AMD announced its first ARM-based server chip yesterday called Seattle. Does this mean Intel is in trouble?
AMD announced today the first of its ARM processors: Seattle. Slated to ship in the second half of 2014, Seattle will have eight ARM Cortex-A57 cores clocked at 2GHz with 16-core versions expected after that. AMD claims that these chips can deliver four times the performance of its Opteron X line.
Seattle will be a 64 bit chip based on ARMv8, will have support up to 128GB of RAM, and will come equipped with AMD’s “Freedom Fabric” protocol, which clusters low-powered CPUs together to make them more efficient.
Still life in old cores?
AMD also announced two new x86 server chips: Berlin and Warsaw.
Berlin is the successor to the Opteron 3300. It appears to be a server intended version of the Kaveri APU, running on the Steamroller microarchitecture with support for GCN.
One of Berlin’s competitive advantages, according to AMD, is the inclusion of heterogeneous Uniform Memory Access (hUMA) technology that helps bridge the gap between GPU and CPU. While eliminating the bottlenecks of GPGPU would be an incredible advantage in server computing, for it to work effectively it would require industry wide changes to commonly used programming models and compilers to work effectively. hUMA might present an advantage in the gaming market, where developers are more agnostic and flexible but this might not happen in the server coding world.
Warsaw is intended to replace AMD’s Opteron 4300 and 6300 series and is scheduled to ship in the first quarter of 2014. In many ways it’s only an incremental update to the last generation of Opteron Piledriver chips. The only changes of note it offers are a 32nm fab, and a better power draw. What’s interesting is what Warsaw lacks: why not a Steamroller core? A 28nm fabrication process? Since Warsaw is based on the aging Piledriver core, it says a lot about the future of big cores, or the lack thereof, from AMD.
Stealing the bread
It’s no secret that Intel’s server business is a real breadwinner for the company, especially in the age of mobile devices requiring backend Internet services (“A Xeon behind every three smartphones”). In the first quarter of 2013, the company’s data center group pulled in $2.3 billion in revenue. While Intel’s PC client group shrinks, it’s data center group is growing; one lost 6 percent of its revenue year-over-year while the other added 7.5 percent more. In light of a declining PC market, it’s one of the few great hopes for Intel as it has proven technology that largely dominates the high margin business.
Now, AMD is looking to shake things up.
In 2012 AMD purchased micro server maker SeaMicro. SeaMicro doesn’t make servers with a few high-powered chips like the Intel Xeon, instead they build serves with hundreds of low powered ARM chips. This means the servers have a lower power draw and have inherent scalability. There’s a plethora of uses for such a type of server, including bringing cloud computing to the next billion users in the developing world.
Servers are an untapped market for ARM. If it can successfully jump the gap into this field via chips like AMD’s Seattle and SeaMicro servers, x86 will be in some serious trouble.