On July 30th, U.S. time, chip giant AMD took the lead in presenting its report card for the second quarter of this year, igniting the chip stock market.

On July 31st, NVIDIA (NVDA.O) surged by 12.81%, Broadcom (AVGO.O) rose by 11.96%, ASML (ASML.O) increased by 8.89%, Qualcomm (QCOM.O) climbed by 8.39%, TSMC (TSM.N) jumped by 7.29%, Micron Technology (MU.O) gained 7.08%, and AMD (AMD.O) increased by 4.36%.

In the second quarter of 2024, AMD's total revenue was $5.835 billion, a year-over-year increase of 9%, and a sequential increase of 7%; net profit reached $265 million, a year-over-year increase of 881%, and a sequential increase of 115%.

The record revenue from the data center division was the backbone of AMD's strong performance growth in the second quarter. AMD's financial report showed that, thanks to the increase in shipments of AI GPUs, the revenue of AMD's data center business unit increased by 115% year-over-year to $2.8 billion.

Dr. Lisa Su, Chairman of the Board and CEO of AMD, stated during the performance communication meeting, "The company's AI chip sales were 'higher than expected', and the revenue from the MI300 chip exceeded $1 billion this quarter."

Advertisement

A reporter from Time Weekly contacted AMD regarding market competition and core competitiveness, but no response was received by the time of publication.

Breaking the $1 billion mark for MI300

Looking at the divisions, AMD's data center division's revenue increased by 115% year-over-year and by 21% sequentially, mainly due to the significant increase in shipments of AMD Instinct™ GPUs, as well as the strong growth in sales of the fourth-generation AMD EPYC™ CPUs.

The client business unit benefited from the sales growth of AMD Ryzen processors, with a quarterly turnover of $1.5 billion, a year-over-year increase of 49%, and a sequential increase of 9%.

However, AMD's gaming division and embedded division saw a decline in revenue. The gaming division's quarterly turnover was $648 million, a year-over-year decrease of 59%, and a sequential decrease of 30%, mainly due to the reduction in semi-custom business income. The embedded division's quarterly turnover was $861 million, a year-over-year decrease of 41%, mainly due to customers continuing to adjust inventory levels, with a sequential increase in turnover of 2%.The contribution of the MI300 series of chips is particularly significant. "We delivered a record data center GPU revenue for the third consecutive quarter, with the MI300 series revenue exceeding $1 billion for the first time," said Dr. Lisa Su.

The MI300 series of chips is a product launched by AMD at the end of 2023, including two chips: MI300X and MI300A. The MI300X is a GPU accelerator, while the MI300A is an APU (Accelerated Processing Unit) that integrates both CPU and GPU functions.

A research report published by Minsheng Securities shows that in terms of computing power, the TF32 floating-point performance of the MI 300X is 653.7 TFlops, and the FP16 and BF16 computing performance is 1307.4 TFlops, with FP8 and INT8 computing performance at 2614.9 TFlops, all of which are 1.3 times that of NVIDIA's H100. In terms of memory, the memory configuration of the MI 300X is 2.4 times that of NVIDIA's H100, with a peak storage bandwidth that is also 2.4 times as much. The inference speed when running Bloom is 1.6 times that of the H100, and the inference speed when running Llama2 is 1.4 times as fast. In terms of power consumption, the overall power consumption of the MI300X is controlled at 750W, which also has an advantage over NVIDIA's H100. Additionally, in terms of pricing, Dr. Lisa Su stated that the purchase and operational costs of the MI300 series chips will be lower than those of NVIDIA.

During the second quarter earnings call, Dr. Lisa Su mentioned that Microsoft has expanded its use of the MI300X accelerator to support GPT-4 Turbo and multiple consortium services, including Microsoft 365 Chat, Word, and Teams. Microsoft also became the first hyperscale user to announce the public availability of MI300X instances.

She stated that the new Azure VM leverages the computing performance and memory capacity of the MI300X and combines it with the latest ROCm software, offering leading inference price-performance when running the latest edge models, including GPT-4.

In addition to Microsoft's expanded use of AMD's chips, on July 30, Apple (AAPL.O) stated in a technical paper that the two artificial intelligence models supporting its artificial intelligence system, AppleIntelligence, were pre-trained on cloud chips designed by Google. Google's Tensor Processing Units were initially created for internal workloads but are now gaining wider adoption. Related reports suggest that in terms of AI training, some large technology companies may be looking for and have found alternatives to NVIDIA's graphics processing units.

