[Hexus] ARM compares A72 vs Intel Broadwell-Y

[simboss]

The A8 already cost Apple 37$ just in pure manufactoring without R&D. In terms of Samsungs SoC in the S6 its 29.5$, again without R&D. Just add a 50% margin on top and see where you land.

I have always wondered what is included and what's not in these estimations, do you have a source that explains it?

From this description: http://press.ihs.com/press-release/...-build-cheaper-buy-comparable-apple-iphone-6- It is hard to tell, but I would say it would be very inaccurate to give a pure manufacturing cost for the CPU, but give the price paid to QC when buying from an external supplier the year before: http://press.ihs.com/press-release/...ries-astronomical-bill-materials-ihs-teardown

ARM isnt really so cheap as its often portraited when performance matters. It also shows on Qualcomm for example.

Based on the above, I don't think that's true, but happy to be proven wrong.

No, you can still charge 100s and 1000s of dollars for CPU design just like you always have been.

You can, but you need to sell them to businesses, the consumer mostly don't care anymore (*).
This is a very different market, both in size and requirements, Intel is doing very well in this market, but loosing the volume advantage of getting both the consumer and the enterprise market at the same time makes a huge difference.
This is exactly where all the other players are at the moment (IBM, Oracle), and they are not doing that well.

(*) The saving grace for Intel is the "does it run windows?" question which is still at the top of the consumer's list.

The reality is that for 90+% of the devices containing ARM CPUs it doesn't matter what is in them. They are only ARM because ARM is cheaper than everything else. Many of the devices are using cores that are almost free at this point. They are using cores that were designed a decade or more ago.

And yet, ARM makes a healthy profit and margin out of them.
This is actually supporting my point, people care less and less about the CPU.

That's actually a market that ARM is very much at risk of losing in the future. Cause when the value of the CPU isn't there, you use the cheapest thing possible. As people start of open source RTL for RISC-V like designs and the tool flows start to support them, there will be little point in actually paying any licensing fees. And that part of the ARM market share will evaporate. And ARM would be massive fools if they don't see it coming.

This is true, although the "let's make HW open source" is based on the assumption that it worked for SW.
It worked on SW in big part because the HW guys were developing the SW, so the "free" SW was partly paid for in the chip price.
Have a look at the top contributors to the Linux kernel:
http://www.linuxfoundation.org/news-media/infographics/who-writes-linux-2015

ARM does not charge that much on each chip, it helps to keep the incentive to move to a cheaper product low, as you are sharing the cost with almost everyone in the industry, it is actually negligible (and in that respect pretty close to the Open-Source business model), whereas going open-source on your own would cost much more.
A concerted and well executed move from all the industry players to move to open-source HW would be a major threat to ARM, but we are still pretty far from it, and the fact that you can't fix bug in HW as you can in SW is a pretty big barrier here, you need someone to take the liability for the chip correctness.

 

[Nothingness]

No, you can still charge 100s and 1000s of dollars for CPU design just like you always have been. The reality is that for 90+% of the devices containing ARM CPUs it doesn't matter what is in them. They are only ARM because ARM is cheaper than everything else. Many of the devices are using cores that are almost free at this point. They are using cores that were designed a decade or more ago.

MIPS is much cheaper than ARM and has been for years. Do you see many companies moving to MIPS?

That's actually a market that ARM is very much at risk of losing in the future. Cause when the value of the CPU isn't there, you use the cheapest thing possible. As people start of open source RTL for RISC-V like designs and the tool flows start to support them, there will be little point in actually paying any licensing fees. And that part of the ARM market share will evaporate. And ARM would be massive fools if they don't see it coming.

What EDA company will have and support flows for RISC-V? Where is the validation suite? The toolchain is stable? What about legacy software? Will RISC-V be small enough to target ARM Cortex-M chips?

 

[soresu]

From the sound of it, RISC-V may be more suited for micro controllers than anything else. I seem to remember Exophase commenting on it lacking several common instructions that would make it less than efficient for more general purpose use cases.

Its highly possible that RISC-V could see use for IoT that have require very efficient core but only a constrained use in conjunction with larger devices.

 

[Systems analyst]

See
http://ir.arm.com/phoenix.zhtml?c=197211&p=irol-presentations
TechCon 2014 presentation, page 6.
This shows projected and actual 16nmFF results, for the A53 and A57 cores.
It may be that ARM are basing their projections for the A72 on similar modelling, and have reason to believe that their modelling is valid.

See also page 8 of the same presentation; it shows some info on 10nm projections.

Seeing the progress of ARM cores:
2012 - first A15 powered device.
2014 - first A57 powered device.
2015 - 2016 - A72 powered devices expected,
is impressive.

 

[B-Riz]

Apple has a whole separate product stack for things based on ARM; iOS and iOS devices.

It is a walled garden, and they had to build everything so people could develop on it.

Guess what, x86 software written in 1980 can be run today on the newest, lowliest x86 SOC from AMD or Intel.

