AMD unveils Instinct MI450X GPU and publishes MI455X inference results
AMD introduces the Instinct MI450X and publishes MI455X inference results, continuing to compete in data centre AI acceleration.
AMD has introduced the Instinct MI450X GPU and published inference results for the MI455X. Both products sit in the AI accelerator line for data centres, where demand for compute for model inference is rising fast.
Publishing inference results for the MI455X shows the focus is on the model operation stage, not only the training stage. Inference is the cost that runs across the whole lifecycle of an AI product, so efficiency at this step directly affects the operating cost of the organisation deploying it.
With the MI450X, the new product introduction continues the upgrade cycle of the Instinct line. AMD announcing a new product and the results of another product in parallel shows the company presenting capability step by step, tied to different chip generations within the same line.
The data centre AI accelerator market has long been concentrated among a few suppliers, so the arrival of one more hardware option matters to organisations that want to avoid dependence on a single platform. Even so, a switch still depends on the software ecosystem and on compatibility with the models in use.
Source: amd.com
