AMD and its primary fab partner GlobalFoundries have signed an updated five-year wafer supply agreement that will extend through the end of 2020. The restructuring simultaneously deepens the commitment between the partners and gives AMD limited freedom to see other foundries. In exchange, GlobalFoundries will get some additional compensation.

Per the terms of the agreement, which pertains to AMD’s microprocessor, graphics processor, and semi-custom products, AMD will make $25 million cash installments to GlobalFoundries over the next four quarters, for a total cash transfer of $100 million. Beginning in 2017, AMD will be required to make quarterly payments to GlobalFoundries based on the volume of certain wafers it is obtaining from another foundry.

The agreement also stipulates annual wafer purchase targets for the five-year period, sets fixed wafer prices for 2016, and provides a framework for yearly wafer pricing. If annual targets are not met, a penalty will be imposed based on the difference between actual wafer purchases and the target for that year.

In further consideration, AMD will issue a warrant to GlobalFoundries parent company Mubadala Development Company granting them rights to purchase 75 million shares of AMD common stock at a purchase price of $5.98 per share so long as ownership is kept under 20 percent. The warrant is valued at $235 million.

This is the sixth time that the partners are updating the wafer supply agreement that was established when AMD spun out its fab business in 2009. The original contract runs until 2024.

AMD said it expects to purchase approximately $650 million in wafers during 2016. $495 million of that falls under the sixth amendment and $155 million falls under the fifth amendment.

No 10nm for GlobalFoundries

AMD’s recently-released Polaris GPUs and upcoming “Zen”-based processors are being made at the GlobalFoundries fab 8 plant in Malta, NY. Like IBM Power9 and Intel Kaby Lake, Zen uses 14nm process technology. Kaby Lake, released this week, is Intel’s third generation of processors to use the 14nm process, while AMD has just begun shipping 14nm production silicon (Polaris GPUs).

Interestingly, we learned this week that GlobalFoundries is skipping the 10nm process. A statement from AMD said that it will collaborate with GlobalFoundries on the 7nm technology node, building on the success of the 14nm node. It’s possible that AMD will use another fab’s services for a 10nm process node, but going from 14nm to 7nm would put the chipmaker in position to leapfrog Intel.

Intel’s 10nm die shrink is expected for the second half of 2017 with the debut of Cannonlake (formerly Skymont). This is the same time frame that IBM is releasing its 14nm next-gen Power9 CPU. AMD plans to launch its first 14nm CPU, the Zen-based “Summit Ridge,” in early 2017.

Zen Expectations

AMD showed off some of its new Zen-based silicon in San Francisco last month and talked up the new high-performance cores. The next-gen microarchitecture design, led by renowned chip architect Jim Keller, is said to enable a 40 percent improvement in instructions per clock over Zen’s predecessor, Excavator.

At the event, AMD put an engineering release of a desktop-class Zen CPU to the test against Intel’s top-end desktop processor the “Broadwell-E” Core-i7. “An 8-core, 16-thread “Summit Ridge” desktop processor (featuring AMD’s “Zen” core) outperform[ed] a similarly configured 8-core, 16-thread Intel “Broadwell-E” processor when running the multi-threaded Blender rendering software with both CPUs set to the same clock speed,” said AMD in a statement.

Of particular relevance to HPC- and datacenter-watchers, AMD also debuted its 32 core (64-thread) “Naples” server CPU based on the Zen microarchitecture. The dual-socket server was running Windows Server OS, but will presumably support Linux when it hits the market (expected in Q2 2017).