ARM Ltd has agreed to be bought by Japanese technology company SoftBank for $32B according to both companies. ARM-based chips are already dominant players in the mobile computing market and recently efforts to push ARM processors into servers, including HPC, have gained momentum. For example, Japan’s next flagship supercomputer, the post K Computer, will be ARM-based and built by Fujitsu.
Many details are still to come. ARM CEO Simon Segars has said the deal is good for shareholders (roughly 43 percent premium over recent share price) and that SoftBank will be able to invest more heavily in the ARM technology roadmap and global expansion of the ecosystem. One observer wondered if currency and stock fluctuations associated with Brexit hasted the deal along. Market watcher IDC offered a quick assessment.
“Softbank’s stated rationale for buying ARM Limited is to exploit ARM’s strong position in the mobile technology market and get a head start in the nascent IoT market. But ARM recently signaled its intent to push harder into the HPC space, and it’s clear that HPC will play a pivotal role in the IoT. So, whether Softbank realizes it or not, the company is going to be an HPC vendor. It’s too soon to tell whether this acquisition will have any effect on Japan’s post-K computer, which will be ARM based,” said Steve Conway, IDC’s research vice president, HPC/HPDA.
The ARM v8 64-bit processor has many fans but so far few big sellers in the server space. Availability of advanced ARM processors is growing steadily as Cavium (ThunderX) and Applied Micro (X-Gene) ramp up production. HPE, of course, has an ARM cartridge for its Moonshot line of datacenter servers. Penguin Computing also has a higher-end ARM-based offering. Still, some have questioned ARM’s ability to fend off Intel in the server space.
Earlier this spring, ARM’s Darren Cepulis told HPCwire, “Fundamentally the system vendors are already confident enough that they have development platforms and things and they are in the middle of that investment and proof of concept pilots and things going through deployment processes with potential customers to evaluate suitability and benefits of the technology, tens of those in the works right now.”
The ARM core has often been criticized as weak but that will change, according to one ARM chipmaker which told HPCwire, “Today the ARM core isn’t equivalent to an Intel core in terms of spec and rate. It will take time. Today the gap is there. No way someone is going to say oh my ARM core’s spec and rate are better than Broadwell or Haswell or Phi. But two years from now when we have the same conversation, things will be different.”
Filippo Mantovani, a researcher at the Barcelona Supercomputer Center and leader in the EU Mont-Blanc project exploring ARM for use in supercomputing, said “Strictly related to HPC, I think it is still too early to guess the consequences. Most of the companies developing ARM based HPC chips are just using ARM ISA, designing their own cores. In this sense, I do not expect a big impact in HPC even if Softbank decides to steer drastically in IoT direction… A longer term effect could be generated if the “new ARM” will decide to resize the effort in developing HPC software ecosystem: a smaller effort could damage the establishment of ARM as an alternative in HPC, a bigger effort would be of course a great benefit.”
Montavani’s group stood up the pilot Mont-Blanc system last June (see HPCwire article,
Filippo Mantovani on What’s Next for Mont-Blanc and ARM). Bull Atos is the contractor for the larger Mont-Blanc machine.
HPCwire will update this article as more becomes clear.