"AMD focuses on software before the release of MI300X"

In June of this year, AMD CEO Lisa Su announced the company's plans to soon launch the Instinct MI300X, which is a replacement for NVIDIA's H100. Considering the global shortage of graphics cards, the timing couldn't be better. In the battle between NVIDIA and Intel, AMD is quietly working and leading in the field of artificial intelligence competition. Currently, the focus is on software, as hardware plans are already in progress.

Recently, AMD decided to support the artificial intelligence software developer Moreh in collaboration with the South Korean telecommunications company KT. This start-up, based in Santa Clara, raised $22 million in Series B funding, bringing the total funding to $30 million. Currently, the company is focused on developing ROCm, a software alternative to NVIDIA's CUDA in the field of artificial intelligence.

Recently, Vamsi Boppana, Senior Vice President of Artificial Intelligence at AMD, stated that the Radeon Open Compute Platform (ROCm) is the company's top priority. "We actually have more resources invested in software. Lisa Su has been very clear that she wants significant and sustained investment in software," he said.

As for Moreh, the company's flagship product MoAI is compatible with PyTorch, TensorFlow, and other applications that previously only ran on NVIDIA hardware. With AMD's investment, the company will further enhance and accelerate AMD's competition in the field of artificial intelligence software.

KT has been working with Moreh since 2021, providing support for its scalable artificial intelligence infrastructure on AMD GPUs and equipped with MoAI software. Currently, the MoAI platform primarily supports AMD's ROCm. KT uses the AMD Instinct MI250 accelerator in conjunction with MoAI, claiming to be 116% faster than NVIDIA's A100.

Software is King

Brad McCredie, Vice President of AMD's Data Center GPU and Accelerated Processing, stated in a statement, "The artificial intelligence software ecosystem supporting AMD AI hardware continues to grow, providing choices for data scientists and other AI users in building AI models and solutions that drive industry growth."

Companies like Lamini have been eagerly awaiting the release of the MI300X with 192GB HBM, which will improve the performance of their models. Lamini states that AMD's ROCm is ready and claims that it "has tremendous potential to accelerate LLM fine-tuning and other areas of artificial intelligence advancement, similar to or even greater than CUDA."

In addition to Moreh, AMD has made significant breakthroughs through the acquisition of the open-source AI software company Nod.ai. "This acquisition is expected to greatly enhance our ability to provide open software to AI customers, enabling them to easily deploy high-performance AI models optimized for AMD hardware," said Boppana.

AMD has been investing in software. In August of this year, the company also announced the acquisition of the French AI start-up Mipsology, which is also a long-term partner of AMD, developing AI software for chip manufacturers, similar to Nod.ai.

Boppana then wrote, "The team will help develop our complete AI software stack, expand our open ecosystem of software tools, libraries, and models, and pave the way for simplified deployment of AI models running on AMD hardware."

Clearly, AMD has found its software path.

ROCm is preparing for MI300X

Boppana stated at the AI Hardware Summit in September that AMD is currently attracting significant customers, which largely determines the company's current strategy. "The plane is flying, so we can't dismantle the engine. However, we are definitely doing something at the foundational level to achieve more unity in our stack," he said in an interview.

He also stated that many customers have already received MI300 samples and are testing their functionality with ROCm. For example, the start-up MosaicML, acquired by DataBricks, has been using AMD's hardware (but with MI250) since the beginning of this year and sharing its system with NVIDIA.

The upcoming supercomputer El Capitan at the Lawrence Livermore National Laboratory in California has installed an unknown number of MI300, further fueling the hype surrounding its release.

Boppana further emphasized that the performance of ROCm is crucial for the success of the upcoming hardware. Although some companies and developers have accepted it, ROCm is still in the early stages of development, "frankly, we still have some areas to improve."

Therefore, while CUDA may currently be dominant, AMD is no longer a passive player. With ROCm, AMD intends to surpass NVIDIA's position in the field of artificial intelligence competition.