記憶體產品

Ultra-Bandwidth Memory Unleashes Graphics-Intensive Applications

作者 Ralf Ebert - 2022-01-19

As PC gaming evolves in sophistication and popularity, gamers demand the highest performance possible in their advanced graphics cards to allow for smooth animation, sharp visuals, high frame rates and real-time ray tracing, a rendering technique critical for realistic light reflections and cinematic effects. Over the past decade, the tech industry has come to realize that the graphics processing units (GPUs) that power these gaming animations are also a good fit for data center applications like artificial intelligence (AI) and for modeling and visualization in high-performance computing (HPC). This is because the mathematical and graphical calculations offloaded to the GPU in those workloads are similar to the GPU functions in advanced gaming.

And what do these gaming, AI and HPC applications all have in common? They rely heavily on memory performance.

That’s why Micron saw a need to design, develop and deliver our GDDR6X ultra-bandwidth solution – the world’s fastest discrete graphics memory solution and the first to power system bandwidth up to 1 terabyte per second (TB/s).

GDDR6X’s Breakthrough Performance

Since 2006, Micron’s engineers and researchers have explored applying multilevel signaling to the memory interface and dramatically increasing system bandwidth. Over 45 patents later, Micron is the first in the industry to implement four-level pulse amplitude modulation (PAM4) in memory, creating a new benchmark for future generations of graphics memory. The Micron memory is the first production memory input/ourput architecture with PAM4.

By using PAM4 multilevel signaling techniques, GDDR6X transfers data much faster, doubling the I/O data rate from two bits to four. Until now, graphics memory was capped at 72 gigabytes per second (GB/s) for each component. GDDR6X pushes beyond the traditional binary standard to employ four distinct levels and transmit two bits of data to and from the memory at a time. As a result, Micron’s GDDR6X increases memory bandwidth to 84 GB/s for each component, translating to a system bandwidth of up to 1 TB/s.

And, unlike traditional memory, GDDR6X’s unparalleled data rates are well-suited for the demand of data-hungry applications in gaming and AI, setting a new benchmark for graphics memory.

Unlocking Efficiency Through GDDR6X

Micron’s GDDR6X is not just the only memory device using PAM4, bringing cutting-edge I/O research to market, it’s also the only one in mass production. With its board design and packaging carefully fine-tuned for simplicity by Micron graphics memory experts, GDDR6X offers customers ease of implementation and less complexity than other high-bandwidth solutions on the market. Notably, GDDR6X doubles the data rate in a channel compared to DDR memory without doubling the required operating frequency, creating a smooth graphics experience for users. Additionally, GDDR6X delivers lower power per transaction than previous generations, making it ideal for energy-sapping workloads like gaming and other high-bandwidth graphics applications that demand both lightning fast and low-power memory. GDDR6X also features the ability to scale power down or up, allowing users to dial back performance to save energy.

Collaborating and Innovating With the Industry

Building solutions like GDDR6X requires deep industry collaboration. Micron announced availability of its GDDR6X memory featuring 8Gb density as part of the company’s Ultra-Bandwidth Solutions portfolio in September 2020. NVIDIA debuted the technology in the company’s GeForce™ RTX 3090 and GeForce RTX 3080 GPUs, which are tailored to support the fast speeds that immersive, high-performance gaming applications demand. In June 2021, it was included in the NVIDIA GeForce RTX 3070 Ti and GeForce RTX 3080 Ti.

Combined with NVIDIA’s GeForce RTX GPUs, GDDR6X delivers the groundbreaking speed necessary to allow gamers to become fully immersed in virtual worlds through photorealistic and reactive effects without lowering frame rates.

Changing the AI Landscape Through Memory

Though commonly paired with graphics, GDDR6X’s uses are expanding beyond graphics cards into other memory-intensive applications like AI. Recently, Micron collaborated with Innosilicon to deliver the world’s first silicon-proven commercial GDDR6X IP. Innosilicon’s GDDR6X-based PHY chip uses PAM4 signaling to boost the high efficiency and data rates that AI applications demand. The AI collaboration demonstrates GDDR6X’s incredible utility beyond simply being a graphics-focused solution, opening a host of possibilities for using GDDR6X’s game-changing bandwidth and signaling interface.

From pushing next-generation graphics acceleration to supporting breakthrough AI solutions, Micron is redefining how memory can push technological boundaries. Our leadership and engagement with standards committees drive a collaborative spirit to enable technology advances like PAM4 to be adopted by broader applications and audiences where the balance of bandwidth, power and design ease fulfills challenging end system requirements.

2021 AspenCore World Electronics Achievement Award (WEAA)

As memory becomes increasingly important to delivering end user experiences, I am truly excited for Micron to receive a 2021 AspenCore World Electronics Achievement Award (WEAA) honoring our GDDR6X ultra-bandwidth solution.

The WEAA recognizes some of the world’s most innovative new electronics products. Winners were determined by AspenCore’s global senior industry analysts and online voters in Asia, Europe and the United States.

With GDDR6X pushing the boundaries of what is possible in graphics acceleration and memory performance, I am honored to see Micron recognized for its creative approach. By receiving this award, we reaffirm our commitment to be a leader in exploring the frontier of memory innovation.

獎項

Micron’s Gan Liu (right), senior director of Global Key Account Sales, attended the 2021 AspenCore WEAA ceremony to accept the company’s award.

Ralf Ebert

Ralf Ebert

Ralf Ebert is the director for GDDR Product Architecture. He is responsible for developing the Micron’s GDDR architecture strategy, defining high-speed memory solutions that serve the game console and high-end graphics market as well as emerging segments like crypto and AI. Previously, Ralf held various positions including director of Micron’s global Graphics Memory Business, senior manager for SDD Business Development, senior manager of marketing, and roles in supply chain and business process excellence.

Ralf has worked in the semiconductor industry since 1997 and spent the last 17 years focusing on memory products in roles at Infineon, Samsung and Qimonda. In 2009, he set up and developed the graphics business for Elpida, which he continued after Micron’s acquisition in 2013.

+
Powered by Translations.com GlobalLink OneLink SoftwarePowered By OneLink