

X86 kills arm in throughput, because arm is very stripped down low power design. ARM has a reduced instruction set, so it's probably an order of magnitude slower than an x86 processor of five years ago. x86 is faster because there are a lot of tricks to squeeze more performance, such as deeper pipes, path prediction, large caches, wider decoders, out of order execution, etc.....and that adds power consumption. To make arm as powerful as x86, would pretty much kill the whole low power ideology behind the design.

For many tasks though and a good gpgpu strategy I could see ARM making some inroads in very specifically tasked devices. I think x86 will fix power consumption and battery technology will improve long before arm gets powerful enough to compete in the x86 space.

In the meantime, ARM is going to stay on tablets.

 

[senseamp]

Unless people start wiring dedicated lines in their houses for their PC's (like we already do for our clothes dryers and stoves), you are going to continue to see the absolute max practical power limit to be around 1725W (15 Amp * 115V) for your standard 15A household NEMA 5-15 plugs and circuits.

Same reason you don't see plug-in electric heaters or hair-dryers above 1800W (typical, unless they are designed for a dedicated NEMA 5-20 or 5-30 line).

Someone should design a PC-dryer combo that dries clothes from the fan exhaust.

 

[CTho9305]

Anyone saying ARM will take over is living in a fantasy land and probably doesn't understand it very well.

X86 kills arm in throughput, because arm is very stripped down low power design. ARM has a reduced instruction set, so it's probably an order of magnitude slower than an x86 processor of five years ago. x86 is faster because there are a lot of tricks to squeeze more performance, such as deeper pipes, path prediction, large caches, wider decoders, out of order execution, etc.....and that adds power consumption. To make arm as powerful as x86, would pretty much kill the whole low power ideology behind the design.

For many tasks though and a good gpgpu strategy I could see ARM making some inroads in very specifically tasked devices. I think x86 will fix power consumption and battery technology will improve long before arm gets powerful enough to compete in the x86 space.

In the meantime, ARM is going to stay on tablets.

I disagree with all the points you seem to base your argument on... can you restate your argument or correct me where I'm wrong?

RISC doesn't mean reduced capability - it means you can reduce the amount of hardware you use to get the same amount of performance out. Until Intel outspent all the RISC companies, the fastest processors in the world were e.g. Alpha, MIPS (SGI), Power/PowerPC, etc. Intel made its money by working from the bottom up, by being willing to accept lower margins than the RISC vendors. ARM is now in the same position, raking in cash from cell phones, tablets, etc to use to attack Intel's markets with lower cost. ARM is arguably in an even safer position, since due to its large ecosystem there's less risk of any one vendor's Netburst disaster sinking the whole ecosystem.

AMD's K8 had a 12 stage pipeline, and I believe that remained through at least Phenom II. ARM's A8 has a 13 stage pipeline; A9 has an 8 stage pipeline and A15 has a 15 stage pipeline. If there's opportunity in deeper pipes, it doesn't look like ARM is missing out. Given that they've both shortened and lengthened pipelines across their products, they probably know what they're doing here.

A9 and A15 both have 32KB L1 instruction and data caches, like all Intel processors I can think of from Pentium 2 through Sandy Bridge, excluding the Netburst designs. A9 and A15 support up to 4MB L2 cache per 4 cores (that's more than many Athlon II's, right? maybe also some modern "Pentium" Core i#-derivatives?)

A15 is 3-wide, which means its decoders are at least as wide as K8 (i.e. through at least Phenom II). Now, many x86 instructions perform two tasks (e.g. read memory and operate on the result), but the decoders are equally wide. If ARM chooses to build an even-wider decoder, it's much easier to do for them since the instructions can't be any arbitrary length from 1 to 15 bytes... an x86 decoder becomes unwieldy above 4-wide, but for RISC instruction sets it's easy to go wider.

A9 is out of order, so ARM has been seeing any power penalties associated with that for a long time now.

