Unveiling the next generation of graphics memory: GDDR7

Loading…

Imagine asking gamers to play with fewer frames per second or endure lag in their games. Consider telling data scientists to wait extra hours for their computations to yield results or suggesting to content creators who use cutting-edge AI tools to expect slower production times. It's certainly not ideal for people who rely on speed and efficiency in their digital tasks to wait longer or expect frequent interruptions as their programs load.  

Our modern world doesn’t wait, so the performance of graphics memory must keep pace with ever greater application demands — everything from gaming and visualization to generative AI and high-performance computing (HPC).

GDDR6 (graphics double data rate 6) memory satisfied the high performance requirements of these workloads for a few generations, with performance scaling as high as 20 Gb/s. Yet over time, as application demands continued to soar, 20 Gb/s was no longer fast enough. And with clock speeds on GDDR6 already running as fast as possible — trending to 100s of picoseconds (1 million times faster than a human blink) — the technical and economic feasibility of continued improvements with GDDR6 did not add up. We could not suggest to our gaming friends that they forgo an immersive, vivid gaming experience where each frame renders smoothly and every detail comes to life. What we needed was a breakthrough in graphics memory to elevate user experience to greater heights. Enter the next generation of graphics memory: GDDR7. 

GDDR7 is a revolution in graphics memory that enables content creators to work on high-resolution video, the latest generative AI tools and intricate 3D models — all in a seamless, uninterrupted workflow. Gamers can experience breathtaking visuals and flawless gameplay. And tech professionals, such as data scientists, can count on speedy turnaround times for complex data processing to uncover insights and make decisions quickly.

Experience a breakthrough in performance with GDDR7 

Built on Micron’s industry-leading 1β (1-beta) node, GDDR7 is the next generation of GDDR memory to advance user experience in graphics and gaming. With initial memory performance enabling data rates of 32 Gb/s and up to 1.5 TB/s of system bandwidth (60% higher than GDDR6), GDDR7 eliminates frustrating bottlenecks and provides the foundation and versatility to conquer increasingly complex problems for a wide range of applications — everything from HPC simulations in scientific research to virtual reality. This stellar performance is accompanied by significant improvements in power efficiency to keep systems running reliably fast yet cool.

We might wonder whether clock speeds have already reached their limit. If we cannot run the clock any faster for graphics memory, how does GDDR7 break through to deliver such outstanding performance? Higher frames per second, reduced processing time and increased throughput are all possible thanks to the innovative technology that powers GDDR7: multilevel signaling. With the foundation already created by Micron’s pioneering work in implementing PAM4 (pulse-amplitude modulation with four signals) on GDDR6X, the DRAM industry was able to further scale the next generation of GDDR by continuing to explore multilevel signaling techniques. GDDR7 is the first DRAM device to feature PAM3 signaling (with three signals), a technique that allows for higher data transmission rates, improving memory bandwidth and overall system performance.  

By using PAM3 on GDDR7, the industry addressed the limits of clock speed by transferring additional data for each cycle. As compared with non-return-to-zero (NRZ) technology on GDDR6, PAM3 encodes more information (1.5 times more) per signal cycle using three distinct voltage levels: –1, 0 and +1. Even though PAM3 transfers fewer bits per cycle (than PAM4 on GDDR6X), it allows for a 50% higher voltage margin and has lower encoder complexity. These effects reduce the need to have higher frequencies for the memory bus and subsequently mitigate the resulting signal loss. The outcome is a powerful technology that delivers a huge performance jump and the ability to further scale GDDR performance in the future. 

Enable a wider range of GDDR7 applications 

As gaming graphics become more lifelike and generative AI moves more workloads to edge devices, the need for high-performance memory for purpose-built devices will only grow. With advanced reliability, availability and serviceability (RAS) features that enhance device reliability and data integrity, GDDR7 use extends beyond gaming to a broader range of applications such as HPC and AI inference at the edge.
 

What’s more, Micron has characterized the industry’s first 40 Gb/s PAM3 data eye. Being able to successfully demonstrate a 40 Gb/s data eye capability with such a new technology (multilevel signaling implemented on DRAM) shows that PAM3 signaling for GDDR7 was the right choice, enabling the industry to continue scaling graphics performance to higher speeds. Against even the most demanding workloads, Micron will continue to lead the way in advancing graphics memory. 

Visit our GDDR7 page to learn more about this high-performance memory. For a quick look at GDDR7 specs and performance, check out our product brief. 

1. As compared to GDDR6/6X trend across 1080p, 1440, and 4K resolutions.
2. Inference projection for GDDR7 32Gb/s at calculated using measured GDDR6 20Gb/s data.