Beyond silicon: AMD evolves AI processor performance, makes play for investment trillions

Long after the chip wars in the 1990s which established the necessity for a close combination of silicon with software, AMD is stepping up its attention on programming as among the keys to future performance. While AMD plays on the pinnacle of hardware development, it is similarly at the sharp end of software, too, as the company's VP of AI Software Anush Elangovan made clear to TechDay.

"The transformation of AMD reflects the reality that AI performance and adoption are no longer determined by silicon alone," Elangovan sets out. "Hardware leadership remains our foundation, but software defines how quickly that performance translates into value."

Steve Ballmer memorably stomped around declaring his love for 'developers, developers, developers' (and the man had a point), and today Elangovan echoes this essential sentiment. "The key driver has been developer accessibility, making compute power usable and optimisable without friction."

Asked how AMD achieves this goal, he says AMD's ROCm open software platform brings together CPUs (central processing unit), GPUs (Graphical PU), NPUs (Neural PU), and adaptive computing solutions into one software environment. "ROCm has matured into the core of AI strategy at AMD, enabling developers to build and deploy models without being tied to a proprietary ecosystem," he explains.

The shift, Elangovan continues, reflects a fundamental elevation of AMD's position in AI. "We now compete not only on hardware capability but on software transparency, interoperability, and scalability. ROCm software's integration with major frameworks like PyTorch, TensorFlow, JAX, vLLM and SGLang, and its day-zero support readiness for Llama, DeepSeek, and OpenAI GPT-OSS 20B and 120B, means AMD hardware can be adopted quickly across the AI lifecycle – from research and prototyping to production training and inference."

This is a step up for a company best known as 'the challenger brand that succeeded' in the previously single-vendor dominated desktop, server and GPU markets, because it puts AMD front and centre in the AI solutions market, alongside that other behemoth: Nvidia and its CUDA ecosystem. "We've evolved from delivering high-performance components to offering end-to-end AI solutions that give developers choice and enterprises the confidence to build sustainable AI infrastructure on open standards," Elangovan confirms.

Again, it's all about empowering developers, specifically those working with AI and high performance computing workloads, 'from local prototyping to large, scale deployment', according to Elangovan. Similarly to Nvidia's CUDA, "ROCm goes from silicon to solution', while of course prominently featuring the hardware by now closely associated with these workloads; AMD is of course known for its Instinct and Radeon product lines.

"ROCm software [also] provides continuity between development and production environments. Its integration into major Linux distros and Windows drivers simplifies setup, often to a single command, while native support for popular AI frameworks offers familiar developer tools with minimal friction," Elangovan adds.

If it sounds a bit like the famously scrappy (and famously successfully so, too) AMD has bigger things in mind, he all but confirms that this is the direction the chipmaker is taking. "The goal of AMD ROCm software is not only to match the competition's capabilities but to surpass them by leveraging the strength of the open community," Elangovan asserts.

"Developers worldwide are eager to collaborate, contribute, and accelerate the progress made in ROCm software. ROCm software's open-source nature and flexibility give it unique advantages that closed ecosystems cannot match. With the combined force of the engineering teams at AMD and the global developer community, ROCm software is evolving rapidly and delivering great performance, rich tooling, and broad functionality that will continue to expand well beyond proprietary alternatives."

Accessible HPC with AMD Developer Cloud

Perhaps ironically, AI development still rests very heavily on human capability. Specifically, developers, and in addition to ROCm, Elangovan describes the AMD Developer Cloud as a major step forward in making high performance computing accessible to anyone building AI.

[Developer Cloud] embodies our belief that innovation should be open, friction-free, and available on demand," he says. "The platform democratises access to AMD Instinct GPUs, bringing compute to ideas, rather than forcing ideas to fit limited infrastructure."

Essentially a platform play, Developer Cloud combines Instinct MI300X GPUs running ROCm, with fully configured environments for frameworks including vLLM, PyTorch, and JAX. Developers log in via GitHub, create containers and get on with training or benchmarking with minimal delay.

Strategically, Elangovan says the platform plays three roles in AMD's broader AI vision. "First, it serves as an onramp for developers, enabling anyone, from startups to academic researchers, to build and test AI workloads without capital investment.

"Second, it acts as an enablement layer for partners such as Hugging Face, Red Hat, and Modular, who use it to validate and optimise models for enterprise deployment.

"Third, it provides a feedback loop for our engineering teams. Real-world usage data from Developer Cloud informs ROCm software's optimisation roadmap, ensuring our software evolves in step with developer needs," he explains.

Ultimately, he says Developer Cloud is a strategic execution layer for AMD's open AI ecosystem. "By giving the global developer community direct access to our latest GPUs and open software stack, we're accelerating 'time to AI', lowering the entry barrier to innovation, and strengthening the connection between AMD hardware, ROCm software, and the people driving the next generation of intelligent systems."

There's no question that competition in the AI platform space is hot; literal trillions of investment dollars beckon. Asked why developers and enterprises should consider AMD versus more entrenched incumbents, Elangovan doesn't hesitate.  "Openness and performance go hand in hand. Developers no longer have to choose between flexibility and speed. With ROCm 7, AMD delivers leadership inference and training performance with broad framework compatibility and day one support for the most popular AI models, all within an open, vendor, neutral software ecosystem," he says.

"AMD gives IT leaders options of where and how to deploy, the flexibility to scale, and the efficiency and performance , backed by an ecosystem built on open standards. Our hardware portfolio spans from AMD Instinct GPUs in data centres to AMD Ryzen AI processors in client systems, all supported by a unified foundation of ROCm software and the AMD XDNA 2 architecture for AI acceleration. Combined with the AMD Developer Cloud and the leading partners [already mentioned], AMD gives customers performance, choice, and transparency."