Aurora exascale supercomputer lead leaves Intel for Samsung
Years of extended delays and changes for late 2022-planned system
Intel is expected to provide next-gen CPUs and GPUs for what will become one of the world's fastest supercomputers later this year, but when the United States' long-delayed Aurora project is finally up and running, it will happen without a top Intel architect who was key to its delivery.
Robert Wisniewski, Intel's technical lead and principal investigator for Aurora, has left the semiconductor giant for a new HPC role at Samsung's research and development division, according to a Tuesday post he wrote on LinkedIn. While Samsung's research arm focuses on many elements important to HPC, the company is not known for a prominent role in supercomputing, although that could be set to change.
In response to questions from The Register, an Intel spokesperson said the chipmaker has "a new leadership team in place" and is "on track" to deliver the US Department of Energy's Aurora exascale supercomputer project at Argonne National Laboratory later this year.
Intel did not specify who has taken over responsibilities for Wisniewski, who has been driving many of the technical efforts around Aurora since 2019. Wisniewski was also an Intel fellow for software on the company's so-called SuperCompute team.
The Aurora supercomputer is strategically important to the US government, in large part because it will be one of the country's first exascale supercomputers. This means capable of delivering at least 1 exaflop, or 1 quintillion floating point operations per second, for critical research projects, ranging from discovering new cancer treatments to simulating nuclear fusion reactions.
This is happening while China has already beat the US to the punch with not one but two systems that reached a peak performance of 1.3 exaflops.
Aurora is also strategically important for Intel, which, as the project's prime contractor, will debut its next-generation Xeon Scalable CPU, code-named Sapphire Rapids, and Ponte Vecchio GPU accelerator in the supercomputer. Intel late last year said these components will allow Aurora to deliver a peak performance of more than 2 exaflops, which would allow it to leapfrog China's systems and rocket to the top of the list of the world's fastest supercomputers.
- Intel ships mystery quantum hardware to national lab
- US to rev fastest supercomputer with powerful test system
- America's Argonne lab buys AI super from HPE, AMD, Nvidia while waiting for Intel
- Look at me. Look at me. I'm the El Capitan now: Cray to build US govt's $600m cray-cray exascale nuke app super
While Intel is expected to install Aurora sometime later this year, it is only after facing delays. The chipmaker's systems partner Cray, which is now owned by Hewlett Packard Enterprise, originally announced in 2015 that it had landed a deal to build a slower version of Aurora with a stated 2018 completion timeline. But that timeline slipped to 2021 when Intel canned Knights Hill, its high-end Xeon Phi accelerator, and transitioned to using its new Xe GPU architecture.
Aurora was then pushed out to 2022 after Intel said in mid-2020 that manufacturing issues had delayed the rollout of its 7nm manufacturing process by 12 months. The company was planning to use its 7nm process for Ponte Vecchio, but the manufacturing woes prompted Intel to pivot and only use its 7nm node for certain parts of the GPU while relying on foundry rival TSMC for other aspects.
It remains to be seen if Intel will truly deliver Aurora this year. There is reason for hope, as the company has reported improvements in its 7nm process over the last several months, and on Monday, the company said it was ahead of schedule in the development of next-generation nodes.
As for Wisniewski's new Samsung job, he said he will create a new US-based systems lab for the Samsung Advanced Institute of Technology, also known as SAIT. His official title at Samsung is senior vice president, chief architect of HPC and head of the SAIT America systems architecture lab.
"We'll be focusing on overcoming the memory and communication walls for AI and HPC applications and looking at these challenges from a system perspective," he wrote, adding that his team is hiring.
"We're grateful for all the contributions that Robert made during his time at Intel as well as his work on Aurora," the spokesperson said. An Argonne representative deferred to Intel on questions about staffing. ®