AMD has confirmed that a next-gen CDNA 4 multi-chip, multi-IP Instinct accelerator is currently in development and scheduled for release by 2023, known as the Instinct MI300 GPU. Technically speaking, this is indeed an APU that will combine the next generation of CDNA 3 cores with the next generation Zen 4 CPU cores.
That’s right – this chip combines CPU and GPU cores in a single package for data centers and AI, and the predicted performance boost is reportedly monstrous.
The Instinct MI300 accelerator features a unified memory APU architecture and new math formats to deliver a 5x performance per watt improvement over the CDNA 2, as well as a projected 8x improvement in AI training performance over its spiritual predecessor , the MI250X.
Instinct will make this possible using a 5nm process node and will feature the 4th generation Infinity architecture. In essence, we should see performance improvements as the unified architecture eliminates redundant memory copies and eliminates the need for a second pool of DRAM memory attached to the CPU.
Of course, these are AMD’s internal stats, so take performance predictions with a pinch of salt until real-world performance can be documented. An official release date has not been provided, although AMD confirmed during Financial Analyst Day 2022 that it plans to release the Instinct MI300 in 2023.
David Wang announced Instinct-class chips for the AI and Data Center segment along with the Compute GPU roadmap, which also includes AMD RDNA 4 Gaming GPUs.
“This is where we think we can win in terms of differentiation,” said Lisa Su, CEO of AMD, in her keynote speech at the event. “It’s about computing technology leadership. It’s about expanding datacenter leadership. It’s about expanding our AI footprint. It’s expanding our software capability. And then it’s really pulling together a broader effort.” of customized solutions, because we think this is an area of growth going forward.”
Analysis: This will not be an easy win for AMD
This is far from AMD’s first rodeo with supercomputing, having released its first data center GPU (the GCN-based Vega II) in late 2018, followed by its first CDNA-based Radeon Instinct MI100 GPU in 2020.
Nvidia also unveiled its GPU roadmap for the year in March with the announcement of its Hopper GPU architecture and has long dominated supercomputer accelerator sales, reporting a 90% market share in 2021.
AMD’s latest foray into the data center market won’t be an easy win, especially with Intel’s Falcon Shores XPU set to add more competition in 2024, although that does mean AMD, Nvidia and Intel will have hybrid CPU-GPU chips entering the mainstream. central market data in the coming years.
Still, Team Red is more confident than ever, so we’ll have to watch that space and see if Nvidia can keep its supercomputing crown.