AMD has prepared for the release of chips for processing tasks of artificial intelligence – the launch of large language models (LLM).
The company introduced the Instinct MI300X Accelerator and the Instinct M1300A Processor (APU), which are designed for LLM training and management. Both new chips have a large amount of memory and improved energy efficiency compared to their predecessors. The company said the MI300X has 1.5 times more memory than the previous M1250X.
During the presentation, AMD CEO Lisa Su said that the MI300X “is the most powerful accelerator in the world.” She stated that the MI300X is comparable to the Nvidia H100 chips in LLM training, but performs better on the output side – 1.4 times better than the H100 when running Meta Llama 2, an LLM with 70 billion parameters.
AMD has partnered with Microsoft to host the MI300X on its Azure virtual machines. Microsoft CTO Kevin Scott also announced that Azure ND MI300X virtual machines, first introduced in November, are available in a preview version. Meta also announced that it will deploy MI300 processors in its data centers.
APU chips combine CPU and GPU for faster data processing. AMD said the MI300A offers more efficient computing, faster model training and a 30x increase in power efficiency. It is also emphasized that compared to the H100, the new chip has 1.6 times more memory. It also has a unified memory, so you no longer need to move data from different devices.
The MI300A APU is “now in production and being integrated into data centers.” Information about prices has not yet been announced.
AMD also announced the Ryzen 8040 mobile processor, which will add more built-in artificial intelligence features to the devices. The company said the 8040 series offers 1.6 times the performance of AI tasks than previous models and includes a Neural Processing Unit (NPU). In addition to II, the Ryzen 8040 processor offers other advantages. For example, it provides a 65% increase in video editing performance and a 77% increase in gaming performance compared to competing solutions.
AMD expects manufacturers such as Acer, Asus, Dell, HP, Lenovo and Razer to release their devices with Ryzen 8040 chips in the first quarter of 2024.
AMD also announced the general availability of the Ryzen AI software platform, which will allow developers to create AI models on Ryzen-powered laptops, offloading the models to the NPU so the CPU can reduce power consumption. Users will get support for basic models like the Whisper recognition model and LLMs like Llama 2.
Source: The Verge