Samsung, ARM Expand Collaboration in HPC at 7nm, 5nm

Samsung, ARM Expand Collaboration in HPC at 7nm, 5nm

ARM has been making a splash in the HPC world lately, thanks to new efforts from companies like Fujitsu and HPE. Now, Samsung has jumped on board with its own announcements about a long-term partnership to extend ARM manufacturing capabilities into 7nm and 5nm nodes. Samsung claims its 7nm LPP (Low Power Plus) and 5nm LPE (Low Power Early) will enable ARM’s Cortex-A76 to hit clock speeds in excess of 3GHz.

In the announcement, Samsung also gave an update on its own foundry nodes and rollout progress. The company will begin rolling out 7nm LPP in the back half of this year, but won’t actually start using EUV until the first half of 2019. That slip suggests that Samsung’s first 7nm designs are going to be built using multi-patterning, with significantly higher manufacturing costs. EUV’s introduction is expected to reduce costs (it won’t improve silicon performance directly), but the manufacturing problems and risks have delayed its introduction for well over a decade. Still, the industry is making progress, albeit in fits and starts. Samsung isn’t giving much detail on 5nm yet, beyond describing it as a further advance and refinement over the design improvements baked into 7nm LPP.

The point of these partnership announcements is for Samsung to tell potential customers that it’s prepared to work with ARM on building bog-standard ARM Cortex designs, albeit designs intended for specific markets. Samsung has been moving into HPC for a while now; British server manufacturer Kaleao has introduced a line of servers built around Samsung’s Exynos 7420, which consists of a quad-core Cortex-A57 CPU and a second cluster of Cortex-A53 cores at lower clock speeds. In this server configuration, the Cortex-A53 cores are repurposed for system and management functions, while the Cortex-A57 cores are reserved for compute. NextPlatform has more details on these systems and the servers.

Samsung, ARM Expand Collaboration in HPC at 7nm, 5nm

“Arm and Samsung Foundry have collaborated on a large number of chips using Artisan physical IP on Samsung Foundry process technologies,” said Kelvin Low, vice president of marketing, Physical Design Group at Arm. “Samsung Foundry’s 7LPP and 5LPE nodes are innovative process technologies which will meet our mutual customers’ needs to deliver the next generation of advanced system-on-chips (SoCs) from mobile to hyperscale datacenters.”

It may have taken several years longer than originally anticipated, but ARM is slowly positioning itself as a force to be reckoned with in both the server and HPC markets. Intel and AMD are both going to have to contend with that — Intel, of course, as the reigning champion of the server market with something like 99% market share prior to the launch of AMD’s Epyc, and AMD as it attempts to gain new market share and retake its position in the larger server market.

Continue reading

GlobalFoundries to Soon Pilot 7nm and EUV Production

GlobalFoundries is pushing ahead with a plan to pilot EUV and 7nm production. We recently got a chance to sit down with the company and talk about its plans for the future.

Troubling News: AMD Moves 7nm GPU Production Back to TSMC

AMD isn't building its 7nm Vega at GlobalFoundries — and that's cause for concern.

AMD Needs More 7nm Capacity Than GlobalFoundries Can Provide

AMD is sourcing 7nm from TSMC and GlobalFoundries because it expects to have more need for 7nm volume than GF can supply.

AMD Demos 7nm Vega for Machine Intelligence at Computex

AMD's new Vega 7nm machine intelligence solution is debuting ahead of schedule.