Advanced Micro Devices’ first ARM chips code-named Seattle will not initially be compliant with specifications from the HSA Foundation, which can boost overall system performance, but support will be added in future generations of the server chips.
The HSA (Heterogeneous System Architecture) Foundation is developing standards and open-source programming tools that enable faster code execution via the joint computing power of CPUs, graphics processors, digital signal processors and other hardware.
HSA support has already been added to AMD’s x86 chips in PCs, but efforts are underway to bring HSA tools to Seattle’s successors, said Suresh Gopalakrishnan, corporate vice president and general manager of the server business unit at AMD.
“There’s a development cycle to it,” Gopalakrishnan said.
AMD announced the first Opteron A1100 Series ARM server chips late last month. The chips, which come in four- and eight-core variants, are initially being packaged in development kits. The chips are expected to be in servers by the end of this year.
But AMD’s upcoming x86 server chips, code-named Berlin, will follow in the footsteps of the PC chips and have HSA support, Gopalakrishnan said. He did not provide dates on when successors to the ARM-based Seattle chips would be released.
Up next: AMD’s ARM chips
Work is still underway among HSA Foundation members to bring compliance to ARM designs, which should pave the way for AMD to add HSA support, Gopalakrishnan said.
AMD is hedging its future bets on ARM servers, but most developers today write applications for x86 chips. AMD’s initial ARM chips will be targeted at servers processing web tasks like search or social networking requests, Gopalakrishnan said. It will also be used in content delivery systems, and also workloads based on Hadoop, a scalable computing environment that deals with large data sets.
The first ARM development kits are being bundled with Fedora Linux, Apache Web server, MySQL database, and also PHP, Java 7 and Java 8 programming languages. But Java 7 and 8 don’t have native support for parallel acceleration on CPUs and GPUs. Instead, extra layers of code are needed to bring parallel execution support to Java virtual machines (JVMs).
Parallel execution across CPUs and graphics processors is a key feature of HSA, and Oracle is working with AMD on an OpenJDK project called Project Sumatra to bring native parallel execution support to Java 9, which is due for release in 2015. Oracle is a member of the HSA Foundation, and the parallel execution feature on Java 9 will work across instruction set architectures.
AMD’s Seattle doesn’t have an integrated graphics core like the other chips from the company and thus no capacity to offload processing tasks, said Jim McGregor, principal analyst at Tirias Research.
But being HSA compliant has its benefits, as all the cores talk to each other and share memory. The HSA has a new uniform memory architecture called HUMA, which makes different memory types in a system accessible to all processors. As a result, developers have access to a larger shared memory pool for code execution.
“It would share cache between cores, and that would be a benefit,” McGregor said.
HSA also helps reduce memory transfers, which improves system performance while saving power. HSA tools also ease programming as coders don’t have to keep track of what’s in each memory pool.
HSA Foundation has set an ambitious goal on hardware and software tools, but it’s going to be a while until programmers and system makers can take full advantage of it, McGregor said.
“They’re still in the process of developing of the tools” and libraries, McGregor said.