Lately, it seems like it's Nvidia's world, and everyone — and certainly anyone in tech and the explosive AI industry — is living in it. Between its well-timed market entry, leading hardware research, and a robust software ecosystem built for its GPUs, the company is dominating AI development and the stock market. Its latest earnings report late today showed quarterly sales tripled, further boosting its share price.
However, longtime rival chipmaker AMD is still pushing hard for a foothold in AI, telling the people behind key technologies in the nascent space to work on AMD hardware as well. can.
“I just wanted to remind everyone that if you're using PyTorch, TensorFlow or JAX, you can use your notebook or scripts, they'll only run on AMD,” said Ian Ferreira, senior director of AMD at Microsoft. Announced at the Build 2024 conference. Wednesday “Also works with BLLM and Onyx.”
The company used its time on stage to show examples of how AMD GPUs can run powerful AI models like Stable Diffusion, and Microsoft Phi, to handle computationally-intensive training tasks without relying on Nvidia's technology or hardware. Can perform effectively.
Conference host Microsoft reinforced the message by announcing the availability of AMD-based virtual machines on its Azure cloud computing platform using the company's high-speed MI300X GPUs. 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's infrastructure.
ML libraries are supported by AMD. Image: Microsoft. Youtube
Nvidia's proprietary CUDA technology, which includes a complete programming model and API designed specifically for Nvidia GPUs, has become the industry standard for AI development. AMD's main message, therefore, was that its solutions could go right into a single workflow.
Seamless compatibility with existing AI systems could be a game changer, as developers can now take advantage of AMD's less expensive hardware without overhauling their code base.
“Of course, we understand that you need more than just the framework, you need a bunch of upstream stuff, you need experimental equipment, distributed training – it's all functional and on AMD. Works,” Ferreira assured.
He then demonstrated how AMD handles various tasks, from running smaller models like ResNet 50 and Phi-3 to debugging and training GPT-2—all using the same code. are those that run Nvidia cards.
Image: Microsoft. Youtube
One of the key advantages AMD touts is the ability to efficiently handle large language models.
“You can load 70 billion parameters on a GPU, eight of them with this example,” he explained. “You can have eight different Llama 70B loads, or take a larger model like the Llama-3 400Bn and deploy it all at once.”
Challenging Nvidia's dominance is no easy feat, as the Santa Clara, California-based company has fiercely guarded its turf. Nvidia has already taken legal action against projects trying to provide CUDA compatibility layers for third-party GPUs like AMD's, arguing that it violates CUDA's terms of service. This has limited the development of open source solutions and made it difficult for developers to adopt alternatives.
AMD's strategy to circumvent Nvidia's blockade is to leverage 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 native support and additional acceleration tools such as Transformers for ROCm-enabled GPUs, Optimum-Benchmark, DeepSpeed ROCm-enabled GPUs, GPTQ, TGI, and more. offers.
Ferreira also pointed out that this integration is local, eliminating the need for third-party solutions or intermediaries that can make the process less efficient.
“You can take your existing notebook, your existing scripts, and you can run them on AMD, and that's important, because a lot of other accelerators would have required transcoding and all kinds of precompiling scripts,” he said. It is,” he said. “Our stuff just works out of the box — and it's really, really fast.”
While AMD's move is undoubtedly bold, it will be quite a challenge to unseat Nvidia. Nvidia isn't resting on its laurels, constantly innovating and making it difficult for developers to migrate from the de facto CUDA standard to new infrastructure.
However, with its open source approach, strategic partnerships, and focus on native compatibility, AMD is positioning itself as a viable alternative for developers looking for more options in the AI hardware market.
Edited by Ryan Ozawa.