7nm GPU, AMD Announces First Sampling

AMD has just confirmed that the transition to 7 nm for its GPU solutions is proceeding without major problems. If everything goes as planned the launch could happen in the near future.

7nm GPU, AMD Announces First Sampling, 7nm GPU, AMD Announces First Sampling, Optocrypto

7nm GPU, AMD Announces First Sampling

The information was launched by AMD’s big boss, Lisa Su during a conference call on the occasion of the publication of the group’s results for the 1st quarter 2018.

We are learning that AMD is working well on a next-generation GPU. It will benefit from a 7 nm engraving. Its design and development are ensured by the Radeon Technologies Group. The file is even advanced since it is planned to send the first samples this year.

Lisa Su said:

“I am also pleased to announce that our next-generation Radeon Instinct 7nm product, optimized for automatic learning workloads, works in our laboratories. We are on track to provide samples to customers later this year.”

GPU 7 nm, VEGA 20 is on rails

These details confirm the various rumors around a 7nm Vega graphics processor. The engine will be optimized to meet specific needs such as AI and Machine Learning. There is no doubt here since Lisa Su talks about a solution for a Radeon Instinct product and not Radeon RX.

It has also been confirmed a partnership with TSMC for manufacturing in 7 nm and not Globalfoundries. Although no explanation came to explain this choice, it is probable that the control of 7 nm by TSMC weighed on the scale.

However, Lisa Su did not specify the exact name of this GPU. It’s hard not to compare your announcements to recent discoveries made in a Linux patch.

They made it possible to discover on one side a name, Vega 20 and on the other side of the ten new material functionalities associated with Vega.

Vega 20 will be an important evolution of Vega 10. The watchword seems to be “always more” with double precision computing power multiplied by a factor of 8, a double memory bus and the possibility of using 32 GB of HBM2.