While Intel has delayed its CPU-GPU superchip, AMD takes the opportunity to indicate that its copy is on the right track and is essential in the current context of technology development.

AMD says it will officially launch its Instinct MI300 in the second half of this year. Wide general availability of the accelerator is planned for 2024. AMD’s CTO Mark Papermaster pointing this out at the Morgan Stanley Technology, Media and Telecom conference is a subtle jab at Intel.

After all, the MI300 will be a superchip, combining 24 Epyc Genoa CPU cores with a CDA-3 GPU and HBM3 memory. Intel also wanted to release such a chip under its GPU offering, codenamed the Rialto Bridge, but that lineup has since been scrapped. The Falcon Shores successor has been pushed back to 2025 and it also appears that a CPU and GPU combo will not be part of the first wave of Falcon Shores hardware. AMD is about to take the lead here in a drastic way.

Indispensable for the future

That matters, according to Papermaster, who points out that chips like the CPU-GPU hybrid MI300 are essential in the current context. Eventually, Moore’s law is no longer very strong, which means the cost of more transistor density goes up. Part of the solution is to build more efficient custom chips. For AI and HPC workloads, a hybrid SoC with a combination of CPU, GPU, and memory is a great solution.

AMD is not alone in this belief: Nvidia calls its version the Grace Superchip and combines ARM-based cores with RAM and GPU computing power. Intel knows that the future of computing power is headed in this direction and has wanted to fulfill this in its roadmaps, but now seems unable to do so. Soon there will only be one superchip left in x86 land: AMD’s.