Why are there no high-power ARM CPUs?

Discussion of all forms of processors, from AMD to Intel to VIA.

Moderators: Flying Fox, morphine

Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 12:56 pm

tl;dr in bold.

ARM CPUs carry a typical TDP of <5W or so, give or take. A "fast" ARM CPU typically hits about 2.0GHz and might have four cores or so packed into a die, and each core is highly efficient (especially at idle). Some ARM chips can sort of compete with an old Core Duo depending on the benchmark. But all of them are designed for rather small TDPs compared to say, a desktop x86 CPU (which can consume well over 100W depending on the make and model).

I know of the exponential relationship between switching (clock) speed and power, etc., but:

  • Why are there no high-TDP ARM chips?
  • If it's because of performance or high-power efficiencies, how would one compare to a modern day Ivy Bridge or Steamroller?
  • And ultimately (edit: from a technical perspective, disregarding funds), what is limiting manufacturers from cranking ARM to >90W? Die size? Performance? Architectural limitations?

I hope this is the right subforum for this. I've been wondering this for awhile and finally figured I'd create an account here to ask. I have some EE and a little CompE background so don't hold back (until I start complaining). Thanks in advance!

EDIT: General answers I've compiled so far:

Wording or points subject to change as conversation evolves

  • Diminishing performance returns: ARM is not as efficient as x86, and remains most efficient at low power. Reduced power to performance efficiency at higher clockspeeds would overshadow net performance gains.
  • Die size to cooling ratio: ARM typically has a tiny die and pumping 100W through 1.5mm^2 or similar is just not feasible--it's a soldering iron.
  • Limited Parallelism: Scaling isn't as easy as just slapping a bunch of cores together, you need to feed them properly. Serial workloads suffer greatly.
  • Beyond clockspeed: Current ARM arches don't have the same level of branch prediction/caching/etc as the x86 competition.
  • Design: A high-TDP ARM CPU would be worth it only if it were designed from the ground up to work as such, similar to what NVIDIA is doing with Project Denver, or more "advanced" cores like the A15.
  • Demand/funding: I dislike this answer because it's not what I'm after, but ok. There's little marketing demand and research into high-TDP ARM.

And here are some things that we might still be disputing:

  • ISA: ARM isn't designed to do "as much" at the ISA level as x86 is.
  • RISC vs CISC: We can bring this up but many argue that line is blurring.
  • It might be worth noting ARM's current lack of 64-bit compatibility or similar hardware limitations for theoretical performance or resource limits.
Last edited by Duct Tape Dude on Thu May 02, 2013 8:47 pm, edited 2 times in total.
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 1:17 pm

Duct Tape Dude wrote:[*]Why are there no high-TDP ARM chips?


Because that's not where the research money's perceived to be, mostly because the market hasn't made much space for such a part yet. Conventional wisdom is that you go x86(-64) for performance, and ARM for power savings, and with the sheer amount of inertia x86 has built and the work put into compilers there, that will be hard to shake.

Duct Tape Dude wrote:[*]If it's because of performance or high-power efficiencies, how would one compare to a modern day Ivy Bridge or Steamroller?


What, a high-TDP ARM part? Or a low-power ARM part in general?

Duct Tape Dude wrote:[*]And ultimately, what is limiting manufacturers from cranking ARM to >90W? Die size? Performance? Architectural limitations?


You could probably argue that architectural limitations inherent to ARM to this point stand in the way, but most of it boils down to money...
AMD FX-8320 | ASRock 970 Extreme4 | 24GB Corsair Ballistic DDR3-1600 | 2048 MB MSI Geforce GTX 660 | 600W Corsair Builder Series |Xubuntu 14.04 x64 | Cooler Master CM690 | Pioneer Blu-ray RW Drive | 240 gig Mushkin SSD | 500 gig Western Digital SATA Drive
Concupiscence
Gerbil
 
Posts: 85
Joined: Tue Sep 25, 2012 7:58 am
Location: Dallas, TX

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 1:40 pm

Concupiscence wrote:Because that's not where the research money's perceived to be, mostly because the market hasn't made much space for such a part yet. Conventional wisdom is that you go x86(-64) for performance, and ARM for power savings, and with the sheer amount of inertia x86 has built and the work put into compilers there, that will be hard to shake.

