It is reported that Samsung Electronics Co., Ltd. has successfully developed the world's first 24Gb GDDR7 DRAM. Not only does it have the highest capacity in the industry, but it is also an ideal solution for next-generation applications due to its blazing-fast speed. In addition to traditional graphics memory applications such as graphics cards, game consoles, and autonomous driving, the high capacity and powerful performance of 24Gb GDDR7 will also be widely used in data centers, AI workstations, and other fields that require high-performance memory.
Here are some specific application scenarios for this memory in the AI field:
Data centers and AI workstations: As AI models and data volumes continue to increase, data centers and AI workstations need to process and store massive amounts of data. The high capacity and speed of 24Gb GDDR7 DRAM can meet the demand for high-performance memory in these scenarios and improve data processing efficiency.
Autonomous driving: In the field of autonomous driving, systems need to process large amounts of data from sensors in real time and make quick decisions. The high-speed data transmission capability of GDDR7 DRAM can support autonomous driving systems to maintain low latency and high responsiveness when processing sensor data and performing complex calculations.
High-performance computing: In fields such as scientific research and engineering simulation, HPC systems require massively parallel computing. 24Gb GDDR7 DRAM provides enough memory bandwidth and capacity to support the performance needs of these systems when performing intensive computing tasks.
AI training and inference: The training and inference process of an AI model requires a large amount of memory, bandwidth, and capacity to store and process data. The high speed and large capacity of GDDR7 DRAM make it ideal for AI accelerators and GPUs, which can improve the efficiency of AI training and inference.
Figure: Samsung successfully developed 24GB GDDR7 DRAM (Source: Samsung)
Graphics processing: In addition to AI computing, 24Gb GDDR7 DRAM can also be used in traditional graphics processing fields, such as graphics cards and game consoles, to provide higher graphics processing performance and more detailed image rendering.
In-Memory Processing (PIM): Samsung Electronics is also exploring the integration of computing logic into in-memory devices, known as in-memory processing (PIM), which can further reduce data movement and improve computing performance and energy efficiency for AI applications.
AI PCs: For AI PCs, 24Gb GDDR7 DRAM can provide enough memory capacity and speed to support AI inference and training tasks while maintaining low power consumption and high performance.
"Following the launch of the world's first 16Gb GDDR7 last year, Samsung has once again strengthened our technology leadership in graphics memory with this innovation," said YongCheol Bae, Executive Vice President, Memory Product Planning at Samsung Electronics. We will continue to drive the market forward with next-generation memory products that meet the rapidly growing needs of the AI market."
The 24Gb GDDR7 uses the fifth-generation 10nm process, which delivers a 50% increase in cell density compared to the previous generation, but the package size remains the same. In addition, it introduces a three-stage Pulse Amplitude Modulation (PAM3) signal that delivers industry-leading speeds of 40Gb per second, a 25% improvement over the previous generation. Under certain conditions, GDDR7 can be further increased to 42.5Gbps.
To improve power efficiency, Samsung is applying advanced technologies in mobile products to graphics DRAM for the first time, with technologies such as clock control management and dual VDD design significantly reducing power consumption and improving power efficiency by more than 30%. In addition, the 24Gb GDDR7 features a power-gated design that minimizes current leakage and ensures stability at high speeds.
Major GPU customers are expected to validate the 24Gb GDDR7 within this year, with plans to commercialize it in next-generation AI computing systems early next year.
