Article: AMD's Mobile Strategy
By: Daniel Bizo (fejenagy.delete@this.gmail.com), December 16, 2011 7:12 am
Room: Moderated Discussions
Wilco (Wilco.Dijkstra@ntlworld.com) on 12/15/11 wrote:
---------------------------
>Daniel Bizo (fejenagy@gmail.com) on 12/15/11 wrote:
>---------------------------
>>Mark Pillsbury (no_spam@gmail.com) on 12/14/11 wrote:
>>---------------------------
>>>I liked your thoughtful analysis but I respectfully disagree with your opinion that
>>>"From a technical perspective, adopting ARM for SoCs makes little sense. It utterly
>>>eliminates one of AMD's core competencies, namely their x86 expertise." I think
>>>ARM is a better technical solution for tablets and netbooks/ultrabooks/notebooks
>>>because power consumption is critical in these markets. A successful company has
>>>to provide what the market needs even if it is not the sweet spot of their core
>>>competency. Otherwise, it is like being the guy who drops his keys in a dark parking
>>>lot and looks for them under a light instead of where he dropped them.
>>
>>There are folks here at RWT who know and can explain this stuff a lot better than
>>me, including David, but the bottom line is that in real life, energy efficiency
>>and power consumption has very little to do with the instruction set architecture.
>>This is because over the last decade the microarchitectural complexity AND integration
>>level of microchips increased in such a big way that the impact of the ISA on device power and efficiency is marginal.
>
>I disagree. The impact on the high-end is smaller nowadays, eventhough it remains
>non-trivial. Nobody would claim that 4-way x86 decode is easy! It has taken a very
>long time for x86 to get there, when 3rd generation OoO ARM is already going to be 4-way.
>
>However at the low end, the x86-penalty is very significant, and this will remain
>so despite increasing complexity and cheaper transistors. The hard fact is that
>you need more transistors to implement an x86 core to get equivalent performance,
>and that means higher power consumption.
>
>>Even if ARM had any advantages over x86 due to its RISC type ISA that enables a
>>cleaner, simpler and more efficient front-end design, architectural design choices,
>>resources on implementation, and available process technology are all much bigger
>>factors on their own, not to speak combined.
>
>If that is the case then why is an Atom core more than double the size of a Cortex-A9
>(could be quadruple, I can't find my old post on this), and ends up slower clock
>for clock, despite Intel's process advantage?
>
>Bobcat is even larger of course, but at least it does beat the A9, although it
>needs significantly more power than Atom to get there. In this case the processes used are the same as for ARM SoCs.
>
>So unless we assume both Intel and AMD are totally incompetent in low power design, the main difference is the ISA.
>
>Wilco
Nobody said that efficient superscalar x86 decode is easy, that wasn't the argument. You also completley missed that the argument didn't said that was no consequence of x86 on design challenges, complexity or power consumption. The argument, however, is, that all those handicaps don't matter any more, because the weight they carry in the overall big picture is subtle at most.
Sorry for being redundant, but it seems you didn't get it for the first take. Today's and future mobile/tablet SoCs are integrating almost all the functions on a single piece of silicon -- GPU, video, audio, air, ODMC, storage IO, etc. As a result, CPU area and power consumption is a fraction of the total device area and power budget. Perhaps even most importantly, these systems typically spend most of their time is low performance or sleep states.
If you cared to do research, you would see that CPU power is typically below 20% of total average system power. You would also see that high-performance ARM designs DO consume a LOT of power running under high-activity CPU code. You would also see that overall average device power is dominated by air AND there are massive differences across implementations in radio efficiency, when it comes to phones. In smartphones, visualisation is a major factor due to 3D GUIs, video and other rich content, and massive displays.
Let's say, given all other things being equal*, ARM CPU consumes 20% less power than x86 for the same, low performance level, just because it's a RISC ISA -- I really believe I'm being very generous. That means roughly 4 percent longer average battery life on a system level, or less. Do you really think that's something that will prevent Intel and AMD to gain traction? Is that something they cannot address?
*And about that all things being equal. It's never true. Design teams differ in size, competence sweet spots, financial and technical resources, methodologies. If there is something that will pose as a major challenge for AMD to reach down to really low power levels, hell it's not going to be the complexity of x86, that's for sure.