I get that. But there exist other high-performance CPUs (here I refer to high-TDP units) like Itanium or even RISC arches like SPARC or POWER, and ARM is not one of them.

Concupiscence wrote:What, a high-TDP ARM part? Or a low-power ARM part in general?

Yes a high-TDP ARM part. That's what my post is about.

Concupiscence wrote:You could probably argue that architectural limitations inherent to ARM to this point stand in the way, but most of it boils down to money...

Is that just speculation or are you basing this architectural limitation on some source?

I just want to know from a technical perspective what is holding ARM back. I'm sorry if I didn't clarify that earlier.
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 3:23 pm

It takes a lot of money and effort to design a new chip. Chip makers don't want to design something unless they are sure if there is an opening in the market for their product to thrive.

it is now, the High-TDP market is flooded with Intel and AMD processors. The healthy competition forces prices way down. Margins are thin. There is no way for a new-comer to recover the heavy investment in development and production of a new chip.

Is there even a market for a high-TDP ARM Chip? As it is now, there is very little in terms of software that will run on ARM and require desktop-class performance. Maybe WindowsRT and Android will bring up demand for ARM on desktop. But that is not a safe enough bet for chipmakers to jump on.

Technically, there is nothing stopping anyone from making a 130w ARM CPU.
Carpe diem quam minimum credula postero
sid1089
Gerbil Team Leader
 
Posts: 290
Joined: Wed Jul 26, 2006 4:56 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 3:32 pm

sid1089 wrote:It takes a lot of money and effort to design a new chip. Chip makers don't want to design something unless they are sure if there is an opening in the market for their product to thrive.

it is now, the High-TDP market is flooded with Intel and AMD processors. The healthy competition forces prices way down. Margins are thin. There is no way for a new-comer to recover the heavy investment in development and production of a new chip.

Is there even a market for a high-TDP ARM Chip? As it is now, there is very little in terms of software that will run on ARM and require desktop-class performance. Maybe WindowsRT and Android will bring up demand for ARM on desktop. But that is not a safe enough bet for chipmakers to jump on.

Thanks for the reply. I understand the lack of demand and funding, but that's not what I am after.

sid1089 wrote:Technically, there is nothing stopping anyone from making a 130w ARM CPU.

Is it really just marketing holding ARM back? I think there would obviously be a performance or maybe efficiency discrepancy since ARM is not dominant over x86 in the high performance market, but if so, what part breaks down to cause this loss of viability? Stagnating clockspeeds? System buses? etc.

I think I might have phrased this wrong as nobody seems to be answering my questions the way I intended, or maybe everyone is more well-versed in marketing than engineering or CPU design concepts. I'll try again:

What technical limitations are there to creating a High-TDP ARM CPU?
How would the performance of an ARM CPU augmented to a high TDP compare to a standard x86 CPU?
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 3:38 pm

Duct Tape Dude wrote:[*]Why are there no high-TDP ARM chips?

From the wiki:
"ARM states that the TSMC 40G hard macro implementation typically operating at 2 GHz; a single core (excluding caches) occupies less than 1.5 mm2 when designed in a TSMC 65 nanometer (nm) generic process[3] and can be clocked at speeds over 1 GHz, consuming less than 250 mW per core."

My bolding. So if the 40G consumes (let's say) 250mW per core, you'd need at least 90/.25 or 360 implementations to consume the 90W you mentioned, and that's assuming complete duplication of memory control, cache and I/O. In reality, you'd probably need closer to 500 cores if you were inclined to build something this weird. So the reason you don't see high-TDP ARM designs is that if you built one you'd have a product for which there is no demand or market.

That's basically it. The ARM is designed to not eat a lot of power. Hell, it's only 1.5mm square mm.
This problem was caused by Windows, which was created by Microsoft Corporation.
sluggo
Gerbil Jedi
Gold subscriber
 
 
Posts: 1539
Joined: Wed Feb 16, 2005 8:44 pm
Location: under the table and dreaming

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 3:44 pm

One major reason is lack of 64-bit support.
Game_boy
Gerbil Elite
 
Posts: 564
Joined: Mon Aug 06, 2007 12:46 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 3:52 pm

Your title asks "Why are there no high-power ARM CPUs". Economics is the answer to that question.

