JEDEC Finalizes GDDR7 Spec With A Massive Bandwidth Boost For Next-Gen GPUs
It's clear by now that consumer graphics card will continue to be fed by Graphics Double Data Rate (GDDR) memory standards rather than making the leap to High Bandwidth Memory (HBM), the latter of which is more costly and more suitable for data center accelerators and high performance computing (HPC) solutions. And as such, the JEDEC Solid State Technology Association (just JEDEC from here on out) has published finalized specifications GDDR7, which will eventually supplant GDDR6 and GDDR6X.
While the release of graphics cards equipped with GDDR7 is not yet imminent, it will likely find its way to flagship NVIDIA Blackwell and AMD RDNA 4 graphics cards, like the presumed GeForce RTX 5090 and Radeon RX 8900 XTX, respectively. We probably won't see those parts until 2025, but when they arrive, we can expect big bandwidth gains over the current generation models.
JEDEC says that GDDR7 will eventually reach up to 192GB/s per device, or 48Gbps per pin. We'll be looking at 32Gbps in the early going, but Samsung and Micron are both already working on faster solutions. Either way, it's an impressive generational lift.
To put it all into perspective, the fastest GDDR6X memory employed right now can be found on the GeForce RTX 4080 Super, which uses 23Gbps memory chips linked to a 256-bit memory bus. NVIDIA's flagship GeForce RTX 4090, meanwhile, employs 21Gbps GDDR6X memory chips linked to a fatter 384-bit memory bus.
Currently, Micron's GDDR6X memory chips can go as high as 24Gbps. That works out to 96GB/s per device. So in other words, GDDR7 effectively doubles the bandwidth, and will scale even higher as companies like Samsung and Micron push the envelope.
What's also interesting about the next-generation VRAM standard is that it is the first DRAM to use a Pulse Amplitude Modulation (PAM) interface for high frequency operations. What this means is that it uses three levels of signaling (-1, 0, +1) to transmit 3 bits over 2 cycles, versus 2 bits over 2 cycles using the traditional non-return-to-zero (NRZ) interface. This is where the bulk of the bandwidth gains come from.
"JESD239 GDDR7 marks a substantial advancement in high-speed memory design," said Mian Quddus, JEDEC Board of Directors Chairman. "With the shift to PAM3 signaling, the memory industry has a new path to extend the performance of GDDR devices and drive the ongoing evolution of graphics and various high-performance applications."
This is important as the tech industry makes a concerted push for more AI-powered experiences that go well beyond AI chatbots and the like. We've seen a glimpse of this with text-to-image generation models and other fancy tools, and you can bet that the major players will keep driving AI into the future. In fact, AMD recently teased that it was adopting AI for its FSR upscaling technology. That's really just the tip of the iceberg, though.
"The groundbreaking GDDR7 memory standard unveiled today represents a pivotal step towards unlocking the potential of next-generation consumer, gaming, commercial, and enterprise devices," said Joe Macri, Compute and Graphics CTO and Corporate Fellow at AMD. "By harnessing the transformative power of GDDR7s, we can collectively unlock a new era of transformational compute and graphics possibilities, paving the way for a future shaped by innovation and discovery."
NVIDIA is equally enthused about the transition to GDDR7 for its next-gen products, and has been pushing for PAM signaling.
"NVIDIA is excited that our work with JEDEC has helped to make PAM signaling the foundational technology for GDDR7, helping customers to get the most performance out of their GPU," said Kaustubh Sanghani, VP GPU Product Management at NVIDIA.
For anyone interested in taking a deeper dive into JEDEC's GDDR7 specifications, the organization has made available a whitepaper. It's free to download, you just need to register first.
