Currently AMD has two GPU architectures on the market, while RDNA 2 is intended for gaming and is really a GPU in the strictest sense of the word. CDNA 2, like its predecessor, does not serve to render graphics and it is not only due to the fact that it lacks a video output, but because all the specialized graphics units have been eliminated.
CDNA architectures are a computationally optimized version of the old GCN architecture with enhancements such as support for 64-bit or double-precision floating-point calculations, typical of HPC GPUs, and the addition of Matrix Core Units, the equivalent of the NVIDIA Tensor Cores that we assume will be ported to RDNA in a future iteration.
AMD’s first MCM GPUs are now shipping
The source is AMD itself, which in the presentation of results for the second quarter of this year and through its set of slides has confirmed that the AMD Instinct MI200 are already being distributed and therefore they are already fully completed.
The particularity of these «graphics» cards? We have known for a long time that for the high-performance computing market there are plans by Intel, with its Xe-HP and Xe-HPC and NVIDIA, with its Hopper architecture, to launch a chiplet-based HPC GPU, but AMD has been ahead of them both with the launch of their Instinct MI200.
In any case we have to clarify one thing, HPC graphics cards do not suffer the same problem as graphics cards where it is very difficult to make several GPUs render the same scene. So we are not facing an MCM GPU by AMD that is based on the same technology that we will see in RDNA 3. Although like the rest of contemporary HPC GPUs it is based on the use of HBM2E memory and therefore is a 2.5DIC configuration with memory chips and stacks on a common interposer.
With chiplets yes, but different from the future RDNA 3
We don’t really know if this AMD Instinct 200 makes use of the silicon bridge of a certain AMD patent or if it is not the classic interposer. Although everything indicates that both GPUs are intercommunicated through a xGMI interface, a faster version of the Infinity Fabric and which is the AMD equivalent of NVIDIA’s NVLink. Each of the two 128 CU GPUs that make up the MCM have 4 HBM2E memory stacks of 16 GB each, making a configuration of 128 GB memory, the highest of a GPU of its kind.
What’s more, everything indicates that AMD could be making use of the CoWoS-S packaging third generation, which was dated on the TSMC roadmap for this year with 8 HBM batteries, which matches the AMD Instinct MI200 that AMD just launched, although you will not be able to buy it, as it is only for certain
What are the specifications of the AMD Instinct MI 200?
The AMD Instinct MI 200 is a Dual HPC GPU, which is composed of two symmetric 128 Compute Unit GPUs each with CDNA 2 architecture. Where each of the Compute Units is composed of 64 ALU and has a tensor unit for matrix operations. A type of key unit for algorithms based on convolutional neural networks. It is therefore a GPU with which AMD intends to compete against the NVIDIA Tesla based on its A100 architecture.
The fact that it is a dual GPU places it above in raw power compared to AMD’s solution and anticipates the proposals of Intel and NVIDIA with MCM GPUs for HPC, which are still undated at this time, while NVIDIA Hopper isn’t expected until at least next year. Will AMD be able to take market share from NVIDIA in HPC GPUs like it has done to Intel in server CPUs? Only time will tell, but they have a tough nut to crack at NVIDIA.