AMD's CPU business has made a huge turnaround in the past few years. Not only does it have to catch up with its old rivals in terms of share, but it also surpasses it in terms of performance. Not long ago, it was claimed to have achieved the status of the strongest CPU in almost all fields. The next focus is of course GPU graphics cards. AMD is also the only two main supplier in this field, but NVIDIA has an overall advantage in both game cards and AI graphics cards, and even captured 95% of the AI market.
Considering the current value and profits of AI graphics cards, how can AMD catch up with its opponents in this field? CEO Su Zifeng also talked about this topic in an interview recently,She was asked about the topic of whether she hopes AMD will become the GPU of choice for companies such as OpenAI and Meta. Su Zifeng said that this is like our position in CPU today.
Judging from the response, AMD's ambitions are not small, but Su Zifeng also said that of course they hope to make a difference in the field of AI, but they are not impatient about it.
In the meaning of these polite words, AMD obviously understands that competing with NVIDIA in the AI market is a long-term process and will not be achieved easily.
AMD's AI graphics cards this year are mainly the MI350 series, but the real killer is next year's MI400 series, which not only upgrades the technology and architecture, but also brings a new generation of HBM4, with a single card capacity of up to 432GB and a bandwidth of 19.6TB/s.
For comparison, it is 50% and 145% higher than the 288GB HEM3E and 8TB/s of the MI350 series respectively, and the average memory bandwidth of each CU unit is also increased to 300GB/s.
In addition to major hardware upgrades,AMD has also made great efforts in the software ecosystem, and the performance of the ROCm7 platform has increased by more than 3 times.It also fully supports Linux and Windows platforms, which will help attract developers to join AMD's AI platform to develop applications.
