Samsung develops 40 Gbps GDDR7 memory with 3 GB capacity

Samsung unveiled a 40 Gbps GDDR7 DRAM chip with 3 GB (24 Gb) capacity built on its 12 nm (10 nm class) node at the 2025 Korea Tech Festival in Seoul, following earlier sampling of 36 Gbps parts.

At the 2025 Korea Tech Festival in Seoul, Samsung showcased a next-generation GDDR7 memory chip that earned a presidential recognition medal. The device is built on Samsung’s 12 nm (10 nm class) DRAM node, operates at 40 Gbps and offers 3 GB (24 Gb) of capacity per chip. Samsung positioned the component for the next generation of graphics cards, highlighting both the raw data rate and the unusual per-chip capacity.

The announcement follows Samsung’s recent sampling of GDDR7 that runs at 36 Gbps and the company’s ongoing mass production of 28 Gbps 3 GB modules. The South Korean company confirmed those 28 Gbps 3 GB parts are in mass production and suggested they are likely intended for NVIDIA’s mid-cycle SUPER refresh of GeForce RTX 50-series GPUs. The report underscores a practical distinction between sampling and mass production: sampling alone is generally insufficient for a product rollout, and manufacturers prefer chips that are already in mass production for immediate use.

Industry observers noted that 3 GB modules are uncommon, so confirmation that Samsung is mass producing 3 GB parts, starting at 28 Gbps, expands higher-speed VRAM options. At the same time, Samsung’s testing of 40 Gbps GDDR7 modules with 3 GB capacity raises questions about where those higher-speed chips will be deployed. Ensuring a proper supply will be a key factor for any vendor planning to adopt these parts, since ample inventory of GDDR7 chips must accumulate before large-scale product launches.

52

Impact Score

What businesses need to know about the EU cyber resilience act

The EU cyber resilience act is turning product cybersecurity into a legal requirement for companies that sell digital products into the European Union. A key compliance milestone arrives in September 2026, well before the full regulation takes effect in 2027.

Claude Mythos and cyber insurance’s next inflection point

Claude Mythos is being treated by governments and regulators as a potential systemic cyber risk with implications for financial stability and insurance markets. Its emergence is intensifying pressure on insurers to clarify whether Artificial Intelligence-enabled cyber losses are covered, excluded, or require new stand-alone products.

OpenAI expands ChatGPT ads with self-serve manager

OpenAI is widening its ChatGPT ads pilot with a beta self-serve Ads Manager, new bidding options and broader measurement tools. The push signals a deeper move into advertising as the company expands the program into several international markets.

OpenAI launches Artificial Intelligence deployment consulting unit

OpenAI has created a new consulting and deployment business aimed at helping enterprises build and roll out Artificial Intelligence systems. The move mirrors a similar push by Anthropic and signals a broader effort by model providers to capture more of the enterprise services market.

SK Group warns DRAM shortages could curb memory use

SK Group chairman Chey Tae-won warned that customers may reduce memory consumption through infrastructure and software optimization if DRAM suppliers fail to raise output. Demand from Artificial Intelligence data centers is keeping the market tight as memory makers weigh expansion against the long timelines for new fabs.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.