Future GPUs might need as much as 48GB VRAM

Samsung has simply introduced the event of the primary 24Gb GDDR7 reminiscence that may at some point be present in a number of the greatest graphics playing cards. An trade first, the brand new DRAM brings enormous enhancements to each capability and efficiency. With these new reminiscence modules, we’d lastly cease seeing GPUs with simply 8GB VRAM within the close to future, with the utmost capability reaching a large 48GB.

Samsung’s latest DRAM will enable GPU makers to maintain the identical variety of reminiscence modules whereas reaching increased capacities. Proper now, GPUs use both 8Gb (1GB) or 16Gb (2GB) reminiscence modules. At 24Gb (3GB), the capability will go up significantly, reaching as much as 48GB. That’s unlikely to occur within the upcoming RTX 5090, however we’d see that sort of quantity in a future RTX 6090.

The corporate revealed that the brand new 24Gb GDDR7 makes use of its Fifth-gen 10nm DRAM, which leads to a 50% enhance in cell density with out growing the scale of the particular reminiscence. Apart from the improved course of node, Samsung can be utilizing three-level pulse-amplitude modulation (PAM3) signaling, which boosts reminiscence pace by a complete lot.

The primary designs will supply speeds of as much as 40Gbps, though Samsung notes that 42.5Gbps will not be out of the query with additional enhancements. That’s nearly twice the reminiscence pace in comparison with the RTX 40-series, which tops out at 22.4Gbps. Which means other than a large enhance in capability, we’ll additionally see enormous enhancements in bandwidth.

As calculated by VideoCardz, a GPU with a 256-bit reminiscence bus and eight 24Gb modules would possibly attain a whopping 1.2TB/s in bandwidth — that’s greater than the RTX 4090 can presently supply. In the meantime, a high-end model with a 512-bit reminiscence bus and 16 GDDR7 modules might go as much as 2.5TB/s. Such speeds are unparalleled on this era, however this bodes properly for future gens.

Samsung stated that the brand new 24Gb reminiscence is presently getting into the validation stage with a few of its main GPU customers earlier than getting into commercialization in early 2025. Does this imply it’ll be prepared in time for Nvidia’s RTX 50-series? Unlikely, however there’s hope for the long run. We’ve already seen leaks that these new 3GB modules would possibly discover a residence in Nvidia’s next-gen GPUs, if not within the desktop variations, then at the least of their laptop computer counterparts. Nonetheless, even when we don’t see them within the preliminary lineup, there’s all the time hope for a Tremendous refresh down the road.