Leaked Nvidia GeForce RTX 5090 specs reveal massive VRAM upgrade; GeForce RTX 5080 still restricted to 16 GB

Prolific leaker Kopite7kimi has finally revealed the specs of Nvidia's upcoming GeForce RTX 5090 and RTX 5080 graphics cards. The former will get a long-overdue bump in VRAM, but the latter will not. And yes, the RTX 5090 will be even more of a power hog than its predecessors. Unfortunately, there's still no word about when they'll launch, but CES 2025 seems like a likely event to show off a flagship product.

Nvidia GeForce RTX 5090

The Blackwell flagship GeForce RTX 5090 (PG144/145-SKU30, GB202-300-A1 GPU) will come with 21,760 CUDA cores. Its SM count cannot be determined now as we don't know how many CUDA cores will be present in one SM. Other notable specs include 32 GB of 28 Gbps GDDR7 VRAM on a 512-bit wide bus for a total memory bandwidth of ~1.5 TB/s. It has a TGP of 600 Watts and will likely require two 12VHPWR cables because pushing that much wattage through a single one could lead to a lot of molten connectors.

Nvidia GeForce RTX 5080

Kopite7Kimi said earlier they believe the GeForce RTX 5080's (PG144/147-SKU45) GB203-400-A1 GPU to have half as many blocks as what the flagship GB202 has, and that seems to be the case indeed. It features 10,752 CUDA cores and 16 GB of 28 Gbps VRAM on a 256-bit bus for a total memory bandwidth of ~750 GB/s. It has a TGP of 400 Watts, much higher than the RTX 4080's 320 Watt figure.

Based on these specs, the GeForce RTX 5090 is shaping up to be an absolute beast. Unfortunately, one can't say the same about the RTX 5080, which seems more like a mid-cycle refresh than anything else. This attitude will likely trickle down to lower-end SKUs, so expect to see more xx70 and xx60 class cards with 8 GB VRAM in the future.