 

[Arachnotronic]

The difference between ARM and Intel is that ARM just licenses IP and has a ton of people building custom SoCs. Intel designed, manufactured, and marketed their own products.

Intel's probably going to go down unless they can make significant strides in microarchitecture at the lower power level. It's all on Silvermont, now. ISA isn't what's important here.

 

[Voo]

I disagree with all the points you seem to base your argument on... can you restate your argument or correct me where I'm wrong?

RISC doesn't mean reduced capability

I totally agree there - RISC doesn't say anything about the performance (or even about the size of the ISA). But I think the rest of his post is valid even without postulating that.

Increasing performance hardly scales linearly with power increases - out of order execution does cost power, widening the front end isn't cheap either (rather hard to idle the frontend) and so on. Also it's not as if a RISC architecture means better performance/watt, cheaper or anything else - heck look at some ARM designs and you see more than one "bad" decision (supporting more than one ISA isn't that much different than having variable width instructions, predicates, implicit barrel shifter,..) - luckily they were still in a situation where they could implement a new ISA that eliminated those problems.

Also not to forget: Just look at how many really experienced, large teams designed architectures that looked good on paper, but just didn't work out in practice. Creating a high performance architecture is much more complicated than the low energy stuff ARM is still funded on (point in case look at the cache architecture of ARM and how that still evolves).

 

[cbn]

http://www.dailytech.com/article.aspx?newsid=23175

"Project Moonshot", HP's Secret ARM Servers become Official.

Curiously AMD is also joining in this program, though it remains to be seen where its interests lie. AMD recently quietly snuck into the DRAM market and also is a top seller of GPUs (and backer of OpenCL). Alternatively, exploration into ARM CPU designs also seems feasible for AMD, given that the company needs something to keep pace with Intel's process-driven energy efficiency improvements.

One big mystery is whether the server OS counterpart to Windows 8, likely to launch next year, will support ARM processors. Microsoft has already made it clear that Windows 8 will support ARM, but given that until right about now no major firms had ARM server plans, there was no word on Windows Server support. An endorsement by Microsoft could help ARM gain even more ground in the server market.

 

[TuxDave]

Anyone saying ARM will take over is living in a fantasy land and probably doesn't understand it very well.

X86 kills arm in throughput, because arm is very stripped down low power design. ARM has a reduced instruction set, so it's probably an order of magnitude slower than an x86 processor of five years ago. x86 is faster because there are a lot of tricks to squeeze more performance, such as deeper pipes, path prediction, large caches, wider decoders, out of order execution, etc.....and that adds power consumption. To make arm as powerful as x86, would pretty much kill the whole low power ideology behind the design.

For many tasks though and a good gpgpu strategy I could see ARM making some inroads in very specifically tasked devices. I think x86 will fix power consumption and battery technology will improve long before arm gets powerful enough to compete in the x86 space.

In the meantime, ARM is going to stay on tablets.

Yeah I know that's the general sentiment but it's always been "he says she says". I want a platform to show an scaled up ARM to settle the rumors.

 

[Dribble]

The difference between ARM and Intel is that ARM just licenses IP and has a ton of people building custom SoCs. Intel designed, manufactured, and marketed their own products.

Intel's probably going to go down unless they can make significant strides in microarchitecture at the lower power level. It's all on Silvermont, now. ISA isn't what's important here.

That's a killer advantage - Intel dictates how you must use their chips - for ARM you can make a custom design with some of their cores doing exactly what you want it too - that has a whole host of advantages.

The other massive advantage is price. Intel is a huge company that needs to make huge amounts of money or the share holders get upset. ARM is little, it doesn't need to sell it's chips for a lot to make a profit. That's a huge plus for ARM as it means Intel can't directly compete with ARM as they wouldn't be able to make enough money.

 

[Idontcare]

Yeah I know that's the general sentiment but it's always been "he says she says". I want a platform to show an scaled up ARM to settle the rumors.

