There’s another ugly bit. AMD hopes to differentiate its 64-bit ARM
server SoC with its time-tested SeaMicro fabric. The so-called Freedom
Fabric has been working in products for more than a year, far ahead of
ARM server SoC rivals Calxeda and Applied Micro. It can be used not only
for connecting chips between boards but for connecting between chassis
for data storage.
But ask Jay Parikh, Facebook's vice president
of infrastructure, what he thinks of these hot fabrics and he’ll tell
you he’s cold on them. “Proprietary fabrics are not something we are
interested in,” he said in conversation after a panel session at AMD’s
ARM server event.
I have talked to Facebook engineers who said they want to use nothing
but standard interconnects like PCI Express in their servers. They don’t
even want Ethernet on server motherboards because it only makes it
harder to upgrade rapidly changing CPUs.
Today, Facebook wants
2.5- to 2.7-GHz processors for its Web tier, which is just one of
hundreds of workloads it maintains. That’s a fast pace for ARM server
SoCs, but not out of range. Parikh said Facebook will not consider any
32-bit ARM designs—its software is all 64-bit--but left a door open
about possible 64-bit ARM chips.
In the end, AMD took a historic
step outside the x86 world today. But it leads down a pretty dark and
twisting path for a company that isn’t packed for a long journey.
AMD has announced that they are launching a new ARM Cortex-A57 64bit ARMv8 Processor in 2014, targetted for the servers market.
ARM is getting hot lately, and they do compete with INTEL on server-side thus, it is not a bad strategy that AMD wants to jump in and ride the waves of ARM. If they have a good execution plan, they can make it work.
Designing their own cores does make sense for them - otherwise they do not have any differentiation. I doubt the one-size-fits-all cores from ARM can address all the markets. I would suspect that ARM's own cores would likely want to retain dominance in the mobile space and are targetted to those market first. This is not to say that they cannot be used in servers but a dedicated server design would compete better.
Based on the news coming out of the ARM camp, it seems to validate this - almost all of them - AppliedMicro, Cavium, Qualcomm, nVidia are indicating they have their own core designs.
AMD does have decades of experience designing CPUs so that is their strength - the question is that of execution and focus when trying to ride two horses at once.
AMD is unique in that in can do both x86 and AMD, but as analyst Nathan Brookwood pointed out to me over lunch today, there's no great advantage to that.
Also, Nathan said he thinks there's no great advantage to designing your own custom core either given ARM is doing a pretty good and timely job covering the waterfront with its own designs.
Nvidia Tegra folks say Qualcomm, for example, went to a lot of trouble to design its own ARM cores but is not getting a huge technical advantage with them.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.