AMD Launches the First 7nm GPUs, Radeon Instinct MI60 and MI50


AMD has launched the Radeon Instinct MI60 and MI5 at its Next Horizon Event held at San Francisco. The company claims the Radeon Instinct MI60 and MI50 to be the world’s first and fastest 7nm data center GPUs.  Both of these accelerators were designed for the most demanding workloads, such as deep learning, high-performance computing, cloud computing and a lot more.

The Radeon Instinct MI60 and MI50 GPUs are based on the architecture of AMD’s Vega 20 GPU and produced under TSMC’s 7nm FinFET manufacturing process. The MI60 and MI50 are also the first GPU to support the latest PCI-SIG’s latest PCIe 4.0 x16 interface, which can go up to 31.51GB/s of bandwidth that is about 2X faster than other x86 CPU-to-GPU interconnect technologies.

Radeon Instinct MI60

The MI60 is built with 64 supercharged compute units with 4,096 stream processors. It has a 1,800MHz of peak engine clock and 32GB of HBM2 ECC (error-correcting code) memory clocked at a speed of 1GHz on a 4,096-bit memory interface. It also has a TDP (thermal design power) rated at 300W and gets its power from a 6-pin and 8-pin PCIe power connectors.

Radeon Instinct MI50

The MI50 is not that far compared to the MI60 as it is equipped with 60 compute units and 3,840 stream processors. The MI510 has a clock speed of 1,747MHz with a 16GB of HBM2 ECC memory operating at a speed of 1GHz on the same 4096-bit memory bus as the MI60.  The MI50 also has a 300W TDP and gets its power from a combination of a 6-pin and 8-pin PCIe power connectors.

The two GPUs provide flexible mixed-precision FP16, FP32 and INT4/INT8 capabilities for a performance focused workloads, such as training complex neural networks or running an inference against the trained networks.

The Radeon Instinct MI60 and MI50 accelerators highlight two Infinity Fabric Links per GPU that can support up to 200 GB/s of peer-to-peer bandwidth, intended for communications to go up to 6X faster than PCIe Gen 3 interconnect speeds. Aside from a faster connection speed, this feature allows the connection of up to four GPUs in a hive ring configuration.

The MI60 is expected to begin shipping to data center customers by the end of 2018, while the MI50 is expected to ship on the market by the end of the first half of 2019. Stay tuned with TechnoCodex to be updated.

For the latest tech news and updates follow TechnoCodex on Facebook, Twitter, Google+. Also, if you like our efforts, consider sharing this story with your friends, this will encourage us to bring more exciting updates for you.

Leave a comment

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More