AMD demos 'Berlin' Opteron, world's first heterogeneous system architecture server chip
Forget CPUs. Forget GPUs. It's all compute, all the time
The company tells The Reg that the demo will consist of X2100 Series Opteron running a Linux environment based on the Fedora Project.
Berlin is not one of your daddy's Opterons, based as they have been on compute cores, but rather it's what AMD calls an APU – accelerated processing unit. This mashup of CPU and GPU cores work together in a heterogeneous system architecture (HSA) with a shared memory.
The compute and graphics cores in the HSA-enabled X2100 work together by means of heterogeneous unified memory architecture (hUMA) and heterogeneous queueing (hQ), which enable threads to run simultaneously and independently on all of the CPU and GPU cores, and sharing the same memory space so that the CPU doesn't have to get involved with feeding the GPU.
AMD's first HSA APU was Kaveri, released this January. At a briefing accompanying that release, AMD CTO Joe Macri said that in the HSA world, it was time to toss out terms like CPU and GPU. "We're going to need to create a new term, because HSA has really changed what can be done on the graphics part of the die." His suggestion: call them both compute cores.
Kaveri, however, is aimed at notebooks and desktops; Berlin is a server-level part, and as such is designed not for the video, games, and other consumer-level usage models at which Kaveri is intended to excel. And that's where the server-software developers come in.
"As servers adapt to new and evolving workloads," said AMD's server unit general manager Suresh Gopalakrishnan, "it's critical that the software ecosystem support the requirements of these new workloads. "We are actively engaged with a broad set of partners in the data center software community who are bringing to market the software infrastructure to seamlessly enable x86 APU based servers."
After all, what good is an AMD hUMA hQ HSA APU without devs to give it some TLC when it ships later this year? ®