Future GPUs might have up to 48GB VRAM

News Room

Samsung has just announced the development of the first 24Gb GDDR7 memory that will one day be found in some of the best graphics cards. An industry first, the new DRAM brings huge improvements to both capacity and performance. With these new memory modules, we might finally stop seeing GPUs with just 8GB VRAM in the near future, with the maximum capacity reaching a massive 48GB.

Samsung’s newest DRAM will allow GPU makers to keep the same number of memory modules while reaching higher capacities. Right now, GPUs use either 8Gb (1GB) or 16Gb (2GB) memory modules. At 24Gb (3GB), the capacity will go up considerably, reaching up to 48GB. That’s unlikely to happen in the upcoming RTX 5090, but we might see that kind of number in a future RTX 6090.

The company revealed that the new 24Gb GDDR7 uses its 5th-gen 10nm DRAM, which results in a 50% increase in cell density without increasing the size of the actual memory. Aside from the improved process node, Samsung is also using three-level pulse-amplitude modulation (PAM3) signaling, which boosts memory speed by a whole lot.

Samsung's next-gen GDDR7 memory.

The first designs will offer speeds of up to 40Gbps, although Samsung notes that 42.5Gbps is not out of the question with further enhancements. That’s almost twice the memory speed compared to the RTX 40-series, which tops out at 22.4Gbps. This means that aside from a massive increase in capacity, we’ll also see huge improvements in bandwidth.

As calculated by VideoCardz, a GPU with a 256-bit memory bus and eight 24Gb modules might reach a whopping 1.2TB/s in bandwidth — that’s more than the RTX 4090 can currently offer. Meanwhile, a high-end version with a 512-bit memory bus and 16 GDDR7 modules could go up to 2.5TB/s. Such speeds are unheard of in this generation, but this bodes well for future gens.

Samsung said that the new 24Gb memory is currently entering the validation stage with some of its major GPU consumers before entering commercialization in early 2025. Does this mean it’ll be ready in time for Nvidia’s RTX 50-series? Unlikely, but there’s hope for the future. We’ve already seen leaks that these new 3GB modules might find a home in Nvidia’s next-gen GPUs, if not in the desktop versions, then at least in their laptop counterparts. However, even if we don’t see them in the initial lineup, there’s always hope for a Super refresh down the line.






Read the full article here

Share This Article
Leave a comment

Leave a Reply

Your email address will not be published. Required fields are marked *