Meta, OpenAI, and Microsoft announced at AMD's investor conference last Wednesday that they will use AMD's latest artificial intelligence chip, InstactMI300X. The new chip released by AMD has over 150 billion transistors, with 2.4 times the memory and 1.6 times the bandwidth of the Nvidia H100 product. AMD is expected to increase the size of the AI accelerator market by nearly twice in 2027 compared to August. This indicates that the technology industry is actively seeking alternatives to expensive Nvidia GPUs. The MI300X is highly anticipated by the industry, as "Nvidia has been suffering for a long time." Nvidia GPUs are not only expensive, but also have limited supply quantities. If MI300X can be widely adopted, it is expected to reduce the cost of developing artificial intelligence models and create competitive pressure on Nvidia.
01 | How much faster than Nvidia's GPU? AMD stated that the MI300X is based on a completely new architecture and has significantly improved performance. Its biggest feature is having 192GB of cutting-edge high-performance memory, known as HBM3, which transfers data faster and can accommodate larger artificial intelligence models. Integrate the MI300X and its built system with Nvidia's (previous generation) flagship GPU
Compare H100. In terms of basic specifications, the floating-point operation speed of MI300X is 30% faster than H100, the memory bandwidth is 60% higher than H100, and the memory capacity is more than twice that of H100. Of course, the MI300X is more benchmarked against Nvidia's latest GPU
H200, although also leading in specifications, MI300X has less advantage over H200, with only one digit more memory bandwidth and nearly 40% more capacity than the latter.
02 | Can AMD replace Nvidia? According to CNBC, Meta (META. US), OpenAI, and Microsoft (MSFT. US) have announced that they will use AMD's latest AI chip, Instact
MI300X may indicate that these tech giants deploying AI tend to seek alternative products to replace NVDA. US's scarce and expensive AI chips. AMD's CEO Su Zifeng predicts that the market value of the AI chip market can reach over $400 billion by 2027, and believes that AMD can occupy a significant market share in it. AMD has not disclosed the pricing of the MI300X, but Nvidia currently costs approximately $40000 per chip, while Su Zifeng revealed that AMD's chips are lower than Nvidia's corresponding products. More importantly, AMD stated that it has improved the software suite ROCm for optimizing AI software stacks
6, in order to compete with Nvidia's industry standard CUDA software, which may be the reason why AI developers are currently more inclined towards Nvidia.
03 | Meta, Microsoft, and Oracle have all expressed support for AMD's new products. According to the latest report from research firm Omidia, Meta, Microsoft, and Oracle are all part of the 2023 Nvidia H100
Important buyers of GPUs. And these tech giants have all signed orders for MI300X. Meta plans to use MI300XGPU for artificial intelligence inference tasks, with Microsoft's Chief Technology Officer Kevin
Scott also stated that the company will deploy MI300X in its cloud computing service Azure, and in addition, Oracle's cloud computing service will also use MI300X. OpenAI will also use AMD in a software product called Triton
GPU. The support of these large companies is of great significance to AMD, not only directly reflected in the revenue generated by large orders, but also means that AMD's software ecosystem has the confidence to compete with Nvidia. However, looking into the future, if AMD wants to attract more Nvidia customers to switch to using its own chips, there is still more work to be done. AMD sued investors and partners, and the company has improved its software suite ROCm against NVIDIA CUDA. The CUDA suite has always been one of the main reasons why AI developers currently favor Nvidia. In addition, price is also important. AMD did not disclose the pricing of the MI300X on Wednesday, but it will definitely be cheaper than Nvidia's flagship chip, which has a unit price of around $40000. Su Zifeng stated that AMD's chips must have lower purchasing and operating costs than Nvidia in order to persuade customers to purchase.
04 Jingtai Opinion | It is not impossible for AMD to replace it. AMD has already revealed Instinct before
MI300A and MI300X
GPU's mass production in the fourth quarter progressed smoothly, and at the third quarter performance conference, it was pointed out that its AI progress was more than expected. It is expected that the data center GPU revenue in the fourth quarter will be about $400 million, and it will exceed $2 billion by 2024. The MI300 is expected to become the product in AMD's history that achieved sales of over one billion dollars in the shortest possible time. It can be seen that the current market has long had expectations for AMD's AI chips, but based on AMD's performance guidance, the strong development of AI may not have been reflected in the fourth quarter and will not be reflected in performance until the 2024 fiscal year. Nvidia, on the other hand, is experiencing an unprecedented increase in revenue and non accounting standard net profit. The strong revenue growth brought about by the shortage of AI chips has been reflected in this year's performance. Nvidia's revenue increased by 205.51% year-on-year and 34.15% quarterly to $18.12 billion in the third quarter of the 2024 fiscal year ending at the end of October 2023; Non accounting standard net profit increased by 588.19% year-on-year to $10.02 billion; And it is expected that the revenue in the fourth quarter will reach $20 billion, and the sustained strong demand for computing power and network will drive strong growth in its data center. Currently, Nvidia's AI chips are in short supply and expensive, which may drive users to switch to AMD. However, in the short term, it will take some time for AMD to replace Nvidia, mainly because Nvidia has already taken the lead in the AI chip field, accumulated many orders, and has a competitive advantage in the platform and ecosystem protection. It is not impossible for AMD to break these barriers and replace them, but it may be difficult to achieve in the short term.
