AMD chips can now do the AI ​​work that Nvidia Tech does.

Lately, it seems to be Nvidia's world, and everyone—and certainly anyone in the tech and explosive AI industry—is living in it. Between timely market entry, leading hardware research, and a robust software ecosystem for its GPUs, the company dominates AI development and the stock market. Its latest earnings report late today showed that quarterly sales tripled, sending the stock higher.

However, longtime rival chipmaker AMD is still pushing hard to establish a foothold in AI, telling the developers behind key technologies that are still in their infancy that they can run their jobs on AMD hardware.

“I just wanted to remind you that if you're using PyTorch, TensorFlow or JX, you can use your notebooks or scripts. It's the same with BLLM and Onyx,” AMD senior director Ian Ferreira said earlier at the Microsoft Build 2024 conference.

The company used its time on stage to show examples of how AMD GPUs can use powerful AI models such as Stable Diffusion, and Microsoft Phi to efficiently perform computationally intensive training tasks without relying on NVIDIA technology or hardware.

Conference host Microsoft reinforced its message by announcing that AMD-based virtual machines using the company's accelerated MI300X GPUs will be available on its Azure cloud computing platform. The chips were announced last June, began shipping in the new year, and were recently implemented in Microsoft Azure's OpenAI service and Hugging Face infrastructure.

Nvidia's proprietary CUDA technology, which includes a complete programming model and an API designed specifically for Nvidia GPUs, has become the industry standard for AI development. AMD's main message was that its solutions can fit into similar workflows.

Its seamless compatibility with existing AI systems can be a game changer.

“Obviously, we understand that you need more than frameworks, that you need more flow stuff, more testing stuff, distributed training — all of that is enabled and running on AMD,” Ferreira confirmed.

He then showed how AMD handles a variety of tasks, from running smaller models like ResNet 50 and Phi-3 to fine-tuning and training GPT-2 – using code that all Nvidia cards run.

One of AMD's stated key advantages is its ability to efficiently handle large language models.

“You can load up to 70 billion parameters on a single GPU, with eight in this example,” he explained. “Eight different Llama 70Bs are installed, or you can take a larger model like the Llama-3 400Bn and deploy that at once.”

Challenging Nvidia's dominance is no easy task as the Santa Clara, California-based company has firmly secured its turf. Nvidia has taken legal action against projects that attempt to provide CUDA compatibility layers for third-party GPUs such as AMD's, claiming that it violates the CUDA terms of service. This has limited the growth of open source solutions and made it difficult for developers to adopt alternatives.

AMD's strategy to avoid Nvidia's blockage is to use its open source ROCm framework, which competes directly with CUDA. The company is making significant progress in this regard by partnering with Hugging Face, the world's largest repository of open source AI models, to provide support for running code on AMD hardware.

This partnership has already yielded promising results, with AMD offering additional overclocking tools such as ONNX models with ROCm-powered GPUs, Optimum-Benchmark, DeepSpeed ​​​​for ROCm-powered GPUs transformers, GPTQ, TGI and others.

Ferreira also pointed out that this integration is native, eliminating the need for third-party solutions or making processes less efficient.

“You can take your existing notebooks, your existing scripts, and you can run them on AMD, and that's important, because a lot of others require instant transcoding and all kinds of precompile scripts,” he said. “Our stuff works out of the box – and it's really fast.”

While AMD's move is bold, dethroning Nvidia will be a big challenge. Navia isn't resting on its laurels, it's constantly innovating and making it difficult for developers to move from the de-facto CUDA standard to a new infrastructure.

However, with its open source approach, strategic partnerships and focus on native compatibility, AMD is positioning itself as an alternative for developers looking for more options in the AI ​​hardware market.

Edited by Ryan Ozawa.