Samsung Electronics' 24 gigabit GDDR7 DRAM / Courtesy of Samsung Electronics

Samsung Electronics developed the industry's first 12-nano 24 gigabit (Gb) GDDR7 DRAM, targeting applications in data centers and artificial intelligence (AI) workstations, the company said Thursday.

GDDR7, or Graphics Double Data Rate 7, is a graphic DRAM, or dynamic random-access memory, designed to boost the performance of devices like gaming consoles and graphic cards.

According to the chipmaker, it plans to commercialize the new chip early next year, beginning its validation for next-generation AI computing systems for major GPU (graphics processing unit) customers within this year.

Samsung said the new chip achieved major upgrades in capacity, performance and power efficiency from its predecessor 16Gb GDDR7, which the company developed last year.

The 24Gb GDDR7 was produced through 12-nanometer class process technology, which enabled increased cell density to achieve the industry's highest capacity of 24Gb while maintaining the same package size as the predecessor.

The new chip also uses three-level Pulse-Amplitude Modulation, or PAM3, signaling technology to enable 40 gigabits-per-second of speed, which is a 25-percent improvement over the previous version. The GDDR7's performance can be further enhanced to up to 42.5Gbps, depending on the usage environment, the company said.

Power efficiency is also enhanced by 30 percent, by applying technologies that were previously used in mobile products, such as clock control management and dual VDD design.

Clock control management technology reduces power consumption by ensuring that each circuit only operates when necessary. Dual VVD design is a power management technique that lowers external voltage during low-speed operation.

"After developing the industry's first 16Gb GDDR7 last year, Samsung has reinforced its technological leadership in the graphics DRAM market with this latest achievement," said Bae Yong-cheol, executive vice president of Memory Product Planning at Samsung Electronics.

"We will continue to lead the graphics DRAM market by bringing next-generation products that align with the growing needs of the AI market."

Samsung Electronics' 12-layer HBM3e chip / Courtesy of Samsung Electronics

Along with high-bandwidth memory (HBM) chips, which are making headlines for their use in AI processors, GDDR chips are also gaining attention for their application in AI services.

HBMs are used for high-performing AI processors, as their larger bandwidth enables transferring more data at once, although their transfer speed is slower than comparable GDDR chips. Most AI servers for data centers use HBM-powered AI processors.

However, HBMs require higher manufacturing costs and come with less flexibility for application compared to GDDR Chips. It is widely known that HBM chips cost three times more than general DRAMs.

Due to these pros and cons, both HBM and GDDR chips are expected to play complementary roles until costs for HBMs decline. HBMs are widely used for high-performance chipsets used for AI learning, while GDDR chips are used for AI inference.