AMD Instinct MI200 to launch by the end of this year
After confirming that its CDNA2 GPU is now shipping to first customers the manufacturer has now announced through its corporate presentation that those processors will launch by the end of this year.
AMD CDNA2 is the successor to CDNA, the first AMD architecture developed specifically for datacenters. A GPU based on this architecture known as Aldebaran, has recently been confirmed by AMD to be shipping to first customers. This means that the launch of this GPU for a global market is imminent. As expected, this has been now also confirmed by AMD’s frequently updated corporate presentations for investors.
Despite the update, the presentation slide still does not mention the exact fabrication node used for this CDNA2 architecture. What is described as ‘Advanced Node’ simply means that the GPU chiplets and I/O might be made using different nodes.
AMD Aldebaran is to be AMD’s first MCM (multi-chip-module) GPU which is now set to debut before AMD’s own gaming architecture featuring MCM GPUs known as RDNA3. The GPU is now rumored to feature two chiplets each with 128 Compute Units. Assuming that each Unit is to offer 128 Stream Processors, the full package would feature up to 16384 cores.
Naturally, the Instinct MI200 will not have all cores enabled to preserve good yields and undisturbed supply chain to customers. Alongside a massive improvement in core count in comparison to first-generation CDNA GPU, Arcturus, the Aldebaran is to be offered with up to 128GB of HBM2e memory. This rumored configuration of Aldebran GPU has been conveniently put in a block diagram by Twitter user Locuza.
AMD Aldebaran GPU Block Diagram, Source: @Locuza_
AMD Instinct MI200 is set to compete with Intel Xe-HP(C) and NVIDIA Hopper accelerators, which are now expected to launch in the next three quarters. Each of those accelerators is to use an MCM package design, signifying a very important transition in the GPU industry.
| AMD Instinct Accelerators | |||
|---|---|---|---|
| Accelerator Name | AMD Radeon Instinct MI60 | AMD Instinct MI100 | AMD Instinct MI200 |
| Architecture | 7nm GCN5 | 7nm CDNA1 (GFX908) | CDNA2 (GFX90A) |
| GPU | Vega 20 | Arcturus | Aldebaran (MCM) |
| GPU Cores | 4096 | 7680 | less than 16384 (?) |
| GPU Clock Speed | 1800 MHz | ~1500 MHz | TBC |
| FP16 Compute | 29.5 TFLOPs | 185 TFLOPs | TBC |
| FP32 Compute | 14.7 TFLOPs | 23.1 TFLOPs | TBC |
| FP64 Compute | 7.4 TFLOPs | 11.5 TFLOPs | TBC |
| VRAM | 32 GB HBM2 | 32 GB HBM2 | 128 GB HBM2E |
| Memory Clock | 1000 MHz | 1200 MHz | TBC |
| Memory Bus | 4096-bit bus | 4096-bit bus | TBC |
| Memory Bandwidth | 1 TB/s | 1.23 TB/s | TBC |
| Form Factor | Dual Slot, Full Length | Dual Slot, Full Length | OAM |
| Cooling | Passive Cooling | Passive Cooling | TBC |
| TDP | 300W | 300W | TBC |
Source: AMD Corporate Presentation (PDF)
