Powered By Zen 4 CPU Cores & CDNA 3 GPU Cores For Blistering Fast HPC Performance

AMD also seems to be working on its first-generation Exascale APU product, the Instinct MI300, powered by Zen 4 CPU & CDNA 3 GPU cores. The details of this HPC chip have also been leaked over in the latest video by AdoredTV.

AMD Instinct MI300 To Be Red Team’s First Exascale APU Product With Zen 4 CPU, CDNA 3 GPU Cores & HBM3 Memory

The first references to AMD’s Exascale APU date all the way back to 2013 with more bits and pieces being revealed in the coming year. Back in 2015, the company revealed its plan to offer the EHP, an Exascale Heteorgenous Processor, based on the then-upcoming Zen x86 cores and Greenland GPU with HBM2 memory on a 2.5D interposer. The original plans were eventually scrapped & AMD went on to release its EPYC and Instinct lineup in their own CPU and GPU server segments. Now, AMD is bringing EHP or Exascale APUs back in the form of the next-gen Instinct MI300.

AMD EPYC Server Roadmap Leaks Out: EPYC Genoa-X With Zen 4 & 3D-V-Cache in 1H 2023, Genoa, Bergamo & Turin With SP6 Support

AMD’s First Exascale APU, the Instinct MI300, has been leaked and detailed. (Image Credits: AdoredTV)

Once again, the AMD Exascale APU will form a harmony between the company’s CPU & GPU IPs, combining the latest Zen 4 CPU cores with the latest CDNA 3 GPU cores. This is said to be the first generation Exascale & Instinct APU. In the slide posted by AdoredTV, it is mentioned that the APU will be taped out by the end of this month which means we can see a potential launch in 2023, the same time the company is expected to unveil its CDNA 3 GPU architecture for the HPC segments.

The first silicone is expected to be in AMD’s labs by Q3 2022. The platform itself is regarded as MDC which could mean Multi-Die Chip. In a previous report, it was stated that the APU will carry a new ‘Exascale APU mode’ and feature support on the SH5 socket which is likely coming in the BGA form factor.

Besides the CPU and GPU IPs, another key driver behind the Instinct MI300 APU would be its HBM3 memory support. Though we still aren’t sure of the exact number of dies featured on the EHP APU, Moore’s Law is Dead has previously revealed die configurations with 2, 4, and 8 HBM3 dies. The die shot is shown in the slide in the latest leak and also shows at least 6 dies which should be a brand new configuration. It is possible that there are several Instinct MI300 configurations being worked upon with some featuring just the CDNA 3 GPU dies and APU designs with both Zen 4 & CDNA3 IPs.

So it looks like we are definitely going to see the Exascale APUs in action after almost a decade of wait. The Instinct MI300 is definitely aiming to revolutionize the HPC space with insane amounts of performance never seen before and with core and packaging technologies that are going to be a revolution for the tech industry. The chip will be competing against NVIDIA’s Grace+Hopper Super Chip and Intel’s Ponte Vecchio HPC accelerators next year.

AMD Radeon Instinct Accelerators 2020

Accelerator Name AMD Instinct MI300 AMD Instinct MI250X AMD Instinct MI250 AMD Instinct MI210 AMD Instinct MI100 AMD Radeon Instinct MI60 AMD Radeon Instinct MI50 AMD Radeon Instinct MI25 AMD Radeon Instinct MI8 AMD Radeon Instinct MI6
CPU Architecture Zen 4 (Exascale APU) N/A N/A N/A N/A N/A N/A N/A N/A N/A
GPU Architecture TBA (CDNA 3) Aldebaran (CDNA 2) Aldebaran (CDNA 2) Aldebaran (CDNA 2) Arcturus (CDNA 1) Vega 20 Vega 20 Vega 10 Fiji XT Polaris 10
GPU Process Node 5nm+6nm 6nm 6nm 6nm 7nm FinFET 7nm FinFET 7nm FinFET 14nm FinFET 28nm 14nm FinFET
GPU Chiplets 4 (MCM / 3D Stacked)
1 (Pere Die)
2 (MCM)
1 (Pere Die)
2 (MCM)
1 (Pere Die)
2 (MCM)
1 (Pere Die)
1 (Monolithic) 1 (Monolithic) 1 (Monolithic) 1 (Monolithic) 1 (Monolithic) 1 (Monolithic)
GPU Cores 28,160? 14,080 13,312 6656 7680 4096 3840 4096 4096 2304
GPU Clock Speed TBA 1700MHz 1700MHz 1700MHz 1500MHz 1800MHz 1725MHz 1500MHz 1000MHz 1237MHz
FP16 Compute TBA 383 TOPs 362 181 TOPs 185 TFLOPs 29.5 TFLOPs 26.5 TFLOPs 24.6 TFLOPs 8.2 TFLOPs 5.7 TFLOPs
FP32 Compute TBA 95.7 TFLOPs 90.5 TFLOPs 45.3 TFLOPs 23.1 TFLOPs 14.7 TFLOPs 13.3 TFLOPs 12.3 TFLOPs 8.2 TFLOPs 5.7 TFLOPs
FP64 Compute TBA 47.9 TFLOPs 45.3 TFLOPs 22.6 TFLOPs 11.5 TFLOPs 7.4 TFLOPs 6.6 TFLOPs 768 GFLOPs 512 GFLOPs 384 GFLOPs
VRAM 192 GB HBM3? 128 GB HBM2e 128 GB HBM2e 64 GB HBM2e 32 GB HBM2 32 GB HBM2 16 GB HBM2 16 GB HBM2 4 GB HBM1 16 GB GDDR5
Memory Clock TBA 3.2 Gbps 3.2 Gbps 3.2 Gbps 1200MHz 1000MHz 1000MHz 945MHz 500MHz 1750MHz
Memory Bus 8192-bit 8192-bit 8192-bit 4096-bit 4096-bit bus 4096-bit bus 4096-bit bus 2048-bit bus 4096-bit bus 256-bit bus
Memory Bandwidth TBA 3.2 TB/s 3.2 TB/s 1.6 TB/s 1.23 TB/s 1 TB/s 1 TB/s 484 GB/s 512 GB/s 224 GB/s
Form Factor OAM OAM OAM Dual Slot Card Dual Slot, Full Length Dual Slot, Full Length Dual Slot, Full Length Dual Slot, Full Length Dual Slot, Half Length Single Slot, Full Length
Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling Passive Cooling
TDP ~600W 560W 500W 300W 300W 300W 300W 300W 175W 150W

Leave a Reply

%d bloggers like this: