AMD's Instinct MI300 has quickly established itself in the AI accelerator market, contributing substantially to revenue growth. While it can't match Nvidia's dominant position, AMD's progress points to a promising outlook in AI hardware.
The newly launched Instinct MI300 graphics card from AMD has become a major source of revenue for the company, rivaling the revenue of its entire CPU product line. It is an important milestone for AMD in the fiercely competitive AI hardware market, where the company has often trailed industry leader Nvidia.
On a recent earnings call, AMD CEO Lisa Su said the data center GPU segment, driven mainly by the Instinct MI300, has exceeded initial expectations. "We're really seeing our [AI] GPU business gradually reaching the same scale as our CPU business," she said.
The achievement is especially impressive because AMD's CPU segment spans a wide range of products for servers, cloud computing, desktops and laptops.
The Instinct MI300, introduced in November 2023, is AMD's first truly competitive graphics card for AI inference and training workloads. Although it launched only recently, the MI300 has quickly drawn attention in the market.
Financial analysts estimate that AMD's AI GPU revenue last September surpassed $1.5 billion, with the following months likely to be even stronger.
The MI300's specs are competitive in the AI accelerator market, offering substantial improvements in memory capacity and bandwidth.
The MI300X graphics processor has 304 GPU compute units and 192 GB of HBM3 memory, delivering a theoretical peak memory bandwidth of up to 5.3 TB/s. It reaches a peak FP64/FP32 matrix compute performance of 163.4 TFLOPS and peak FP8 performance of up to 2,614.9 TFLOPS.
The MI300A APU integrates 24 Zen 4 x86 CPU cores alongside 228 GPU compute units. It has 128 GB of unified HBM3 memory and the same theoretical peak memory bandwidth as the MI300X, 5.3 TB/s. The MI300A's peak FP64/FP32 matrix compute performance reaches 122.6 TFLOPS.
The success of the Instinct MI300 has attracted interest from major cloud providers such as Microsoft. The Windows giant recently announced general availability of its ND MI300X VM series, which uses eight AMD MI300X accelerators. Earlier this year, Microsoft Executive Vice President of Cloud and AI Scott Guthrie said AMD's accelerators are currently the most cost-effective GPUs based on their performance in Azure AI Services.
Impressive as AMD's growth in the AI GPU market is, the company still hasn't caught up with Nvidia in overall market share. Analysts predict that Nvidia could reach $50 to $60 billion in AI GPU sales revenue in 2025, while AMD could reach a maximum of only $10 billion.
Still, AMD Chief Financial Officer Jean Hu noted that the company is working with more than 100 customer relationships for the MI300 lineup, including major tech companies such as Microsoft, Meta and Oracle, as well as a broad set of enterprise customers.
The rapid success of the Instinct MI300 in the AI market raises questions about AMD's future direction for consumer graphics cards, since higher revenue from AI accelerators could affect how the company allocates its resources. However, AMD has confirmed its commitment to the consumer GPU market, with Su announcing that the next-generation RDNA 4 architecture, likely to debut on the Radeon RX 8800 XT, will be unveiled in early 2025.
Source: Techspot
