Today, at their New Horizon event in San Francisco, AMD has officially announced their first GPUs based on the 7nm node; the Radeon Instinct MI50 and MI60. Both the GPUs are a big leap ahead from the current technology, however, they are intended for data center use. According to AMD, the Radeon Instinct GPUs will offer unparalleled performance in things like HPC, deep-learning, weather research, etc.
Combining world-class performance and a flexible architecture with a robust software platform and the industry’s leading-edge ROCm open software ecosystem, the new AMD Radeon Instinct™ accelerators provide the critical components needed to solve the most difficult cloud computing challenges today and into the future.
– David Wang, Senior VP of Engineering at AMD Radeon.
Both the Radeon Instinct MI50 and MI60 are optimized for data centers, with high floating point performance and advanced compute units. Aside from the hardware prowess of these GPUs, AMD also emphasized a lot on the software side of things. With a new version of the ROCm open software platform on the way, things are looking great for these GPUs at launch.
Now, onto the sweet stuff. Both GPUs boast some serious specifications.
|AMD Radeon Instinct MI50||AMD Radeon Instinct MI60|
|Memory Interface: HBM2||Memory Interface: HBM2|
|Memory Bandwidth: 1TB/s||Memory Bandwidth: 1TB/s|
|Memory Size: 16GB HBM2||Memory Size: 32GB HBM2|
|ECC Memory: Yes||ECC Memory: Yes|
|FP64 Performance: 6.7 TFLOPS||FP64 Performance: 7.4 TFLOPS|
|PCI Interface: PCIe 4.0||PCI Interface: PCIe 4.0|
|AMD Infinity Fabric™ Link Support: Yes||AMD Infinity Fabric™ Link Support: Yes|
|Process Node: 7nm||Process Node: 7nm|
AMD also introduced the Infinity Fabric Link technology with the Radeon Instinct GPUs. In multi-GPU configurations, the Instinct GPUs will be up to 6 times faster than standard PCIe 3.0 peer-to-peer transfers. AMD also talked about the security features of the Instinct GPUs. Both the MI50 and the MI60 support AMD’s MxGPU technology, which is a hardware-based virtualization technology that makes hacking very difficult at the hardware level. This is huge in cloud-computing environments.
In the end, AMD talked about the availability of these upcoming GPUs. While they didn’t discuss the pricing, they did announce that both of the Radeon Instinct GPUs will be available and shipped to data centers by the end of 2018. The new version of the AMD ROCm software will also go live as the Instinct GPUs ship out, by the end of 2018.
It is great to see AMD starting to make an appearance back in the GPU market, after their success in the CPU market. It is certainly impressive that AMD is squeezing this much performance out of the current Vega architecture. The potential in these GPUs is great. With the upcoming Navi architecture and the implementation of HBM3 or GDDR6 memory along with 7nm process node & PCIe 4.0, we expect performance numbers to go through the roof.
This is huge for data centers that demand the highest level of computational power. With both Nvidia and AMD making leaps in GPU technology, complex problems like weather predictions, disease prevention, genome mapping, computational biology, etc., can be solved much more quickly and effectively. Also, AMD’s use of Error Correcting (ECC) memory in these GPUs is a big thing for a lot of users.