The further maturation of AMD and Google's chips has made the market realize that NVIDIA's competitors are getting stronger, and NVIDIA's stock price plummeted by 7% on July 30, with its market value evaporating nearly $200 billion overnight.

Wu Quan, Dean of the HuaXin JinTong Semiconductor Industry Research Institute, told a reporter from Times Weekly that Apple's pre-training on Google's cloud chips is more about valuing Google's unique advantages in speed, energy efficiency, and data optimization in deep learning and machine learning.

Wu Quan believes that there is both competition and cooperation among tech giants, and this relationship sometimes manifests as cross-competition and cooperation. For example, the cooperation between Google and Apple, AMD and Microsoft, and NVIDIA and Meta. Although they intersect in fields such as artificial intelligence, GPUs, and TPUs, there are often differences in specific directions, which certainly reduces direct competition and leverages their respective strengths.

"The cooperation between these companies indicates that in terms of AI acceleration and large model training, their competition is staggered, and the space for cooperation is greater than that of competition," said Wu Quan. The pre-training process is part of their experimentation, which helps to explore different dimensions of AI, and whether in cloud or terminal applications, the giants are all seeking differentiated development. However, this cooperation does not mean that any party will give up cooperation with other companies.Supply Chain Issues Remain a Challenge

LLisa Su anticipates that MI300 revenue will continue to increase in the second half of the year, with growth expected in both the third and fourth quarters. AMD will continue to expand deployments with existing customers.

Where Does AMD's Advantage Lie as NVIDIA Takes the Lead?

Lisa Su believes that a significant feature of MI300 is its leading position in the industry in terms of memory bandwidth and capacity. From an inference perspective, AMD's early deployments have primarily been focused on inference, where it has performed exceptionally well. Concurrently, AMD is optimizing the ROCm software stack to make it easier for customers to train on AMD platforms.

"I anticipate that we will continue to increase our deployments in training. Inference will be larger than training, but from AMD's perspective, I see both inference and training as growth opportunities," Lisa Su stated.

In Wu Quan's view, AMD is more "comprehensive" compared to NVIDIA. AMD offers a range of computational solutions that include CPUs, GPUs, NPUs (Neural Processing Units), DPUs (Data Processing Units), and MPEs (Multi-Processing Engines), which is a notable advantage for AMD. Although comprehensiveness is an advantage, it may also present challenges in terms of complexity and cost.

From NVIDIA's perspective, its main "shortcoming" might lie in the CPU domain. NVIDIA's traditional strength is in GPUs, but it is relatively weaker in the CPU area. However, NVIDIA is also actively working to address this deficiency by expanding its CPU capabilities.

Wu Quan mentioned that AMD's MI300 series is not significantly different in performance from NVIDIA's H100 series chips. While it may not have an absolute advantage in terms of comprehensiveness, considering the price difference, AMD's chips may still offer better value for money. It can be said that the factor of price is very likely to influence the choice between AMD and NVIDIA for Microsoft and other companies.

Of course, Microsoft uses not only AMD's chips but also NVIDIA's products, reflecting the diversity of market applications. "As AI and High-Performance Computing (HPC) demands continue to grow, companies are starting to seek more flexible computational power configurations. In the past, there was a general demand for high-end computational power, but with the proliferation of AI applications, the market's demand for medium or balanced computational power is increasing. This has created the so-called 'olive-shaped' demand distribution, which means that the demand in the mid-range market is expanding, not just the high-end market," Wu Quan stated.

However, supply chain issues remain a significant challenge for AMD. Lisa Su stated that the company is working hard to improve supply chain conditions and expects supply to continue to increase in the second half of the year. Despite this, the overall supply chain for MI300 remains tight, and it is anticipated to stay so until 2025.AMD also announced that it will launch the MI325 later this year. The MI325X will leverage the same infrastructure as the MI300, offering double the memory capacity and 1.3 times the peak computing performance. Additionally, AMD plans to introduce the MI350 series in 2025, based on the new CDNA 4 architecture, which is expected to deliver a 35-fold performance increase over CDNA 3, competing with NVIDIA's Blackwell solutions. In 2026, AMD intends to release the MI400 series, powered by the CDNA "Next" architecture.

Dr. Lisa Su stated that data centers will continue to be the main driver of revenue growth in the second half of the year, with gross margins expected to continue to improve. In the long term, the gross margin of the MI300 is anticipated to exceed the company's average. AMD forecasts that data center GPU revenue will surpass $4.5 billion in 2024, up from the $4 billion expectation in April.