RTX 50 Series Specs Leaked: RTX 5090 to 20,000+ Cores

NVIDIA is reportedly gearing up to launch its next-generation Blackwell architecture with the RTX 50 series graphics cards, expected to deliver significant performance enhancements. Recently, renowned leaker Kopite7kimi unveiled specifications for the GeForce RTX 5090 and GeForce RTX 5080 models.

The GeForce RTX 5090 is built on the GB202-300-A1 architecture and features an estimated 21,760 CUDA cores, contrary to previous rumors of 24,576 cores. This model boasts 32GB of GDDR7 memory with a 512-bit memory interface, and it comes with a power consumption of up to 600W. For comparison, the current flagship GeForce RTX 4090 has a power draw of 450W.

On the other hand, the GeForce RTX 5080 utilizes the GB203-400-A1 architecture, with a power requirement of 400W—up from the 320W of the RTX 4080 series. It includes 10,752 CUDA cores, 16GB of GDDR7 memory, and a 256-bit memory interface. Overall, the performance of the RTX 5080 is expected to be nearly half that of the RTX 5090.

While the reasoning behind NVIDIA's product line strategy remains unclear, speculation suggests that the company may aim to position certain consumer-grade GPUs as alternatives for AI applications. With the significant memory bandwidth improvement from GDDR7, the RTX 5080 is likely to outperform its predecessor in high-resolution scenarios.

From a technical perspective, it is possible that the GB202 processor is composed of two GB203 processors, implying that the RTX 5090 could be akin to a dual-chip version of the RTX 5080.