AMD unveils new lineup of data center and consumer chips

Superior Micro Gadgets Inc. is increasing its product portfolio with a variety of latest processors designed to be used in information facilities and private computer systems.

AMD CEO Lisa Su (pictured) logged in processors on the CES 2023 shopper electronics occasion in Las Vegas late Wednesday.

Expanded information middle portfolio

AMD is increasing its information middle chip lineup with the addition of the Intuition MI300, referred to as the adaptive processing unit, or APU, containing 146 billion transistors. It combines components of a central processing unit and a graphics processing unit in a single package deal. In accordance with AMD, the chip is designed to run high-performance computing purposes and synthetic intelligence software program.

The MI300 is manufactured from chips, that are compact computing modules that may be combined and matched in several methods to create customized processors. MI300 has 13 such modules. They’re organized in two layers stacked on high of one another in a three-dimensional configuration.

In accordance with AMD, the primary layer of the MI300 consists of 9 CPU and GPU chips made utilizing a five-nanometer course of. The CPU modules are based mostly on the corporate’s newest Zen 4 processor structure. GPU modules, alternatively, characteristic AMD’s CDNA 3 structure, which powers information middle graphics playing cards.

The MI300’s CDNA 3 and Zen 4 circuits are stacked on high of one other semiconductor layer containing 4 extra chips. In accordance with AMD, the second chipset deal with supporting duties equivalent to information enter and output operations. They’re produced utilizing a six-nanometer course of.

MI300 shops the info it processes in a 128 gigabyte onboard reminiscence pool. The chip’s GPU and CPU modules share reminiscence, which AMD says is extra environment friendly than conventional information middle processors.

Conventional CPUs and GPUs are applied as separate chips and consequently retailer info in separate reminiscence circuits. Every chip retains its personal copy of the info being processed. AMD says the MI300’s shared reminiscence eliminates the necessity to create a number of copies of information for the CPU and GPU modules, which makes processing extra environment friendly.

The corporate claims that the MI300 can prepare AI fashions eight instances sooner than the present era MI250X graphics card. Moreover, the chip guarantees to ship a five-fold efficiency increase per watt for AI workloads.

“The MI300 can prepare a lot bigger AI fashions sooner, at decrease prices, and with much less energy,” mentioned Su. CES keynote. “The MI300 can cut back coaching time for these fashions from months to weeks and dramatically cut back vitality prices. And extra importantly, it might additionally assist a lot, a lot bigger fashions that can be utilized for much more superior and extra highly effective AI companies sooner or later.”

AMD unveiled the MI300 at CES, together with one other new information middle chip referred to as the Alveo V70. It’s optimized to carry out the duty of inferring or operating AI purposes in manufacturing after coaching. AMD says the Alveo V70 is predicated on an structure developed by the Xilinx subsidiary and might carry out as much as 400 trillion calculations per second.

New desktop and laptop computer chips

Alongside the info middle processors it previewed at CES, AMD introduced just a few new additions to its portfolio of shopper chips. Among the chips are designed to be used in desktop computer systems, whereas others goal the laptop computer market.

AMD is launching a brand new line of desktop CPUs often known as the Ryzen 7000X3D sequence. The sequence consists of three chips providing 8 to 16 cores with a most clock pace of 5.7 gigahertz. AMD says the chips can carry out file compression 52% sooner than Intel Corp.’s high-end Core i9-13900K processor and run video video games as much as 25% sooner.

The Ryzen 7000X3D sequence is considered one of AMD’s first CPU sequence to characteristic V-Cache expertise. First launched in Might 2021, the expertise makes it doable to put a cache module on high of a CPU in a three-dimensional configuration.

A CPU’s cache shops information close to its logic circuits to enhance efficiency. When information is saved close to logic circuits, it may be retrieved sooner, which quickens calculations. AMD’s V-Cache expertise additional reduces the gap info has to journey earlier than it may be processed.

“The Ryzen 7000X3D is our first 16-core Ryzen processor with V-Cache expertise and our quickest 3D stacked chip ever,” mentioned Su.

AMD is launching its new Ryzen 7040 processor lineup for laptops. It’s based mostly on the corporate’s newest Zen 4 CPU structure and can be manufactured utilizing a 4 nanometer course of. Moreover, AMD has added an AI module that guarantees to run machine studying purposes 20% sooner than the M2 chip in Apple Inc.’s newest MacBooks.

“This chip makes use of four-nanometer course of expertise and has greater than 25 billion transistors, which is sort of double our Ryzen 6000 era,” Su detailed.

AMD is making the newest additions to its shopper CPU portfolio alongside a brand new line of graphics playing cards for the laptop computer market. The Radeon RX 7000 sequence options 4 GPUs pushed by the RX 7600M XT. In accordance with AMD, the RX 7600M XT can run some video video games as much as 31% sooner than Nvidia Corp.’s RTX 3060 desktop graphics card.

The chip has 2,048 shader processors, which have the same function as a CPU core. Shader processors are supported by 32 megabytes of cache and eight gigabytes of GDDR6 reminiscence. In accordance with AMD, the RX 7600M XT is manufactured utilizing a six-nanometer course of.