Keep telling us ARM is the future, its not; it as an appliance CPU.

Google, Facebook, Amazon and Apple run their back-end operations on x86 hardware and software, and will in the future; they may try out some ARM stuff, but it will never replace Intel's design and supply chain dominance.

A few good designs from ARM is not going to change 30+ years of established industry software / hardware.

 

[Nothingness]

From the sound of it, RISC-V may be more suited for micro controllers than anything else. I seem to remember Exophase commenting on it lacking several common instructions that would make it less than efficient for more general purpose use cases.

Its highly possible that RISC-V could see use for IoT that have require very efficient core but only a constrained use in conjunction with larger devices.

Unless RISC-V uses variable-length instruction, its instruction density will be worse than ARM Thumb used in Cortex-M and this matters a lot in that segment.

 

[Nothingness]

Guess what, x86 software written in 1980 can be run today on the newest, lowliest x86 SOC from AMD or Intel.

Keep telling us ARM is the future, its not; it as an appliance CPU.

Google, Facebook, Amazon and Apple run their back-end operations on x86 hardware and software, and will in the future; they may try out some ARM stuff, but it will never replace Intel's design and supply chain dominance.

A few good designs from ARM is not going to change 30+ years of established industry software / hardware.

You're perhaps too young to remember but the same was told of Intel back then.

 

[krumme]

You're perhaps too young to remember but the same was told of Intel back then.

We just have to go back to the end 90ties when x86 servers hit. Same story.

If anything the numbers speak for themselves also today. 4b loss on mobile in a year and Intel have gotten nowhere long term.

 

[B-Riz]

(Am too young to have been there for Intel's rise (born early 80's), only been around around for their established dominance)

A confluence of events occurred that has cemented x86 for now and the future.

Intel + MS for corporate and home desktops, and Intel / AMD + Linux for web stuffs.

While ARM works well for what it does; people are thinking it is going to repeat what x86 has already done.

I do not see a "killer app" for it now or in the future; Windows + Office (and PC Gaming) and cheap plentiful x86 CPU's established the PC as we know it, and people are not going to replace it with something probably more limited but powered by ARM.

 

[Khato]

Intel rose to dominance in the server market from the low end up. They didn't start out selling high margin Xeons, they started out selling pretty much the same thing that they sold to the consumer market. This worked because the competition wasn't really offering anything comparable, especially not on pricing. Effectively, Intel leveraged their consumer volume to get a foot hold in "low cost" servers and built it up from there.

How is ARM going to make headway? It's not really in the traditional server market actually. They're going to have far more luck competing against Intel in areas like telecoms/networking servers which work great with a larger number of slower cores and, more importantly, don't have software barriers of entry. But even there it's going to be difficult for ARM to get a foothold as Intel has been quite adept thus far at preemptively releasing Xeon products to negate potential threats from ARM. And the projected performance of A72 doesn't change that in the slightest - Intel's high margin big-core Xeon products will still be completely unaffected while many-small-core Xeon products can keep ARM boxed in.

 

[imported_ats]

Unless RISC-V uses variable-length instruction, its instruction density will be worse than ARM Thumb used in Cortex-M and this matters a lot in that segment.

It used to matter, these days, not so sure.

 

[imported_ats]

We just have to go back to the end 90ties when x86 servers hit. Same story.

If anything the numbers speak for themselves also today. 4b loss on mobile in a year and Intel have gotten nowhere long term.

If by end 90ties you mean mid to late 80s, sure...

 

[krumme]

(Am too young to have been there for Intel's rise (born early 80's), only been around around for their established dominance)

A confluence of events occurred that has cemented x86 for now and the future.

Intel + MS for corporate and home desktops, and Intel / AMD + Linux for web stuffs.

While ARM works well for what it does; people are thinking it is going to repeat what x86 has already done.

I do not see a "killer app" for it now or in the future; Windows + Office (and PC Gaming) and cheap plentiful x86 CPU's established the PC as we know it, and people are not going to replace it with something probably more limited but powered by ARM.

From what perspective are you talking. As a mobile or desktop user?

Its google play store that rules and is used today. X86 is already the marginalised on consumer side. Look at desktop sales and even notebooks.
And it happened within something like 4 years.

 

[krumme]

If by end 90ties you mean mid to late 80s, sure...

It happened in first of august 1998
But hey it was frased wrong. P pro brought the solid server cpu and by the late 90ties Intel had won the major part of the market.

 

[krumme]

It used to matter, these days, not so sure.

And what are these days excactly? The tune of 4b vs dcg profit?

What myth is going on here. The one promoted by Anand and later demoted by Johan? - labeled:
"Intel engineer with a voltmeter visits. Explains ohms law and bust the x86 power myth"

 

[krumme]

Intel rose to dominance in the server market from the low end up. They didn't start out selling high margin Xeons, they started out selling pretty much the same thing that they sold to the consumer market. This worked because the competition wasn't really offering anything comparable, especially not on pricing. Effectively, Intel leveraged their consumer volume to get a foot hold in "low cost" servers and built it up from there.

