Samsung Unveils 24 Gb GDDR7 DRAM for Next-Gen AI

On October 17, Samsung Electronics announced the development of the industry's first 24 Gb GDDR7 DRAM, setting a new benchmark in both capacity and speed. This cutting-edge GDDR7 chip is poised to be the optimal solution for next-generation applications, finding extensive use in high-performance memory solutions across diverse sectors, including data centers and AI workstations. It goes beyond traditional graphics DRAM applications found in graphics cards, gaming consoles, and autonomous driving technologies.

YongCheol Bae, Executive Vice President of DRAM Product Planning at Samsung Electronics, stated, "Following our introduction of the world's first 16 Gb GDDR7 last year, this latest achievement reinforces Samsung's technological leadership in the graphics DRAM market. We remain committed to driving the advancement of graphics DRAM and delivering next-generation products that meet the growing demands of the AI market."

The 24 Gb GDDR7 utilizes fifth-generation 10-nanometer (nm) DRAM technology, which enhances unit density by 50% while maintaining the same packaging size as its predecessors.

In addition to advanced manufacturing processes, Samsung has implemented three-level Pulse Amplitude Modulation (PAM3) signaling to achieve an industry-leading speed of 40 Gb/s (gigabits per second) for graphics DRAM—an increase of 25% over previous versions. The performance of GDDR7 can be further boosted to up to 42.5 Gb/s, depending on the operating conditions.

Moreover, by applying technologies previously used in mobile products to graphics DRAM for the first time, energy efficiency has significantly improved. Techniques such as clock control management and dual VDD design have led to a reduction in unnecessary power consumption, increasing power efficiency by over 30%.

To enhance operational stability during high-speed performance, the 24 Gb GDDR7 employs power-gating design techniques to minimize current leakage.

Validation of the 24 Gb GDDR7 for next-generation AI computing systems by major GPU clients will commence this year, with commercialization expected in early 2025.