Technically, there is no limitation. ARM, the instruction set, does not dictate how large a chip can be or how much power it can consume.

You could design your own ARM core that is comparable in size/TDP to an intel x86 processor. If you design it good enough, I don't see why it cannot hit the same performance level as the Intel chip. But keep in mind that Intel has decades of experience developing and tuning the performance of their chips.
Carpe diem quam minimum credula postero
sid1089
Gerbil Team Leader
 
Posts: 290
Joined: Wed Jul 26, 2006 4:56 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:02 pm

Duct Tape Dude wrote:What technical limitations are there to creating a High-TDP ARM CPU?
How would the performance of an ARM CPU augmented to a high TDP compare to a standard x86 CPU?
Well, there aren't any. You could create a massively parallel ARM processor with high TDP. It would just run like bollocks.

ARM design is focused on space and power efficiency. If you look at Phoronix synthetic benchmarks (I know, synthetic, but they serve this purpose), Ivy Bridge architecture is over 600% more efficient in terms of performance-per-watt than Cortex-A9, it's just difficult to scale an Ivy Bridge down to the very low power envelope occupied by ARM chips without losing critical functionality that makes it so fast. Desktop chips use a lot of power, but they are also an order of magnitude faster.

Simply scaling up the ARM design wouldn't provide great performance; the chips are certainly very impressive given their size, heat, and power envelopes, but don't think that they are competitors for AMD or Intel's big x86 designs. They perform very poorly in comparison to the full-fledged superscalar out-of-order superpipelined SMT desktop chips. Those chips are massive and hot because desktop power and heat envelope are not a limiter for them, so the designers are much more free to use exotic means to improve performance.
i5-3570K @ 4.4 (NH-C14), 4x8GB DDR3-1866, GA-Z68MA-D3H-B2, ASUS GTXTITAN-6GD5, 128GB Vertex 4 / 2x60GB Vertex Plus R2 / 2x2TB Barracuda 7200.14 RAID0 / ANS-9010 (4x4GB), SST-DA1000 (PSU), 2x VS229H-P, 1x VG248QE, 1x MIMO 720F, Corsair Vengeance K90+M95
auxy
Gerbil Elite
 
Posts: 781
Joined: Sat Jan 19, 2013 4:25 pm
Location: the armpit of Texas

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:24 pm

One could build a CPU compatible with the ARM instruction sets that was high power and high performance if you designed it 100% from the ground up for that purpose.

There are no technical limitations, and really there never were. Scaling up current designs to that high TDP would be more of a technical problem than a new design from the ground up, really.
I.S.T.
Gerbil First Class
 
Posts: 178
Joined: Wed Sep 14, 2005 5:18 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:25 pm

It exists now.

The A15 is the first server class ARM chip and at the highest frequencies (phone versions use much lower frequencies as well as big.Little to keep power use down) it's fairly high-powered for ARM.

But still not 90W high powered because ARM wants to maintain their power use advantage.
ChronoReverse
Gerbil Elite
 
Posts: 736
Joined: Wed Dec 12, 2007 4:20 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:29 pm

I'm NOT a computer engineer, but I think it comes down to how it was designed. As others have stated, it was designed to be low power/"low" performance and so it probably wouldn't scale well. Look at it from the flip side. The Pentium 4 with Netburst architure was designed to be VERY fast and roadmaps had it going to 10Ghz. Although as Intel found out, thermals play a big role. But you'd never be able to scale a P4 down to a low-power (properly) chip.

So from the way I see it, technically, nothing is stopping ARM from being a high performance, except that it'd be rubbish. To use the car analogy (first!), I guess one could look at it from the point of view of taking the engine from a SMART car and trying to scale it up to 1000hp. Not going to happen.

Of course, I could be wrong. Since I'm not that techie...and all.
Venii, vidii, vicii
Wii came, Wii saw , Wii conquered
Sargent Duck
Grand Gerbil Poohbah
Silver subscriber
 
 
Posts: 3068
Joined: Thu Mar 13, 2003 8:05 pm
Location: In my secret cave that has bats

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:51 pm

sluggo wrote:So if the 40G consumes (let's say) 250mW per core, you'd need at least 90/.25 or 360 implementations to consume the 90W you mentioned, and that's assuming complete duplication of memory control, cache and I/O. In reality, you'd probably need closer to 500 cores if you were inclined to build something this weird.

