AMD gave few details on Hierfalcon. On the face of it, it's hard to see how it will differentiate itself over the Intel Rangley chip shipping today, a 2-8 core 64-bit Atom with cryto and other accelerators for comms systems--let alone what I woulod expect to see in this timeframe from TI, Freescale and others.
In short, I've yet top be convinced ARM is any kind of savior for AMD.
Actually till now most of the ARM comms players have been smaller fry. bigger player can grab market share through non technical advantages like easy of availability, larger user community, lower MOQ, no NDA requirement. Try getting datasheets for comm processors from Cavium or Broadcom as a small oem !
I often ended up with Freescale even though there were better parts around simply because You could whip out a credit card and order dev kits or samples.
On a tech note, not sure what AMD gains by going ARM. probably marketing and SW Eco system reasons. The x86 ISA decode penalty is a non issue these days.
If AMD can move these parts through commodity channels and make their availability ubiquitous, they have a good chance of a 10 -20% market share. But execution s everything and that seems a big ask for AMD these days. Would also help if they went open with their freedom interconnect.
What would also be interesting is repurposing the GCN for packet processing, ala IXP/Netronome
Purely on a technical basis , AMD's ARM choice makes sense. the v8 ISA is easier to implement than x86 and at smaller sizes efficiency wins out.
Is AMD planning to use Arm's cores or going custom ? if custom, I wonder
how much of the x86 code can be reused ? components like bus, interconnect, bottom part of the execution pipeline, SIMD components should be reusable to a large degree. So should branch code.
Maybe it is not that big a deal for them to switch to ARM.
@Junko - The only logical conclusion to that question is that AMD feels that ARM provides something that x86 cannot. AMD is a member of the exclusive club of 2 that can make x86 processors so for it to give up all the advantage of the established ecosystem around x86 to move to ARM is a testament that they feel they can compete better with ARM CPUs.
They did try low power CPUs with bobcat but probably found the ARM CPUs to be much better at the low power level. At the high end where it is performance at any cost they will still bring out x86 cpus.
@Tarra, yes, in theory, AMD giving up on "all the advantage of the established ecosystem around x86" does seem to indicate that they must have found something more compelling in ARM. The question is "what"?
@Junko, That is the multi-million dollar question. Probably the development cost of a new x86 internal core did not result in an advantage in terms of performance/cost/power of an off the shelf ARM core. I am sure the AMD brass did their homework before making the switch.
This does validate what the ARM camp has been claiming in terms of architectural simplicity of ARMv8 architecture and its inherent advantage in building low power processors.
My guess is that AMD's plan is the following:
- Continue to push x86 based processors at the high end of the performance arena in the server and enterprise
- Push ARM based SoCs into markets where AMD has not had much success to begin with - this includes the embedded market as validated by the hiero-falcon announcement and low power servers using the freedom-fabric.
- Integrate their ARM and ATI graphics technology to gain a foothold in the lucrative mobile and eventually ultra-mobile segments.
Reasonable thinking in my mind as x86 has not had success in the lower two arenas inspite of Intel trying for so long.
Not sure how AMD will differentiate with other ARM providers. Rick, you did not mention X-Gene from Applied. It seems they have similar features in term of core count and interfaces and have silicon today.
There are also the existing PPC and MIPS players in this crowded market, do you think ARM or x86 will gain market share there?
Sometimes I do forget that we still have AMD. This company is truly trying. My problem has been that AMD has good technology but poor marketing unit. They make nice products but they have a very bad branding strategy.
I wonder if choice of processor cores will be any more important than choice of other components on the SoC. Rangeley does not have the 10G Ethernet backplane interface. On the other hand, Rangeley does have hardware encryption capable of 10Gbps throughput, 5 Gbps in the 15W part. Rangley also has VT-x and no mention was made of hardware virtualization features of the AMD chip's cores. These things could be more important than the processor core, paticularly for software-defined networking.
@Jay: Good points on how AMD's Hierofalcon is ahead of Intel's Rangeley in some ways --and needs to be since it will have to compete with a Rageley 2 late next year.
And agreed, as long as the core roughly meets the perf/watt needs and more importnatly is the ISA the OEM/user sw is written for, it's fine. Much of the differentiation is increasingly in SoC peripherals.
I often ended up with Freescale even though there were better parts around simply because You could whip out a credit card and order dev kits or samples. On a tech note, not sure what AMD gains by going ARM. probably marketing and SW Eco system reasons. The x86 ISA decode penalty is a non issue these days. If AMD can move these parts through commodity channels and make their availability ubiquitous, they have a good chance of a 10 -20% market share. moncler jackets But execution s everything and that seems a big ask for AMD these days. Would also help if they went open with their freedom interconnect.