Advanced Micro Devices Inc. is expanding its product portfolio with an array of new processors designed for use in data centers and personal computers.
AMD Chief Executive Officer Lisa Su (pictured) debuted the processors late Wednesday at the CES 2023 consumer electronics event in Las Vegas.
Expanded data center portfolio
AMD is expanding its data center chip lineup with the addition of the Instinct MI300, a so-called adaptive processing unit or APU that features 146 billion transistors. It combines elements of a central processing unit and a graphics processing unit in a single package. According to AMD, the chip is designed to run high-performance computing applications and artificial intelligence software.
The MI300 is built from chiplets, compact computing modules that can be mixed and matched in different ways to create specialized processors. There are 13 such modules in the MI300. They’re organized into two layers that are stacked atop one another in a three-dimensional configuration.
According to AMD, the first layer of the MI300 comprises nine CPU and GPU chiplets made using a five-nanometer process. The CPU modules are based on the company’s latest Zen 4 processor architecture. The GPU modules, in turn, feature the CDNA 3 architecture that powers AMD’s data center graphics cards.
The MI300‘s CDNA 3 and Zen 4 circuits are stacked atop another semiconductor layer that includes four more chiplets. According to AMD, the latter chiplets manage supporting tasks such as data input and output operations. They’re produced using a six-nanometer process.
The MI300 stores the data it processes in a 128-gigabyte onboard memory pool. The chip’s GPU and CPU modules share the memory, which AMD says makes them more efficient than traditional data center processors.
Traditional CPUs and GPUs are implemented as separate chips and, as a result, store information in separate memory circuits. Each chip keeps its own copy of the data being processed. AMD says that the MI300’s shared memory removes the need to create multiple data copies for the CPU and GPU odules, which makes processing more efficient.
The company claims that the MI300 can train AI models eight times faster than its current-generation MI250X graphics card. Additionally, the chip promises to provide a fivefold performance-per-watt increase for AI workloads.
“The MI300 can train much larger AI models faster, at lower costs and with less power,” Su during a CES keynote. “MI300 can reduce the time to train these models from months to weeks with dramatically lower energy costs. And more importantly, it can also support much, much larger models that can be used for even more advanced and more powerful AI services in the future.”
AMD debuted the MI300 at CES alongside another new data center chip called the Alveo V70. It’s optimized to perform inference, or the task of running AI applications in production after they’ve been trained. AMD says that the Alveo V70 is based on an architecture developed by its Xilinx subsidiary and can perform up to 400 trillion calculations per second.
New desktop and laptop chips
Alongside the data center processors that it previewed at CES, AMD announced several new additions to its portfolio of consumer chips. Some of the chips are designed for use in desktops, while others target the laptop market.
AMD is rolling out a new lineup of desktop CPUs known as the Ryzen 7000X3D series. The lineup comprises three chips that offer 8 to 16 cores with a maximum clock speed of 5.7 gigahertz. AMD says that the chips can perform file compression 52% faster than Intel Corp.’s high-end Core i9-13900K processor, while running video games up to 25% faster.
The Ryzen 7000X3D series is one of the first CPU lineups from AMD to feature its V-Cache technology. The technology, which was first demonstrated in May 2021, makes it possible to place a cache module atop a CPU in a three-dimensional configuration.
A CPU’s cache stores data near its logic circuits to increase performance. When data is kept near the logic circuits, it can be retrieved faster, which speeds up calculations. AMD’s V-Cache technology further reduces the distance that information must travel before it can be processed.
“The Ryzen 7000X3D is our first 16-core Ryzen processor with V-Cache technology and our fastest 3D stacked chip ever,” Su stated.
For laptops, AMD is rolling out the new Ryzen 7040 processor series. It’s based on the company’s latest Zen 4 CPU architecture and will be manufactured using a four-nanometer process. Additionally, AMD has added an AI module that promises to run machine learning applications 20% faster than the M2 chip in Apple Inc.’s newest MacBooks.
“This chip uses four nanometer process technology and it has more than 25 billion transistors, almost twice as many as our Ryzen 6000 generation,” Su detailed.
AMD is rolling out the latest additions to its consumer CPU portfolio alongside a new series of graphics cards for the laptop market. The Radeon RX 7000 series features four GPUs headlined by the RX 7600M XT. According to AMD, the RX 7600M XT can run some video games up to 31% faster than Nvidia Corp.’s RTX 3060 desktop graphics card.
The chip features 2,048 shader processors, which have a similar role as the cores of a CPU. The shader processors are supported by a 32-megabyte cache and eight gigabytes of GDDR6 memory. According to AMD, the RX 7600M XT is manufactured using a six-nanometer process.