You do realize you just described a wonderful massively parallel general processing unit, right? The blend of GPU and CPU... Who wouldn't want 500 cores working for them at 90W? Or even a fraction of that running noticeably faster? That sounds incredible to me.

sluggo wrote:That's basically it. The ARM is designed to not eat a lot of power. Hell, it's only 1.5mm square mm.

Honestly I think this is the best answer so far: it's only 1.5mm^2. Everyone here might sooner recognize this problem as the temperature increase from Sandy to Ivy even with the TIM and heatspreader fixes. Thanks for this!

Game_boy wrote:One major reason is lack of 64-bit support.

True, but even the new ones that are taped out won't be high-TDP. You bring a valid point though: they support many fewer features than x86.


I.S.T. wrote:One could build a CPU compatible with the ARM instruction sets that was high power and high performance if you designed it 100% from the ground up for that purpose.
There are no technical limitations, and really there never were. Scaling up current designs to that high TDP would be more of a technical problem than a new design from the ground up, really.

Do you have any sources or references on this? Particularly the "no technical limitations?" I'm just curious, not trying to shoot you down!

ChronoReverse wrote:It exists now.
The A15 is the first server class ARM chip and at the highest frequencies (phone versions use much lower frequencies as well as big.Little to keep power use down) it's fairly high-powered for ARM.
But still not 90W high powered because ARM wants to maintain their power use advantage.

Aha, thanks! The lack of the 90W-class part makes me wonder what wall they hit first, is it just a raw performance-per watt disadvantage at high power?

Sargent Duck wrote:The Pentium 4 with Netburst architure was designed to be VERY fast and roadmaps had it going to 10Ghz. Although as Intel found out, thermals play a big role. But you'd never be able to scale a P4 down to a low-power (properly) chip.

Netburst was a CISC arch with an insanely long pipeline (20something steps?) and ARM has never been limited by the ability to cool the chip. It's been limited by how little power they want to feed it. Two totally different design limitations since one was designed to sip power, the other was designed disregarding it.

Sargent Duck wrote: I guess one could look at it from the point of view of taking the engine from a SMART car and trying to scale it up to 1000hp. Not going to happen.

I raise you this: http://www.youtube.com/watch?v=UQnz6AyPjgs
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:56 pm

Well, the thing is, ISA doesn't really limit you that much in terms of how big you want to go, or how small. x86 started out as a set of chips for a calculator for Pete's sake.
I.S.T.
Gerbil First Class
 
Posts: 178
Joined: Wed Sep 14, 2005 5:18 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 4:59 pm

auxy wrote:Well, there aren't any. You could create a massively parallel ARM processor with high TDP. It would just run like bollocks.

I'm most curious as to why though... And by how much?

auxy wrote:ARM design is focused on space and power efficiency. If you look at Phoronix synthetic benchmarks (I know, synthetic, but they serve this purpose), Ivy Bridge architecture is over 600% more efficient in terms of performance-per-watt than Cortex-A9, it's just difficult to scale an Ivy Bridge down to the very low power envelope occupied by ARM chips without losing critical functionality that makes it so fast. Desktop chips use a lot of power, but they are also an order of magnitude faster.

Aha, this is probably one of the best answers bridging financial feasibility and technical reasoning. ARM would only lose efficiency at higher TDPs. This makes sense, thanks.

auxy wrote:Simply scaling up the ARM design wouldn't provide great performance. They perform very poorly in comparison to the full-fledged superscalar out-of-order superpipelined SMT desktop chips. Those chips are massive and hot because desktop power and heat envelope are not a limiter for them, so the designers are much more free to use exotic means to improve performance.

These are all great points, and your post probably settles this issue.... but I still would love to see benchmarks because science. Also, as someone else pointed out, newer ARM cores improve a few of these aspects (out-of-order processing). But overall I think your post might settle a good chunk of the practicality problem with the Phoronix benches.
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:00 pm

Duct Tape Dude wrote:
Sargent Duck wrote: I guess one could look at it from the point of view of taking the engine from a SMART car and trying to scale it up to 1000hp. Not going to happen.

