SAN JOSE, Calif. – Breaking with its long history as an x86 PC chip maker, Advanced Micro Devices is formally throwing its hat into the ring of ARM server chip makers. Just how far the chip maker goes down the ARM road, how fast--and how successfully--remains unclear.
The long-time x86 CPU maker will design a 64-bit ARM SoC based on ARM’s own V8-compliant core. It will ship both as a merchant chip and in its own SeaMicro branded systems in 2014.
AMD is not designing its own ARM cores initially. It would not comment on whether it will take an ARM architecture license to design custom ARM cores in the future. The decision to use ARM’s own core design—the Atlas 64-bit core that is a successor to the ARM A15—is apparently driven by AMD’s time-to-market goals.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
The effort marries AMD’s deep server expertise with Sea Micro’s unique interconnect fabric to compete in a market for 64-bit ARM server SoCs that is rapidly getting crowded. Whether the products will help AMD pull out of its slump in the x86 sector remains to be seen.
AMD last week announced the layoff of 15 percent of its employees amid 10 percent revenue declines, expanding its focus on embedded systems. Given AMD's dire situation, it’s surprising the company is not moving even more aggressively into ARM-based servers and custom ARM core designs.
Lisa Su implies but does not specifically say AMD's 2014 chip will embed the ARM cores AND the SeaMicro fabric.
She says nothing to clarify if this is a system only or merchant chip, too. But the implication is this is also a merchant chip.
From hot-off-the presses AMD press release:
The first ARM technology-based AMD Opteron processor will integrate the AMD SeaMicro Freedom supercompute fabric, the industry's premier high-performance fabric.
Seamicro fabric or servers are used by sub 1% of market today, not sure this architecture is extensible to heavy workload's such as DB systems, in Data Center, majority of there use case is around Web servers. If one wants to do Web servers they can use Multicore ARM server SOC from Calxeda and others coming to the show if market is going to grow types.
So am not sure what advantage AMD has over others?
I think they are done! they cannot kick the bucket to 2014, need to pay bills now.
This is Rory Read's hail mary pass. ARM does make sense for AMD and maybe even in servers eventually but not in the immediate future.
I thought they would likely use the strength of their GPU/APU to come out with low power ARM processors for the mobile/ultramobile markets. That is certainly a huge market.
On the other hand AMD is laying off 15% of its work-force and now designing x86 AND ARM servers. It will be interesting to see how they train their sales force to sell both ARM and x86 processors into servers.
x86 is not their problem, the issue for AMD has been been in execution - remember the performance lead they had with opteron and the big delays in bringing out the bulldozer designs by which time sandy bridge and ivy bridge leap-frogged them.
I don't see how this would change with the same team in place. What they need is a new design team to work on this in isolation if this has to succeed.
I was a young design engineer in AMD's analog products group when founder Jim Giles was still in it. They deleted that product area, and I left.
I rejoined AMD's networking group to design and release a compliant IEEE 802.3 10BASE-2/5 tap transceiver, and supervised the design and release of a 10BASE-T predecessor StarLAN transceiver.
I left AMD again a couple years before they eliminated the networking products division.
I hope AMD finds a way to focus on something other than x86 processors. ARM could be one of those ways.
I'd hate to see them eliminate any more product areas. There aren't too many of them left for the company.
This is like testing the waters before taking the plunge. Perhaps they are not too sure of what to make of the ARM 64bit in server domain, but given their excellence in interconnect fabric and traditional experience in server workloads optimization they should be able to to make it take off. But it all depends on timing and positioning and if they get it right then it could be the proverbial light at the end of the tunnel they are looking for.
Could someone clarify the use of terminology a little? Does the term "SoC" as used here just reflect the integration on a single chip of IP that includes more than one vendor, ie ARM plus someone else's fabric IP? It just seems confusing since if the fabric is part of the "system" then it's not really usable as a system until you have a least two of these talking to each other, then it's not really a "SoC" if you see what I mean. Forgive me if I'm nitpicking a bit here!