AMD and NVIDIA's battle spreads to the AI field! Meta and Oracle announced a huge investment in the acquisition of AMD AI chips.

Zhitong App learned that AMD, one of the major players in the CPU and GPU industries and a competitor of NVIDIA (NVDA.US), held an "Advancing AI" conference on Wednesday. AMD officially announced the launch of its flagship AI GPU accelerator, MI300X, which means that the intense competition between AMD and NVIDIA has expanded from the PC field to the AI field. AMD's test data shows that its overall performance is 60% higher than NVIDIA's H100. In addition to the expected release of new products such as Instinct MI300X and MI300A, some global leaders in the AI field, such as OpenAI, Microsoft, and Meta, also attended the event and expressed their intention to deploy a large number of AMD Instinct MI300X in the future.

At the "Advancing AI" conference, the Instinct MI300X accelerator took the stage first. Since the parameters of the MI300 series AI chips were announced in the product promotion activities six months ago, this "Advancing AI" conference focused more on the performance of the entire system in practical applications and a comprehensive performance comparison with NVIDIA's H100 AI chip, which is the most popular in the AI training/inference field. In addition, AMD has abruptly raised its global AI chip market size expectation for 2027 from $150 billion to $400 billion.

Global leaders in the AI field, OpenAI, Microsoft (MSFT.US), and Meta (META.US), stated at the AMD event on Wednesday that they will use AMD's latest AI chip, Instinct MI300X. This is the most obvious sign so far that global tech companies are looking for an alternative to the expensive NVIDIA H100 AI chip. The NVIDIA H100 is crucial for creating and deploying generative artificial intelligence (generative AI) applications such as OpenAI's ChatGPT. Now, the NVIDIA H100, which has almost a monopoly position in the AI chip field, finally has a strong competitor, which is AMD Instinct MI300X.

If AMD's latest high-end AI chips start shipping in early next year, they will be able to meet the computing power needs of tech companies and cloud service providers for building large-scale AI models, while reducing the cost of developing AI models for tech companies. This will undoubtedly pose significant competitive pressure on NVIDIA's AI chip sales, which have been soaring.

Lisa Su, CEO of AMD, said on Wednesday, "The interests of potential large customers are basically focused on large processors and large GPUs in the field of cloud computing." AMD announced that the MI300X is based on a new architecture, which typically brings significant performance improvements. The most notable feature of AMD's new AI chip is its cutting-edge high-performance HBM3 memory with 192GB, which enables faster data transfer and can accommodate larger-scale AI models.

During the presentation, Su Zifeng, affectionately known as "Mom Su" by AMD fans, directly compared the MI300X and its system with Nvidia's main AI chip, the H100. "This performance directly translates into a better user experience," Su Zifeng said. "When you make demands on the model, you want it to respond faster, especially as the complexity of the response increases."

The main challenge for AMD is whether companies that have been relying on Nvidia's software and hardware will invest time and money to trust another AI chip supplier. On Wednesday, AMD told investors and partners that the company has improved its software suite called ROCm to compete with Nvidia's industry-standard CUDA software, partially addressing a key deficiency that AMD has faced in the AI chip field - the software and hardware ecosystem, which has been one of the main reasons why AI developers currently prefer Nvidia's products.

AMD did not disclose the pricing of the MI300X on Wednesday, but Nvidia's AI chips are priced as high as $40,000 per unit. Su Zifeng stated that AMD's chips must have lower purchasing and operating costs than Nvidia's chips in order to convince potential large customers to make a purchase.

Although the content of the presentation was largely in line with market expectations, it did not seem to boost AMD's stock price. Against the backdrop of a collective decline in US tech stocks on Wednesday, AMD's stock price turned from gains to losses before the end of the presentation, ultimately closing down more than 1%. However, it rose more than 1% after hours.

Due to AMD's upcoming high-performance AI chip that will compete with Nvidia's H100, Wall Street analysts generally have a bullish outlook on AMD's stock price. The consensus rating and target price compiled by Seeking Alpha show that Wall Street analysts have a "Buy" rating on AMD, with an average target price expectation of $132.24, implying a potential increase of up to 13% in the next 12 months, with the highest target price being $200.