I raise you this: http://www.youtube.com/watch?v=UQnz6AyPjgs


Haha, I had heard of some people doing that a few years back. But still, if ARM = the SMART engine, then what you're doing is replacing the ARM architecture for something built more for performance.
Venii, vidii, vicii
Wii came, Wii saw , Wii conquered
Sargent Duck
Grand Gerbil Poohbah
Silver subscriber
 
 
Posts: 3068
Joined: Thu Mar 13, 2003 8:05 pm
Location: In my secret cave that has bats

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:02 pm

Duct Tape Dude wrote:You do realize you just described a wonderful massively parallel general processing unit, right? The blend of GPU and CPU... Who wouldn't want 500 cores working for them at 90W? Or even a fraction of that running noticeably faster? That sounds incredible to me.
Realize that if your single-threaded performance is bad, you still have a slow processor. Parallelism is great and all, but you can't disregard single-threaded performance. Look at what's happened to AMD with their FX chips. Even on embarrassingly parallel tasks like graphics, you're still held up by the slowest part of the whole operation.
Duct Tape Dude wrote:Netburst was a CISC arch with an insanely long pipeline (20something steps?) and ARM has never been limited by the ability to cool the chip. It's been limited by how little power they want to feed it. Two totally different design limitations since one was designed to sip power, the other was designed disregarding it.
CISC and RISC are meaningless terms these days; Intel chips since the P6 have been RISC internally. ARM chips absolutely are limited by the ability to cool the chip; have you not heard of the numerous throttling issues on high-end smartphones? (Also, Prescott had a 31-stage pipeline.) The point Sargent Duck makes is very apropos.
Duct Tape Dude wrote:I'm most curious as to why though... And by how much?
See my first quote-reply in this post.
i5-3570K @ 4.4 (NH-C14), 4x8GB DDR3-1866, GA-Z68MA-D3H-B2, ASUS GTXTITAN-6GD5, 128GB Vertex 4 / 2x60GB Vertex Plus R2 / 2x2TB Barracuda 7200.14 RAID0 / ANS-9010 (4x4GB), SST-DA1000 (PSU), 2x VS229H-P, 1x VG248QE, 1x MIMO 720F, Corsair Vengeance K90+M95
auxy
Gerbil Elite
 
Posts: 781
Joined: Sat Jan 19, 2013 4:25 pm
Location: the armpit of Texas

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:08 pm

Really, I just think that if an ISA designed for a programmable calculator can become the most profitable and highest overall performance in the world, there's no way ARM can't hypothetically do the same(Reality will determine what happens).
I.S.T.
Gerbil First Class
 
Posts: 178
Joined: Wed Sep 14, 2005 5:18 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:09 pm

Why?

Because that was never in the design docket for ARM in the first place. It's like asking a P4 to run at 5C over ambient.
It is one of the blessings of old friends that you can afford to be stupid with them. Ralph Waldo Emerson.
Captain Ned
Global Moderator
Gold subscriber
 
 
Posts: 20120
Joined: Wed Jan 16, 2002 7:00 pm
Location: Vermont, USA

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:10 pm

I should apologize, auxy: I didn't include you in the previous reply since there were so many good points to go through. But I made another one and here's this.

auxy wrote:Realize that if your single-threaded performance is bad, you still have a slow processor. Parallelism is great and all, but you can't disregard single-threaded performance. Look at what's happened to AMD with their FX chips. Even on embarrassingly parallel tasks like graphics, you're still held up by the slowest part of the whole operation.

I think you know what you're talking about but I'll go ahead and disagree with the last part for things like OpenCL or parallel compute-heavy jobs like mining--GPUs are generally regarded as faster than CPUs... though I suppose their TDPs are a little higher, too. The point is they have a bunch of weak units.

auxy wrote:CISC and RISC are meaningless terms these days; Intel chips since the P6 have been RISC internally.

Arguably, after decoding variable-width CISC instructions. I think x86 engineers would be a bit ticked if you summed up their current arches as just RISC.

auxy wrote:ARM chips absolutely are limited by the ability to cool the chip; have you not heard of the numerous throttling issues on high-end smartphones? (Also, Prescott had a 31-stage pipeline.) The point Sargent Duck makes is very apropos.

Actually, no, I haven't! I believe you though. Just wasn't aware of it.

auxy wrote:See my first quote-reply in this post.