How is ARM going to make headway? It's not really in the traditional server market actually. They're going to have far more luck competing against Intel in areas like telecoms/networking servers which work great with a larger number of slower cores and, more importantly, don't have software barriers of entry. But even there it's going to be difficult for ARM to get a foothold as Intel has been quite adept thus far at preemptively releasing Xeon products to negate potential threats from ARM. And the projected performance of A72 doesn't change that in the slightest - Intel's high margin big-core Xeon products will still be completely unaffected while many-small-core Xeon products can keep ARM boxed in.

I agree in this view. But the change is perhaps not comming because of arm but eg because of gpu tech, hsa and the use of new type of programming that can do parallelization. Just an example.
The point is history does not repeat in the same way. We only know one thing fore sure. It will not be like the past.

 

[soresu]

Extremetech have the full details on Helio X20
http://www.extremetech.com/computing/205066-eight-isnt-enough-mediatek-plans-10-core-mobile-soc

Its not 16nm, but with T880 MP4 it should come close to Exynos 7420 in GPU and beat it handily for single threaded CPU tasks.

It could be interesting to see how the power consumption differs between the two chips given one uses more efficient cores on 28nm, and one uses less efficient cores on 14nm.

 

[imported_ats]

And what are these days excactly? The tune of 4b vs dcg profit?

What myth is going on here. The one promoted by Anand and later demoted by Johan? - labeled:
"Intel engineer with a voltmeter visits. Explains ohms law and bust the x86 power myth"

You know, context might be useful for you. Esp when there is a nice quote included indicating exactly what was responded to. AKA, you seem to think thumb still matters in markets...

 

[imported_ats]

It happened in first of august 1998
But hey it was frased wrong. P pro brought the solid server cpu and by the late 90ties Intel had won the major part of the market.

Intel was into servers by the late 80s at a minimum. There were x86 servers of course sold substantially before that of course. A pretty good peg is probably 1986 which was the release point of Netware 2.0. And the rest was pretty much history.

 

[386DX]

It happened in first of august 1998
But hey it was frased wrong. P pro brought the solid server cpu and by the late 90ties Intel had won the major part of the market.

I remember those days at my co-op job. Was working with DEC Alpha's, Sun Sparc servers and SGI Onyx Workstations then one day we were sent a Pentium Pro server to try out and benchmark and it was pretty impressive at the time. I remember nobody expected it to be any good compared to the Alpha but it put up strong numbers but ultimately the Alpha's EV6 bus gave it the advantage for our workload.

 

[Idontcare]

I remember those days at my co-op job. Was working with DEC Alpha's, Sun Sparc servers and SGI Onyx Workstations then one day we were sent a Pentium Pro server to try out and benchmark and it was pretty impressive at the time. I remember nobody expected it to be any good compared to the Alpha but it put up strong numbers but ultimately the Alpha's EV6 bus gave it the advantage for our workload.

DEC. Great product, excellent engineering. Piss poor executive team and BoD. How do you set yourself up to have it all but end up with nothing in the end?

History repeats on occasion I suppose.

 

[B-Riz]

From what perspective are you talking. As a mobile or desktop user?

Its google play store that rules and is used today. X86 is already the marginalised on consumer side. Look at desktop sales and even notebooks.
And it happened within something like 4 years.

Hmm, I have yet to spend on money in Google Play, but Steam, Origin, MS and pay what you want PC software companies have gotten money from me, I even donated to Wikipedia last year!

Just massive lulz about mobile toys being relevant and marginalizing x86.

They are the latest market, but they do not run anything, they just display what is created and served by x86 PC's / servers...

 

[Dribble]

How is ARM going to make headway?

In the same way they make headway elsewhere - Intel is expensive and restrictive. You have to buy the products they give you at the prices they force you to pay.

ARM is much cheaper and open - you can make whatever custom setup that you like and tweak those arm cores to be extra fast for your specific use cases, and you deciding pricing (ARM licenses cost very little).

 

[monstercameron]

In the same way they make headway elsewhere - Intel is expensive and restrictive. You have to buy the products they give you at the prices they force you to pay.

ARM is much cheaper and open - you can make whatever custom setup that you like and tweak those arm cores to be extra fast for your specific use cases, and you deciding pricing (ARM licenses cost very little).

That's not quite true, even Intel is doing semi custom xeons.

 

[B-Riz]

In the same way they make headway elsewhere - Intel is expensive and restrictive. You have to buy the products they give you at the prices they force you to pay.

ARM is much cheaper and open - you can make whatever custom setup that you like and tweak those arm cores to be extra fast for your specific use cases, and you deciding pricing (ARM licenses cost very little).

LOOOL

Backblaze would disagree that Intel is "expensive and restrictive".

https://www.backblaze.com/blog/storage-pod-4-5-tweaking-a-proven-design/

Standard boring office comp i3 cpu powers their storage servers; not Xeons or quad anything...

Seriously, do you even computer?