Prolific leaker Kopite7kimi has disclosed the specifications for Nvidia's upcoming GeForce RTX 5090 and RTX 5080 graphics cards. The RTX 5090 will finally receive a significant increase in VRAM, while the RTX 5080 will not see this upgrade. Additionally, the RTX 5090 is expected to consume even more power than its predecessors. The exact launch date remains uncertain, but CES 2025 is anticipated to showcase this flagship product.
GeForce RTX 5090 Specifications
The Blackwell flagship GeForce RTX 5090 (PG144/145-SKU30, GB202-300-A1 GPU) will feature 21,760 CUDA cores. The total number of Streaming Multiprocessors (SM) is still unknown, as it depends on how many CUDA cores each SM will include. Other significant specifications consist of 32 GB of 28 Gbps GDDR7 VRAM on a 512-bit bus, offering an impressive total memory bandwidth of approximately 1.5 TB/s. The graphics card is expected to have a Total Graphics Power (TGP) of 600 Watts, likely necessitating two 12VHPWR cables to avoid overheating issues associated with high wattage through a single connector.
GeForce RTX 5080 Specifications
Kopite7kimi previously mentioned that the GeForce RTX 5080 (PG144/147-SKU45, GB203-400-A1 GPU) would be half of the AD202 architecture, and this holds true. It will include 10,752 CUDA cores and 16 GB of 28 Gbps VRAM on a 256-bit bus, resulting in a total memory bandwidth of around 750 GB/s. The TGP for this card is set at 400 Watts, which is significantly higher than the RTX 4080's 320 Watts.
Performance Expectations
Given these specifications, the GeForce RTX 5090 is expected to be a powerhouse. In contrast, the RTX 5080 seems more like a mid-cycle refresh rather than an exciting new entry. This trend may extend to lower-tier models, suggesting that we can expect more xx70 and xx60 class cards with 8 GB of VRAM in the near future.