Given how we know businesses to operate, I'd actually be willing to take the counter-position on this one which is to say that the very absence of such a "scaled up ARM" design is attributable to the fact that the feasibility concepts of such a design have been done time and again and the internal conclusions have always come to the same endpoint - it wouldn't be competitive if scaled up so its never been pursued to the point of releasing a commercialized product.

The history of IC design is replete with examples of designs that were attempted before their time (Apple's Newton anyone?), and they failed for various reasons owing to their die size, power consumption, absolute performance, and production cost...but rarely (I can't think of a single instance to be honest) has there been the case where a design comes out that is a smashing success but people are left scratching their head wondering why it wasn't attempted 10yrs prior.

Usually the battle with aggressive IC designs is that seeding time period that exists shortly after the time of introduction, the time in which the technology's fate is critically dependent on reaching a minimum critical mass in terms of established market share.

Failure to reach that critical mass means the difference between an iPod and a Zune.

So when I see the hypothetical arguments surrounding a "what if" scenario of a scaled up ARM device, or a scaled down x86 chip for that matter, I'm left with relying on the historically consistent observation that if it made sense (even on paper, which can sometimes be wrong in the false-positive sense, see bulldozer) then it would have already been done and we would have such examples of scaled up/scaled down IC's out in the marketplace right now.

The fact that we don't see such designs reduced to practice is most telling IMHO. Be it for technological reasons or economic ones, there must be some good reason why no one is attempting to field such designs in the pursuit of profits.

If there was even a remote chance of it working out then you can bet someone would already have attempted it even at the risk of it being a Newton or a Zune.

 

[zlejedi]

Probably it will work for some kind of niche market like those atom based servers:

http://www.theregister.co.uk/2011/07/18/seamicro_sm10000_64hd_upgrade/

 

[Voo]

So when I see the hypothetical arguments surrounding a "what if" scenario of a scaled up ARM device, or a scaled down x86 chip for that matter, I'm left with relying on the historically consistent observation that if it made sense (even on paper, which can sometimes be wrong in the false-positive sense, see bulldozer) then it would have already been done and we would have such examples of scaled up/scaled down IC's out in the marketplace right now.

Well only if you assume everyone has infinite amount of work and cash available to them. Otherwise you have to prioritize your projects based on market projections (especially considering how long it takes to come up with an design and have the finished product ready to sell), i.e. not "What will sell great today" but "What will sell great in 2+years".

I argue that if Intel had known how the market would look today around 2006 or whenever they thought about their first Atoms they would've designed them quite differently and given them much higher priority and well, we'd certainly have more suitable x86 designs today.

 

[Idontcare]

Well only if you assume everyone has infinite amount of work and cash available to them. Otherwise you have to prioritize your projects based on market projections (especially considering how long it takes to come up with an design and have the finished product ready to sell), i.e. not "What will sell great today" but "What will sell great in 2+years".

I asserted quite the opposite, it is because of the lack of infinite resources that companies must prioritize their efforts to pursue products that stand the best chance of being successful, and the fact that scaled-up ARM and scaled-down x86 have not been aggressively pursued is evidence of the conclusions made by the teams created to determine the market viability of those projects at every revisitation of project prioritization (usually on a quarterly and annual cycle).

(if you haven't worked as a professional in the industry then this might seem like gibberish, I've tried to boil it down to laymen terms, but in general there is a universally employed project management and ranking methodology that is used when prioritizing resource allocation for IC designs, the things that are not pursued are generally not being pursued for good reasons)

Some good ideas are a matter of timing, what isn't a good idea now may be a great idea in 10yrs time because of changing usage patterns in the market (disposable income and free-time) and/or changing production technologies in the meantime (node shrinks). But some ideas are simply bad ideas no matter the timing.

