Nvidia GeForce RTX 5090 could feature a significantly higher memory bus than the RTX 4090

Nvidia's next generation of graphics cards, codenamed Blackwell (or Ada Lovelace next), aren't scheduled to hit shelves until at least 2025. Like Ada Lovelace, it is rumoured to use a custom node from TSMC (likely 3 nm). It is also rumoured to offer the "biggest" generation-to-generation performance leap. Prolific leaker Kopite7kimi has now revealed one of the many reasons why that could indeed be the case.

The flagship GeForce RTX 5090 will supposedly feature a 512-bit memory interface, the highest ever seen on a consumer graphics card. That, combined with new GDDR7 modules, could result in a twofold increase in memory bandwidth. It potentially opens up the possibility of Nvidia packing up to 32 GB of video memory on its flagship card and up to 64 GB on its AI-centric products. Blackwell is also said to feature a new Streaming Multiprocessor (SM) design, which could potentially up the number of CUDA cores.

A modest increase in clock speed is almost inevitable, given the major node shift and improved architecture. Whether or not it hits the "2.6x" performance increase figure remains to be seen. Then again, it won't matter much as the GeForce RTX 5090 will remain an enthusiast-only card, and the real concern is will the new tech trickle down to lower-end offerings.

Buy the Asus TUF Gaming GeForce RTX 4080 on Amazon