GDDR7 vs GDDR6 – What’s the difference?

GDDR7 is the newest type of video card memory for high-end graphics cards from NVIDIA, and its arrival marks a new era for GPU memory. Just like how desktop memory has gone from DDR4 to DDR5 in the past few years, GPUs are now set to move from GDDR6 to GDDR7 with the arrival of the NVIDIA GeForce RTX 50-series of GPUs. Though the newest GPUs from Intel still use older GDDR6 memory, and AMD’s RDNA 4 cards will likely follow suit, both companies will likely eventually move to GDDR7 memory as well, and it will become the de facto memory standard for GPUs in the coming years.

Since GDDR6 is the current standard, GDDR7 is the next generation of GPU memory as its name implies. It will offer significant improvements in bandwidth and efficiency over GDDR6, which was first introduced in 2018 with the GeForce 20-series GPUs. In terms of transfer rates, GDDR7 has the potential to offer almost double the speed of the GDDR6 memory it’s replacing.

For example, GDDR6 currently tops out at around 20Gb/s in terms of transfer rates with GDDR6X taking that up to 23Gb/s, which is the pinnacle of GDDR6 memory speed. However, GDDR7 will arrive with speeds as high as 28Gb/s on the RTX 5090 and 30Gb/s on the RTX 5080 and has the potential to go as high as 40Gb/s and 48Gb/s in the next few years as well, making it a huge advancement in terms of throughput.

For a real-world example of the benefits of going from GDDR6 to GDDR7, we can examine the RTX 5080. This GPU has the same 16GB of memory on a 256-bit bus as its predecessor, the RTX 4080, so the only change is it now has faster memory. It’s going from 22Gb/s GDDR6 to 30Gb/s GDDR7, and the result is its memory bandwidth has been increased, going from 716.8GB/s to 960GB/s. What impact that will have on frames-per-second is hard to judge since we are comparing two different architectures, but the memory bandwidth gains are crystal clear.

GDDR7 will also offer big gains in efficiency over GDDR6 as well. For example, Samsung claims its GDDR7 memory will offer a 30% improvement in power efficiency over GDDR6, while Micron and SK Hynix both claim their GDDR7 will offer up to 50% better efficiency, which is quite a jump. This boost to efficiency will arguably be more important for mobile 50-series GPUs with GDDR7, but desktop users will also benefit from it as well.

GDDR7 is likely to be much more expensive than GDDR6 due to the advanced manufacturing it requires along with R&D costs incurred to develop it over the past few years, which is why it is only being used for high-end graphics cards at first. Since Intel is not targeting the high-end GPU market with its latest GPUs, it has chosen to keep using GDDR6, likely in order to keep costs down. AMD has teased its next-generation RDNA 4 GPUs but has not revealed specs yet, but it’s also expected to continue to use GDDR6 as well, which we assume is to keep its GPU prices somewhat reasonable.

However, as the cost of manufacturing GDDR7 is reduced in the years to come, it’s expected to become the new standard for all GPUs. This is good news for all gamers, but will especially benefit midrange and entry-level GPUs since those cards are typically hamstrung by narrow 128-bit and 192-bit memory buses. Having memory with faster data transfer rates could allow for a decent amount of memory bandwidth even across a narrow bus. These cards could still be hamstrung by having just 8GB of memory, so it remains to be seen if GDDR7 will help boost performance given that limitation.