.jpg){kind=link}
It looks like AMD is trying to raise our hopes with its upcoming MI325X processors. These processors are designed specifically for data centers and can handle intensive AI workloads. According to a Bloomberg report, AMD"s CEO Lisa Su says that these chips would outperform Nvidia"s popular H100 processors.
One of the key features of the MI325X system-on-chip (SoC) is its use of 256 GB of HBM3E memory, which is a 1.8x increase in memory capacity over its predecessor, the MI300X. The HBM3E memory has an impressive 6 terabytes per second (TB/s) bandwidth, which is crucial for handling large datasets and complex computations involved in AI tasks.
Nvidia"s older H100 processors, which debuted in 2022, used HBM3 instead of the newer HBM3E memory, with a memory bandwidth of 3TB/s. However, the upcoming Blackwell-based processors by Team Green use 288GB of HBM3E memory that has a peak bandwidth of around 13.8 TB/s.
Demand for advanced SoCs has risen over the past few months as tech companies are looking to train newer and advanced AI models. Last week, Nvidia and Foxconn announced that they"re building the world"s largest facility to manufacture Nvidia"s GB200 superchips based on the Blackwell architecture.
Nvidia has even begun shipping samples of these chips to partners and anticipates generating billions in revenue from them by the end of this year. This is also the reason why the company"s stocks are shooting upwards, only for it to become the world"s second most valuable company, just behind Apple.
Look what showed up at our doorstep.
— OpenAI (@OpenAI) October 8, 2024
Thank you to @nvidia for delivering one of the first engineering builds of the DGX B200 to our office. pic.twitter.com/vy8bWUEwUi
AMD is also trying to capture some of this growing market. AMD hasn"t announced when these new MI325X processors will be available, but production of these chips is expected to begin by the end of 2024. AMD says that the AI chip market is projected to reach a value of $500 billion by 2028.
Via Bloomberg