We can't conclude whether scaled-up ARM or scaled-down x86 are bad ideas right now for reasons of timing or for reasons that they simply aren't fundamentally scalable, but we can safely make conclusions regarding why those products don't exist at this time the same as we can might conclude for why flying cars and moonbases aren't common even in the 21st century when they were predicted to be common by the 1990's.

 

[TuxDave]

Given how we know businesses to operate, I'd actually be willing to take the counter-position on this one which is to say that the very absence of such a "scaled up ARM" design is attributable to the fact that the feasibility concepts of such a design have been done time and again and the internal conclusions have always come to the same endpoint - it wouldn't be competitive if scaled up so its never been pursued to the point of releasing a commercialized product.

This is why the OP is actually interesting because HP must've done some internal studies to conclude that they will work towards ARM servers because they found some market for them.

So while I am also entitled to viewing internal studies as well and since those aren't up for discussion, the benchmarks of an HP server will be.

 

[cbn]

We can't conclude whether scaled-up ARM or scaled-down x86 are bad ideas right now for reasons of timing or for reasons that they simply aren't fundamentally scalable, but we can safely make conclusions regarding why those products don't exist at this time the same as we can might conclude for why flying cars and moonbases aren't common even in the 21st century when they were predicted to be common by the 1990's.

http://www.anandtech.com/show/5027/appliedmicro-announces-xgene-arm-based-socs-for-cloud-computing

I find it interesting that the 3 Ghz ARMv8 only uses 2 watts per core.

Of course, how much performance ARM is able to give us at those two watts will be very telling for any future ARM Scaling up plans.

[A little off topic] I think it will also be very interesting to see if ARMv8 can put some competition on Haswell for Windows and Mac Ultra Books. (Although it would seem to me the impact of this battle will be small compared to the one for smartphones, which will involve more Operating systems and a far greater number of devices)

 

[cbn]

Speaking of Timing for scaling up/scaling down,

It seems (on very basic terms) like this is boiling down to a competition of process technology vs. X86 decoder penalty.

ARM stays safe from Intel by staying low power....thus keeping Intel's x86 decoder penalty proportionally large if Intel even wants to directly compete at low watts.

Intel's Process technology advantage could help then over come that, but the low profit of ARM's market makes them not want to use it. (Maybe a good example of Intel delaying low power processors is 32nm Medfield.)

A couple of twists in the plot for scaling down:

1. Consumers are wanting smaller and smaller devices. This increases demand and profits for ARM.

2. Intel wants profits, but as the the devices become smaller the Touch Interface becomes more important relative to x86 legacy compatibility. This puts Intel in a real predicament with respect to maintaining x86 compatibility at these lower power levels.

A twist in the plot for scaling up:

1. As ARM produces CPUs for the server market, it produces a CPU suitable for consumer content creation. This typically involves a keyboard interface and much larger display. But at what point can this be done satisfactorily with a docked smartphone for most (non power) users? How many users would ARM need at this level to begin forcing x86 out of the mainstream content creation level? (thus paving the way to higher ARM CPU power levels for the future)

Conclusions:

1. At this point it appears the case could be made for ARM not scaling up so much (or at least not so quickly).

2. A case could also be made for Intel not scaling x86 down too far if it can get its Silvermont phones powerful enough to play nicely with legacy apps in a docked configuration.

 

[Voo]

but in general there is a universally employed project management and ranking methodology that is used when prioritizing resource allocation for IC designs, the things that are not pursued are generally not being pursued for good reasons

Well Intel has said that they bet on netbook/tablet designs when creating the first Atoms (in some of Anands old articles about the first atom I think) and I really don't see how that was a smart move in retrospective. Because quite clearly netbooks are a dying breed and the only successful tablets at the moment are upscaled smartphones.

So I really don't see how one could claim that Intel's Atom strategy makes sense in retrospective - if they had started 2005 with designing a CPU not for netbooks but smartphones I think their current smartphone market share wouldn't be basically zero. But then who'd have predicted 2005 the advance of the iPhone?