AMD only just unveiled its upcoming Instinct MI300 line of GPU accelerators, packing the CDNA 3 GPU and Zen 4 CPU architectures with a whopping 192GB of super-fast HBM3 memory. The AMD Instinct MI300X accelerator was designed from the ground up for large language models (LLMs) that require huge amounts of memory capacity and memory bandwidth.

The new AMD Instinct MI300 accelerators use the company's impressive and innovative die-stacking technology, with Zen 4 CPU cores mixed with CDNA 3 GPU cores all in a single package. AMD has the Instinct MI300A which is made for HPC and AI workloads, with its 128GB of on-board HBM3 memory used well here. The unified memory architecture is a huge win for AMD, which looks to continue it into the next-gen Instinct MI400 series of GPU accelerators.

Now we're hearing about the next-gen, high-end, data-center GPU accelerator -- the Instinct MI400 series -- which should be based on the next-gen CDNA 4 GPU architecture. We should expect to see Zen 5 CPU cores mixed with CDNA 3 GPU cores, for a monster new accelerator. Not only that, but we should expect AMD to use its new XSwitch interconnect technology, which will fight off its competitor with NVIDIA and its NVLink technology.

AMD CEO Dr. Lisa Su teased: "When you look across those workloads and the investments that we’re making, not just today, but going forward with our next generation MI400 series and so on and so forth, we definitely believe that we have a very competitive and capable hardware roadmap. I think the discussion about AMD, frankly, has always been about the software roadmap, and we do see a bit of a change here on the software side".

AMD's new Instinct MI400 GPU accelerator won't go inside of your new Gaming PC, but for HPC and AI systems... oh man, AMD will actually have a high-end GPU competitor... unlike the Radeon RX 8000 series GPUs that we're hearing won't come in high-end gaming GPU form.

Source