While Nvidia’s RTX 40-series (Ada Lovelace) generation of consumer gaming GPUs has been praised for their efficiency, the company was simultaneously heavily criticized for the lack of video memory (VRAM) in them. The low frame buffer meant that modern titles that are fairly memory-heavy had exhibited issues, like textures not properly streaming and loading, among others.

As such, the company was sort of forced to release a mid-cycle refresh with its RTX 4000 Super series, though we are not sure if this was always the strategy from Nvidia to offer the bare minimum of VRAM initially and then add more when it absolutely had to.

However, rumor has it that Nvidia is not looking to repeat the same mistake again. Last year, some alleged memory specs of the RTX 5000 series (codenamed “Blackwell”) were out. Apparently, the chip giant is looking to go for a 512-bit wide memory interface on the next generation, which makes it likely that the RTX 50-series flagship GeForce GPUs, the RTX 5090 and 5090 Ti, could be packing 32GB of VRAM.

Reputed leakster and Twitter (now X) user kopite7kimi believes that Nvidia is seemingly sticking to such a configuration:

In addition, they also claim that Nvidia could be using 28Gbps memory in the RTX 5000 series:

What this means is that Nvidia’s RTX 5090 or 5090 Ti may end up being an absolute monster in terms of VRAM specifications. The company is reportedly going from a 384-bit wide bus to a 512-bit while the memory is going up to 28Gbps from 21Gbps.

Hence, the memory bandwidth of the hypothetical next-gen flagship should be ~1,771 GB/s, which is over 75% higher than the 1,001 GB/s on the RTX 4090.

There is also a slim possibility that Nvidia’s RTX 5090/Ti could have 48GB of VRAM, though that seems rather unlikely, even though Micron, which is expected to supply the GDDR7 memory for the next-gen, has both 2GB (16Gb) and 3GB (24Gb) chips.

Micron roadmap showing GDDR7 HBM3 memory and more
via AnandTech

But what good is VRAM without the core itself? Previous unconfirmed reports have suggested that the RTX 50 series could see a bump of 50% in terms of CUDA core count. Thanks to the increased bandwidth and IPC (instructions per cycle) improvements, we could potentially see a performance increase in the range of around 60-70%.