AMD unveiled its complete end-to-end built-in AI platform imaginative and prescient and launched its open, scalable rack-scale AI infrastructure constructed on {industry} requirements at its annual Advancing AI occasion.
The Santa Clara, California-based chip maker introduced its new AMD Intuition MI350 Collection accelerators, that are 4 instances sooner on AI compute and 35 instances sooner on inferencing than prior chips.
AMD and its companions showcased AMD Intuition-based merchandise and the continued development of the AMD ROCm ecosystem. It additionally confirmed its highly effective, new, open rack-scale designs and roadmap that carry management Rack Scale AI efficiency past 2027.
“We can now say we are at the inference inflection point, and it will be the driver,” stated Lisa Su, CEO of AMD, in a keynote on the Advancing AI occasion.
In closing, in a jab at Nvidia, she stated, “The future of AI will not be built by any one company or within a closed system. It will be shaped by open collaboration across the industry with everyone bringing their best ideas.”
Lisa Su, CEO of AMD, at Advancing AI.
AMD unveiled the Intuition MI350 Collection GPUs, setting a brand new benchmark for efficiency, effectivity and scalability in generative AI and high-performance computing. The MI350 Collection, consisting of each Intuition MI350X and MI355X GPUs and platforms, delivers a 4 instances generation-on-generation AI compute improve and a 35 instances generational leap in inferencing, paving the best way for transformative AI options throughout industries.
“We are tremendously excited about the work you are doing at AMD,” stated Sam Altman, CEO of Open AI, on stage with Lisa Su.
He stated he couldn’t imagine it when he heard in regards to the specs for MI350 from AMD, and he was grateful that AMD took his firm’s suggestions.
AMD stated its newest Intuition GPUs can beat Nvidia chips.
AMD demonstrated end-to-end, open-standards rack-scale AI infrastructure—already rolling out with AMD Intuition MI350 Collection accelerators, fifth Gen AMD Epyc processors and AMD Pensando Pollara community interface playing cards (NICs) in hyperscaler deployments similar to Oracle Cloud Infrastructure (OCI) and set for broad availability in 2H 2025. AMD additionally previewed its subsequent era AI rack known as Helios.
It is going to be constructed on the next-generation AMD Intuition MI400 Collection GPUs, the Zen 6-based AMD Epyc Venice CPUs and AMD Pensando Vulcano NICs.
“I think they are targeting a different type of customer than Nvidia,” stated Ben Bajarin, analyst at Artistic Methods, in a message to GamesBeat. “Specifically I think they see the neocloud opportunity and a whole host of tier two and tier three clouds and the on-premise enterprise deployments.”
Bajarin added, “We are bullish on the shift to full rack deployment systems and that is where Helios fits in which will align with Rubin timing. But as the market shifts to inference, which we are just at the start with, AMD is well positioned to compete to capture share. I also think, there are lots of customers out there who will value AMD’s TCO where right now Nvidia may be overkill for their workloads. So that is area to watch, which again gets back to who the right customer is for AMD and it might be a very different customer profile than the customer for Nvidia.”
The most recent model of the AMD open-source AI software program stack, ROCm 7, is engineered to satisfy the rising calls for of generative AI and high-performance computing workloads— whereas dramatically bettering developer expertise throughout the board. (Radeon Open Compute is an open-source software program platform that permits for GPU-accelerated computing on AMD GPUs, notably for high-performance computing and AI workloads). ROCm 7 options improved assist for industry-standard frameworks, expanded {hardware} compatibility, and new improvement instruments, drivers, APIs and libraries to speed up AI improvement and deployment.
In her keynote, Su stated, “Opennesss should be more than just a buzz word.”
The Intuition MI350 Collection exceeded AMD’s five-year aim to enhance the power effectivity of AI coaching and high-performance computing nodes by 30 instances, in the end delivering a 38 instances enchancment. AMD additionally unveiled a brand new 2030 aim to ship a 20 instances improve in rack-scale power effectivity from a 2024 base yr, enabling a typical AI mannequin that at this time requires greater than 275 racks to be skilled in fewer than one absolutely utilized rack by 2030, utilizing 95% much less electrical energy.
AMD additionally introduced the broad availability of the AMD Developer Cloud for the worldwide developer and open-source communities. Objective-built for speedy, high-performance AI improvement, customers could have entry to a totally managed cloud setting with the instruments and adaptability to get began with AI initiatives – and develop with out limits. With ROCm 7 and the AMD Developer Cloud, AMD is reducing obstacles and increasing entry to next-gen compute. Strategic collaborations with leaders like Hugging Face, OpenAI and Grok are proving the facility of co-developed, open options. The announcement acquired some cheers from people within the viewers, as the corporate stated it might give attendees developer credit.
Broad Companion Ecosystem Showcases AI Progress Powered by AMD
AMD’s ROCm 7
AMD prospects mentioned how they’re utilizing AMD AI options to coach at this time’s main AI fashions, energy inference at scale and speed up AI exploration and improvement.
Meta detailed the way it has leveraged a number of generations of AMD Intuition and Epyc options throughout its knowledge middle infrastructure, with Intuition MI300X broadly deployed for Llama 3 and Llama 4 inference. Meta continues to collaborate intently with AMD on AI roadmaps, together with plans to leverage MI350 and MI400 Collection GPUs and platforms.
Oracle Cloud Infrastructure is among the many first {industry} leaders to undertake the AMD open rack-scale AI infrastructure with AMD Intuition MI355X GPUs. OCI leverages AMD CPUs and GPUs to ship balanced, scalable efficiency for AI clusters, and introduced it’ll provide zettascale AI clusters accelerated by the most recent AMD Intuition processors with as much as 131,072 MI355X GPUs to allow prospects to construct, prepare, and inference AI at scale.
AMD says its Intuition GPUs are extra environment friendly than Nvidia’s.
Microsoft introduced Intuition MI300X is now powering each proprietary and open-source fashions in manufacturing on Azure.
HUMAIN mentioned its landmark settlement with AMD to construct open, scalable, resilient and cost-efficient AI infrastructure leveraging the complete spectrum of computing platforms solely AMD can present.Cohere shared that its high-performance, scalable Command fashions are deployed on Intuition MI300X, powering enterprise-grade LLM inference with excessive throughput, effectivity and knowledge privateness.
Within the keynote, Pink Hat described how its expanded collaboration with AMD permits production-ready AI environments, with AMD Intuition GPUs on Pink Hat OpenShift AI delivering highly effective, environment friendly AI processing throughout hybrid cloud environments.
“They can get the most out of the hardware they’re using,” stated the Pink Hat exec on stage.
Astera Labs highlighted how the open UALink ecosystem accelerates innovation and delivers higher worth to prospects and shared plans to supply a complete portfolio of UALink merchandise to assist next-generation AI infrastructure.Marvell joined AMD to share the UALink swap roadmap, the primary actually open interconnect, bringing the final word flexibility for AI infrastructure.
