Article: AMD's Mobile Strategy
By: Paul A. Clayton (paaronclayton.delete@this.gmail.com), December 15, 2011 9:14 pm
Room: Moderated Discussions
Exophase (exophase@gmail.com) on 12/15/11 wrote:
---------------------------
[snip]
>Just the same, it's silly to think that ARM64, an
>architecture, won't be available to profiles as low power
>as ARMv7a is today, because:
>
>a) ARM64 takes LESS resources to decode than ARMv7,
>especially Thumb-2, so there's absolutely no reason to
>believe it'll be limited to heavier weight CPUs
Well, heavier than M-series. :-)
>b) And the ISA's inclusion of 32-bit ops for power saving
>clearly shows that it doesn't want implementations to be
>at a power disadvantage vs Cortex-A series
I thought that was kind of neat and perhaps not
unreasonable with 32-bit fixed-length instructions (i.e.,
adequate space for opcodes).
>c) ARM's strategy for servers is clearly emphasizing
>perf/W per core and is pointless without an advantage here
While the ISA may provide some power-efficiency features
(like 'b'), for greater power efficiency with an
inferior process technology and less design resources, I
suspect that ARM will design for a lower performance point
that Intel is less inclined to pursue aggressively.
(IMO, Intel has not handled Atom as a server well.)
The freedom to innovate without fear of cannibalizing more
profitable market segments may be ARM's greatest strength
in servers.
>d) Anyone can see that with the way tablets are
>progressing 64-bit will be strongly desirable in only a
>few years, possibly as few as two or three
If persistent memory technologies prove more scalable than
DRAM in the near future, this might further push for
support of larger memories. (With zero-power memory,
capacity can be increased without sacrificing battery life.)
---------------------------
[snip]
>Just the same, it's silly to think that ARM64, an
>architecture, won't be available to profiles as low power
>as ARMv7a is today, because:
>
>a) ARM64 takes LESS resources to decode than ARMv7,
>especially Thumb-2, so there's absolutely no reason to
>believe it'll be limited to heavier weight CPUs
Well, heavier than M-series. :-)
>b) And the ISA's inclusion of 32-bit ops for power saving
>clearly shows that it doesn't want implementations to be
>at a power disadvantage vs Cortex-A series
I thought that was kind of neat and perhaps not
unreasonable with 32-bit fixed-length instructions (i.e.,
adequate space for opcodes).
>c) ARM's strategy for servers is clearly emphasizing
>perf/W per core and is pointless without an advantage here
While the ISA may provide some power-efficiency features
(like 'b'), for greater power efficiency with an
inferior process technology and less design resources, I
suspect that ARM will design for a lower performance point
that Intel is less inclined to pursue aggressively.
(IMO, Intel has not handled Atom as a server well.)
The freedom to innovate without fear of cannibalizing more
profitable market segments may be ARM's greatest strength
in servers.
>d) Anyone can see that with the way tablets are
>progressing 64-bit will be strongly desirable in only a
>few years, possibly as few as two or three
If persistent memory technologies prove more scalable than
DRAM in the near future, this might further push for
support of larger memories. (With zero-power memory,
capacity can be increased without sacrificing battery life.)