Intel

AMD CEO Lisa Su teases Instinct MI200 MCM accelerator

AMD Instinct MI200 shown for the first time?

AMD is preparing for an announcement of its next-generation HPC products.

The company CEO and President Dr. Lisa Su is eager to showcase its new EPYC and Instinct accelerators at AMD Accelerated event on November 8th. AMD is expected to unveil its Zen3 based EPYC processors codenamed Milan-X as well as CDNA2 based Instinct MI200 accelerator series. Both products might have just been shown in a teaser image:

AMD Instinct MI200 teaser, Source: Lisa Su

The Instinct MI200 is a GPGPU designed to accelerate algorithms used in data centers. It will be the first Multi-Chip-Module GPU featuring two compute dies. The accelerator known as Instinct MI250X is to offer 110 Compute Units and up to 128 GB of HBM2e memory.

We know for the fact that this is a MI200 series GPGPU because it was already confirmed to use an OAM (OCP Accelerator Module) connector. The current Instinct series is only offered in the PCI-Express form factor. When it comes to EPYC CPU added to the teaser image this could be the Milan-X, the updated Zen3 based EPYC CPU with 3D V-Cache. The only thing that can be concluded from the picture is that it does not look like the Zen4 “Genoa” SP5 package and it has 8-channel memory support.

Looking forward to showcasing our upcoming @AMDServer #EPYC and @AMDInstinct accelerators. Join us for our Accelerated Data Center Premiere on November 8th at 11 a.m. ET. https://t.co/O2xZjXctfj pic.twitter.com/ezlWMNO9gK

— Lisa Su (@LisaSu) November 5, 2021

Both products would be offered together for the next-generation server environment utilizing a high-bandwidth interconnect system such as AMD Infinity Fabric.  The platform is set to compete with NVIDIA Hopper and Intel Ponte Vecchio/Sapphire Rapids servers in 2022.

AMD Accelerated Data Center Premier is now scheduled for November 8th, just a day before NVIDIA GTC 2021 Fall keynote by its CEO Jensen Huang.

RUMORED AMD Instinct Accelerators Specifications
VideoCardzAMD Radeon Instinct MI60AMD Instinct MI100AMD Instinct MI250AMD Instinct MI250XAMD Instinct MI300
Architecture7nm GCN5 (GFX906)7nm CDNA1 (GFX908)7nm CDNA2 (GFX90A)7nm CDNA2 (GFX90A)CDNA3 (?)
CPUZen4 (?)
GPUVega 20ArcturusAldebaran (MCM)Aldebaran (MCM)? (MCM)
Compute Tiles11224
Compute Units64 (64)120< 1101104x (?)
FP32 Cores (Full GPU)4096 (4096)7680 (8192) TBC TBC 4x (?)
GPU Clock Speed1800 MHz~1500 MHz TBC ~1700 MHz TBC
FP16 Compute29.5 TFLOPS185 TFLOPS TBC 383 TFLOPS TBC
FP32 Compute14.7 TFLOPS23.1 TFLOPS TBC 47.9 TFLOPS TBC
FP64 Compute7.4 TFLOPS11.5 TFLOPS TBC 47.9 TFLOPS TBC
VRAM32 GB HBM232 GB HBM2 TBC 128 GB HBM2E TBC
Memory Clock1000 MHz1200 MHz TBC TBC TBC
Memory Bus4096-bit4096-bit TBC TBC TBC
Memory Bandwidth1 TB/s1.23 TB/s TBC TBC TBC
Form FactorDual Slot, Full LengthDual Slot, Full Length TBC OAM TBC
CoolingPassive CoolingPassive Cooling TBC TBC TBC
TDP300W300W TBC 500W TBC