That I did.
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:12 pm

Captain Ned wrote:Why?

Because that was never in the design docket for ARM in the first place. It's like asking a P4 to run at 5C over ambient.

Yes, that is true! But my point/question is what are the performance characteristics like at high TDPs? I think several others have made some quality detailed points supporting you though.
Duct Tape Dude
Gerbil First Class
Gold subscriber
 
 
Posts: 116
Joined: Thu May 02, 2013 12:37 pm

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:17 pm

Sargent Duck wrote:So from the way I see it, technically, nothing is stopping ARM from being a high performance, except that it'd be rubbish. To use the car analogy (first!), I guess one could look at it from the point of view of taking the engine from a SMART car and trying to scale it up to 1000hp. Not going to happen.


I think a more accurate analogy would be to take that SMART car engine and then adding a huge air intake to it when the intake isn't the bottleneck. It can only use whatever air it takes in if it is designed to utilize it. Similarly, as it stands today, most ARM CPUs aren't design to even make use of any additional TDP. If they were provided it, they wouldn't know what to do with it aside for exhaust it.

Another question to ask is: What are Intel/AMD processors doing that cause them to use more power and be so much faster? I'm going out on a limb here with this, but I think the answer is in how much work it's actually doing (work = heat). Generally speaking, each x86 cycle does a lot more work than an ARM's CPU cycle and a lot more efficiently. Until ARM CPUs begin to integrate all of the same (or better) branch predictors, instruction buffers/fetchers, optimizations, etc. as more developed x86 CPUs (and without stepping over patents), they'll continue to be lighter weight and slower.
bjm
Gerbil
Silver subscriber
 
 
Posts: 39
Joined: Sat Sep 01, 2007 1:54 am

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:32 pm

Duct Tape Dude wrote:
auxy wrote:Realize that if your single-threaded performance is bad, you still have a slow processor. Parallelism is great and all, but you can't disregard single-threaded performance. Look at what's happened to AMD with their FX chips. Even on embarrassingly parallel tasks like graphics, you're still held up by the slowest part of the whole operation.
I think you know what you're talking about but I'll go ahead and disagree with the last part for things like OpenCL or parallel compute-heavy jobs like mining--GPUs are generally regarded as faster than CPUs... though I suppose their TDPs are a little higher, too. The point is they have a bunch of weak units.
GPUs are obviously faster because they are simply much larger and faster chips. Compare:
  • Ivy Bridge at 77W -- 4 cores with 3 ALUs each at 3.4Ghz
  • GK104 at 195W -- 1536 ALUs at 1Ghz
I'll give you a moment to let that sink in.

Each "core" in a GPU is MUCH smaller and simpler than a modern x86-64 CPU core, and because of this, they're a lot less versatile. Obviously, for a parallel task, you want parallel processors, but games don't run merely on the GPU. GPUs are ASICs, or application-specific integrated circuits, they are not general-purpose processors and are very bad at that sort of computing. A game can do its general purpose processing on the CPU, and use the GPU to accelerate the dense-but-simple math workload of the graphics and physics processing.

bjm wrote: I think the answer is in how much work it's actually doing (work = heat). Generally speaking, each x86 cycle does a lot more work than an ARM's CPU cycle and a lot more efficiently. Until ARM CPUs begin to integrate all of the same (or better) branch predictors, instruction buffers/fetchers, optimizations, etc. as more developed x86 CPUs (and without stepping over patents), they'll continue to be lighter weight and slower.
This is correct.
i5-3570K @ 4.4 (NH-C14), 4x8GB DDR3-1866, GA-Z68MA-D3H-B2, ASUS GTXTITAN-6GD5, 128GB Vertex 4 / 2x60GB Vertex Plus R2 / 2x2TB Barracuda 7200.14 RAID0 / ANS-9010 (4x4GB), SST-DA1000 (PSU), 2x VS229H-P, 1x VG248QE, 1x MIMO 720F, Corsair Vengeance K90+M95
auxy
Gerbil Elite
 
Posts: 781
Joined: Sat Jan 19, 2013 4:25 pm
Location: the armpit of Texas

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:36 pm

Duct Tape Dude wrote:Yes, that is true! But my point/question is what are the performance characteristics like at high TDPs?