Intel needs time and more design cycles to push its low power x86 implementation to those territories. It doesn't happen overnight, their latest 32nm LP SoC is only their 2nd iteration targeted at mobile phones.
---------------------------
>Daniel Bizo (fejenagy@gmail.com) on 12/15/11 wrote:
>---------------------------
>>Mark Pillsbury (no_spam@gmail.com) on 12/14/11 wrote:
>>---------------------------
>>>I liked your thoughtful analysis but I respectfully disagree with your opinion that
>>>"From a technical perspective, adopting ARM for SoCs makes little sense. It utterly
>>>eliminates one of AMD's core competencies, namely their x86 expertise." I think
>>>ARM is a better technical solution for tablets and netbooks/ultrabooks/notebooks
>>>because power consumption is critical in these markets. A successful company has
>>>to provide what the market needs even if it is not the sweet spot of their core
>>>competency. Otherwise, it is like being the guy who drops his keys in a dark parking
>>>lot and looks for them under a light instead of where he dropped them.
>>
>>There are folks here at RWT who know and can explain this stuff a lot better than
>>me, including David, but the bottom line is that in real life, energy efficiency
>>and power consumption has very little to do with the instruction set architecture.
>>This is because over the last decade the microarchitectural complexity AND integration
>>level of microchips increased in such a big way that the impact of the ISA on device power and efficiency is marginal.
>
>I disagree. The impact on the high-end is smaller nowadays, eventhough it remains
>non-trivial. Nobody would claim that 4-way x86 decode is easy! It has taken a very
>long time for x86 to get there, when 3rd generation OoO ARM is already going to be 4-way.
>
>However at the low end, the x86-penalty is very significant, and this will remain
>so despite increasing complexity and cheaper transistors. The hard fact is that
>you need more transistors to implement an x86 core to get equivalent performance,
>and that means higher power consumption.
>
>>Even if ARM had any advantages over x86 due to its RISC type ISA that enables a
>>cleaner, simpler and more efficient front-end design, architectural design choices,
>>resources on implementation, and available process technology are all much bigger
>>factors on their own, not to speak combined.
>
>If that is the case then why is an Atom core more than double the size of a Cortex-A9
>(could be quadruple, I can't find my old post on this), and ends up slower clock
>for clock, despite Intel's process advantage?
>
>Bobcat is even larger of course, but at least it does beat the A9, although it
>needs significantly more power than Atom to get there. In this case the processes used are the same as for ARM SoCs.
>
>So unless we assume both Intel and AMD are totally incompetent in low power design, the main difference is the ISA.
>
>Wilco
Nobody said that efficient superscalar x86 decode is easy, that wasn't the argument. You also completley missed that the argument didn't said that was no consequence of x86 on design challenges, complexity or power consumption. The argument, however, is, that all those handicaps don't matter any more, because the weight they carry in the overall big picture is subtle at most.
Sorry for being redundant, but it seems you didn't get it for the first take. Today's and future mobile/tablet SoCs are integrating almost all the functions on a single piece of silicon -- GPU, video, audio, air, ODMC, storage IO, etc. As a result, CPU area and power consumption is a fraction of the total device area and power budget. Perhaps even most importantly, these systems typically spend most of their time is low performance or sleep states.
If you cared to do research, you would see that CPU power is typically below 20% of total average system power. You would also see that high-performance ARM designs DO consume a LOT of power running under high-activity CPU code. You would also see that overall average device power is dominated by air AND there are massive differences across implementations in radio efficiency, when it comes to phones. In smartphones, visualisation is a major factor due to 3D GUIs, video and other rich content, and massive displays.
Let's say, given all other things being equal*, ARM CPU consumes 20% less power than x86 for the same, low performance level, just because it's a RISC ISA -- I really believe I'm being very generous. That means roughly 4 percent longer average battery life on a system level, or less. Do you really think that's something that will prevent Intel and AMD to gain traction? Is that something they cannot address?
*And about that all things being equal. It's never true. Design teams differ in size, competence sweet spots, financial and technical resources, methodologies. If there is something that will pose as a major challenge for AMD to reach down to really low power levels, hell it's not going to be the complexity of x86, that's for sure.
Intel needs time and more design cycles to push its low power x86 implementation to those territories. It doesn't happen overnight, their latest 32nm LP SoC is only their 2nd iteration targeted at mobile phones.