GeForce RTX 5090 Leak: 600W TGP, 32GB VRAM And More Monster Specs

GeForce RTX graphics card.
We never did end up seeing a Ti or Super variant of NVIDIA's flagship GeForce RTX 4090 like there was with the previous generation's top SKU (GeForce RTX 3090 Ti). That ship has sailed at this point and instead, all eyes are looking ahead to NVIDIA's next-generation Blackwell lineup. While we wait for the eventual launch, an historically reliable leaker posted several supposed specs for NVIDIA's upcoming GeForce RTX 5090 and GeForce RTX 5080.

Now, it's still a little early, presuming that we're in store for an unveiling on or around the Consumer Electronics Show (CES) in January 2025. And as we've seen with NVIDIA numerous times in the past, specifications can and often do change at the last minute. And of course this falls into the leak category, so take it for what it's worth.

Disclaimer out of the way, what are we looking at? According to X/Twitter user kopite7kimi, NVIDIA's flagship GeForce RTX 5090 will be a monster graphics card built around a GB202-300-A1 Blackwell GPU with a 600W total graphics power rating. That's a 33% uptick from the GeForce RTX 4090's 450W TGP.

Kopite7kimi X/Twitter post outlining specs for the GeForce RTX 5090 and 5080.

Kopite7kimi is also under the impression that it will feature 21,760 CUDA cores, which is 33% more than the 16,384 CUDA cores found on the GeForce RTX 4090. That's a smaller increase than what we got with the current generation flagship—the GeForce RTX 4090 flexes 56% more CUDA cores than the GeForce RTX 3090, and 52% more compared to the GeForce RTX 3090 Ti variant.

Still, 21,760 is a lot of CUDA cores, and of course there's the added benefited for an upgraded GPU architecture (Blackwell versus Ada Lovelace). Additionally, kopite7kimi claims the GeForce RTX 5090 will wield an impressive 32GB of GDDR7 VRAM linked to a 512-bit memory bus. If that memory ends up being clocked at 28Gbps, we'll be looking at around 1.7TB/s of memory bandwidth.

That would be a massive 68% increase in memory bandwidth versus 1.01TB/s on the GeForce RTX 4090, which features 24GB of GDDR6X on a 384-bit bus. So overall, the GeForce RTX 5090 looks like a substantial upgrade over the GeForce RTX 4090, assuming the leaked specifications are right on the money. The trade off, of course, is the power draw.

Meanwhile, it's said the GeForce RTX 5080 will feature a GB203-400-A1 GPU with 10,752 CUDA cores, 16GB of GDDR7 linked to a 256-bit memory bus, and a 400W TGP. To put those specs into perspective, the GeForce RTX 4080 is equipped with 9,728 CUDA cores and 16GB of GDDR6X memory on a 256-bit bus, with a 320W TGP. As for the GeForce RTX 4080 Super, it carries the same 320W TGP, but boasts 10,240 CUDA cores and the same memory config (albeit slightly faster).

There's no mention of clock speeds or features related to either SKU. However, according to the folks at Benchlife, both card will utilize a single 12V-2x6 PCIe power connector rather than dual connectors as some rumors have indicated. It's also said that the cards will support PCIe 5.0 and DisplayPort 2.1a.

It's anyone's guess what pricing ends up looking like. As points of reference, the GeForce RTX 4090 and GeForce RTX 4080 launched at $1,599 and $1,199, respectively, while the GeForce RTX 4080 Super debuted at $999.