As others have pointed out, at the die size of an ARM processor there's simply no thermal margin. ARM was designed to run small and cool. Trying to force 90W out of an ARM die is simply impossible absent cryogenic cooling.
It is one of the blessings of old friends that you can afford to be stupid with them. Ralph Waldo Emerson.
Captain Ned
Global Moderator
Gold subscriber
 
 
Posts: 20120
Joined: Wed Jan 16, 2002 7:00 pm
Location: Vermont, USA

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:45 pm

Duct Tape Dude wrote:Why are there no high-power ARM CPUs?

because to get true high-power, you need an arm and a leg?
You don't have to feel safe to feel unafraid.
danny e.
Maximum Gerbil
Gold subscriber
 
 
Posts: 4371
Joined: Thu Apr 25, 2002 3:09 pm
Location: Indonesia/Nebraska/Wisconsin

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:50 pm

danny e. wrote:
Duct Tape Dude wrote:Why are there no high-power ARM CPUs?

because to get true high-power, you need an arm and a leg?

Oh no, you didn't!
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9934
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 5:51 pm

morphine wrote: Oh no, you didn't!

danny e. = "the noise in the TR signal to noise ratio". :D
You don't have to feel safe to feel unafraid.
danny e.
Maximum Gerbil
Gold subscriber
 
 
Posts: 4371
Joined: Thu Apr 25, 2002 3:09 pm
Location: Indonesia/Nebraska/Wisconsin

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 6:15 pm

If you need some power consumption numbers-- albeit in current tablet power envelopes:
http://www.anandtech.com/show/6536/arm- ... l-showdown

Task energy on the ARM A15 doesn't look all that great, it actually looks awful in some areas vs. ARM A9 and Intel Atom.

However, since A15 has somewhat different design targets in mind, it will probably behave differently at higher powers-- I am not qualified to speculate there...
continuum
Gerbil First Class
Gold subscriber
 
 
Posts: 140
Joined: Mon Jun 09, 2003 1:42 am
Location: California

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 6:15 pm

And noone has even mentioned Project Denver?

The Internet claims it's due in 2015

(I don't know this site, just went through some links after a search http://www.brightsideofnews.com/news/20 ... ategy.aspx )
Meow.
Savyg
Gerbil Elite
Silver subscriber
 
 
Posts: 613
Joined: Thu Aug 26, 2004 6:18 am
Location: Between desert and tundra

Re: Why are there no high-power ARM CPUs?

Postposted on Thu May 02, 2013 6:32 pm

Savyg wrote:And noone has even mentioned Project Denver?
A lot of people think "Project Denver" is vaporware -- and nobody even really knows what the purpose is. I'm still on the fence as to whether it will ever become a real product, but look at this:
Wikipedia wrote:The existence of Project Denver was revealed at the 2011 Consumer Electronics Show. In a March 4, 2011 Q&A article CEO Jen-Hsun Huang revealed that Project Denver is a five year 64-bit ARM architecture CPU development on which hundreds of engineers had already worked for three and half years and which also has 32-bit ARM architecture backward compatibility
That would mean it was targeted for 2013, but that's clearly not going to happen; Maxwell won't even be ready this year. I personally still think "Project Denver" is just Nvidia getting back at Intel for years of coarse relationships between the two.

continuum wrote:Task energy on the ARM A15 doesn't look all that great, it actually looks awful in some areas vs. ARM A9 and Intel Atom.
Cortex-A15 is a joke -- the performance per watt is bad, the power consumption is not good enough for phones or even tablets, but the raw performance isn't good enough to compete with AMD or Intel designs. So where is the market for it?
i5-3570K @ 4.4 (NH-C14), 4x8GB DDR3-1866, GA-Z68MA-D3H-B2, ASUS GTXTITAN-6GD5, 128GB Vertex 4 / 2x60GB Vertex Plus R2 / 2x2TB Barracuda 7200.14 RAID0 / ANS-9010 (4x4GB), SST-DA1000 (PSU), 2x VS229H-P, 1x VG248QE, 1x MIMO 720F, Corsair Vengeance K90+M95
auxy
Gerbil Elite
 
Posts: 781
Joined: Sat Jan 19, 2013 4:25 pm
Location: the armpit of Texas

Next

Return to Processors

Who is online

Users browsing this forum: No registered users and 1 guest