Outperforming Nvidia's H100! AMD significantly raises market size expectations

Comparative performance data against Nvidia's H100, the most popular AI chip in the AI training/inference field, shows that the MI300X offers performance improvements of up to 20% in general LLM core TFLOPs in FlashAttention-2 and Llama 2 70B. **From a platform perspective, comparing the 8x MI300X solution with the 8x NVIDIA H100 solution, AMD found that the performance improvement of Llama 2 70B is much greater, reaching 40%, and the improvement under the Bloom 176B benchmark is as high as 60%.

Specific performance comparison data between AMD Instinct MI300X and NVIDIA H100:

In a 1v1 comparison, the overall performance is 20% higher than H100 (Llama 2 70B)

In a 1v1 comparison, the overall performance is 20% higher than H100 (FlashAttention 2)

In an 8v8 server, the overall performance is 40% higher than H100 (Llama 2 70B)

In an 8v8 server, the overall performance is 60% higher than H100 (Bloom 176B)

The driving force behind the latest MI300 AI chip is AMD ROCm 6.0. The software stack has been updated to the latest version, with powerful new features, including support for various AI workloads such as generative AI and large language models (LLM).

Memory is another area of significant upgrade for AMD, with the HBM3 capacity of MI300X increased by 50% compared to its predecessor MI250X (128 GB). To achieve a memory pool of up to 192GB, AMD equips MI300X with 8 HBM3 stacks, each stack being 12-Hi and integrating 16Gb ICs, with each IC having a capacity of 2GB, or each stack having a capacity of up to 24GB.

This level of memory capacity will provide up to 5.3TB/s bandwidth and 896GB/s Infinity Fabric bandwidth. In comparison, the upcoming NVIDIA H200 AI chip offers 141GB capacity, while Intel Gaudi3 will offer 141GB capacity. In terms of power consumption, the rated power of AMD Instinct MI300X is 750W, which is a 50% increase over Instinct MI250X's 500W and 50W more than NVIDIA H200.

AMD made bold predictions about the future market size of AI chips in the field of AI chips on Wednesday, believing that the AI chip market will expand rapidly. Specifically, AMD expects the overall size of the AI chip market to exceed $400 billion by 2027, nearly double the $150 billion it provided a few months ago, highlighting the rapid change in expectations for AI hardware among major global companies, as these companies are rapidly deploying new AI products.

Which tech giants will use the MI300X?

At the press conference on Wednesday, AMD stated that it has signed agreements with some of the tech companies that have the greatest need for GPUs to use this chip. According to a recent report by research firm Omidia, Meta and Microsoft are the largest buyers of NVIDIA's H100 AI chip in 2023.

At the press conference, Meta, the parent company of Facebook and Instagram, publicly stated that it will heavily use the MI300X GPU for processing AI inference workloads, such as processing AI stickers, image editing, and operating its AI assistant. They also mentioned combining the ROCm software stack to support AI inference workloads.

Microsoft's Chief Technology Officer, Kevin Scott, publicly stated that Microsoft will provide technical access to the MI300X chip through its Azure network services. In addition, news on the same day showed that Microsoft will evaluate the demand for AMD's AI chip products and assess the feasibility of adopting this new product.

OpenAI, the developer of ChatGPT, stated that it will support AMD MI300 and other GPUs in an important software product called Triton. Triton is not a large language model like GPT-4, but it is a very important product in the field of AI research.

One of NVIDIA's largest customers, Oracle, stated that it will use the Instinct MI300X accelerator in its cloud computing service system and plans to develop generative AI services based on AMD Instinct MI300X.

AMD has not yet predicted the long-term sales of this AI chip. Currently, it has only provided expectations for 2024, with an estimated total revenue of $2 billion from data center GPUs in 2024. In just the last quarter, NVIDIA's data center business revenue exceeded $14 billion, although this figure includes business other than GPUs. However, AMD stated that in the next four years, the total market size of the AI chip field may climb to $400 billion, which is twice the company's previous forecast. Su Zifeng also told reporters that AMD does not believe it needs to defeat NVIDIA to achieve good results in the market. When it comes to the artificial intelligence chip market, Su Zifeng told reporters, "It is obvious that NVIDIA currently occupies the majority." "We believe that by 2027, the AI chip market could exceed $400 billion, and we will play an important role."