AMD will host an event called "AdvancingAI" at 10 a.m. Pacific time on December 6 (2 a.m. on December 7, Beijing time). It is reported that,AMD will release InstinctMI300 series products at this event, including MI300A, MI300X, etc., and will highlight the company's growth momentum in the field of artificial intelligence hardware and software.At that time, AMD Chairman and CEO Su Zifeng will discuss AMD products and software with company executives, AMD partners and customers.

Analysts from investment bank Wedbush said in a recently released research report that at this event, AMD will compare the product performance of MI300 with Nvidia's H100, especially in terms of memory and bandwidth.

Wedbush also believes that,Microsoft may participate in AMD's event because Microsoft has announced that it will use AMD's new chips in its Azure cloud computing business.

According to multiple foreign media reports, in addition to Microsoft, many technology giants such as Amazon, Oracle, and IBM are also considering AMD products. Ming-Chi Kuo pointed out in a recent market briefing that AMD’s largest customer is Microsoft, accounting for more than 50% of shipments, followed by Amazon. In addition, Meta and Google are currently testing AMD products, and Meta is expected to have a high probability of becoming an AMD customer.

In addition, technology media Wccftech stated that AdvancingAI is an event focused on artificial intelligence, so mainstream consumer products are unlikely to be unveiled at this event, but information about consumer applications of artificial intelligence and data center business is expected to be released at this event.

How will the MI300 market perform?

Currently, many analysts are optimistic about the market performance of AMDMI300 series products.

Analysts at Jefferies said,With the launch of MI300, AMD will become the "first choice" for technology companies together with Nvidia.Bank of America also wrote in its latest report that AMD’s products are attractive to industry giants, ordinary enterprises, OEMs and AI startups.

In terms of shipments, Ming-Chi Kuo estimates that AMD’s AI chip shipments next year will be about 10% of Nvidia’s (based on CoWoS), mainly MI300A. If the cooperation between Microsoft and AMD goes smoothly and AMD obtains orders from Meta and Google,AMD's AI chip shipments are expected to reach 30% of Nvidia's (based on CoWoS) in 2025.

Taiwan Electronic Times pointed out in today’s report,AMD’s MI300 shipments are expected to reach 300,000 to 400,000 units next year.The largest customers are Microsoft and Google.

Su Zifeng also clearly pointed out before that a number of ultra-large-scale cloud service providers have promised to deploy MI300 chip products. Data center GPU products will bring $400 million in revenue to AMD in the fourth quarter, and revenue is expected to exceed $2 billion in 2024.MI300 will be the fastest product to achieve sales of US$1 billion since the company was founded.

From a performance perspective, MI300 is sufficiently competitive compared to competing products.

According to data released by AMD in June this year, products optimized for large language modelsMI300X has a memory of 192GB and a memory bandwidth of 5.2TB/s.InfinityFabric has a bandwidth of 896GB/s and 153 billion transistors.

in addition,The HBM (high bandwidth memory) density provided by MI300X is 2.4 times that of NVIDIA H100, and the HBM bandwidth is 1.6 times that of H100.

It is reported that Nvidia H100 has a memory capacity of 80GB and a bandwidth of 3.35TB/s. But the company's products are also constantly iterating. Recently, Nvidia's latest AI chip H200 has been released. The computing power of H200 is basically the same as that of H100, its memory capacity reaches 141GB, and its bandwidth is 4.8TB/s.

However, the barriers to the software ecosystem are still one of the challenges AMD needs to face. Karl Freund, chief analyst at Cambrian-AIResearch LLC, pointed out that AMD’s software ecosystem is not as complete as Nvidia’s. Training and running large AI models not only depends on GPU performance, but system design is also particularly important.

Su Zifeng once said, "We hope to become an important player in this (artificial intelligence) market." She mentioned that AMD is working on improvements in the AI ​​software suite.