AMD Advancing AI event: 6 key details you need to know...
At AMD's Advancing AI event in San Jose, California, United States, the brand unveiled a range of hardware as well as software-centric announcements, which at large serves as the vision for its open AI ecosystem. Here are the key announcements:
1. AMD Instinct MI350 series GPUs announced:
AMD announced the Instinct MI350X and Instinct MI355X GPUs and platforms, which the company claims allow for a four times generation-on-generation AI compute boost for better AI solutions across various industries.
2. End-to-end open standards rack-scale infrastructure:
At the keynote, AMD also showed its open standards rack-scale AI infrastructure, which is already rolling out with AMD Instinct MI350 series accelerators, 5th Gen AMD EPYC processors, and AMD Pensando Pollara NICs in hyperscaler deployments, including Oracle's Cloud Infrastructure. The company says the broad availability for the same is going to be the second half of 2025.
3. AMD unveils its next-generation AI rack called Helios:
AMD stated that this is going to be built on the next-gen AMD Instinct MI400 series GPUs. When you compare this to the former generation, it can allow for 10 times more performance when running inference on mixture-of-experts models.
4. AMD announced the broad availability of its Developer Cloud:
The company says this is purpose-built for fast, high-performance AI development and allows for developers to have access to a fully managed cloud environment with tools, allowing for fast development. The company says the combination of ROCm 7 and the AMD Developer Cloud enables the company to expand access to next-generation compute. It is already collaborating with AI leaders like Hugging Face, OpenAI, and Grok.
5. AMD's latest version of its AI software stack, ROCm 7:
The company says that this is to serve the growing demands of generative AI and the demand for more compute. The company says that this is going to enhance the experience for developers and features improved support for industry-standard frameworks and more.
6. AMD also revealed its partner ecosystem:
Seven out of ten of the largest AI model builders, including the likes of Meta, OpenAI, Microsoft, and xAI, are now on board with AMD for training their AI models, the company said. For instance, it was detailed how the Instinct MI300X is used for Llama 3 and for Llama 4 inference. Sam Altman, OpenAI CEO, also discussed hardware, software, and more. Other companies like Microsoft and Red Hat also joined them.
