NVIDIA GeForce RTX 5080 SUPER leaks again: full GB203 GPU, 24GB GDDR7 at 32Gbps, 420W+ power

NVIDIA's beefed-up GeForce RTX 5080 SUPER graphics card is back in the headlines, with leaked specs suggesting we can expect 24GB of GDDR7 memory clocked at 32Gbps, 10752 CUDA cores, and 400W+ of power.

It was only earlier this month that leaked Moore's Law is Dead said that we were to expect the new RTX 5080 SUPER with 24GB and the new RTX 5070 SUPER with 18GB later this year, with between 7% and 15% performance gains over their non-SUPER counterparts.

But in a new post by inside @kopite7kimi on X, we've got some fresh specs that include the PG147-SKU35, GB203-450-A1 GPU, 10752 CUDA cores, 24GB of GDDR7 clocked at 32Gbps on a 256-bit memory bus, and a rather hefty 400W+ of power consumption.

Moore's Law is Dead pumped out a new video explaining that he was told by his AIB sources that in a "last briefing on the 5080 SUPER detailed: full GB203 with 10752 CUDA cores, 24GB of 32Gbps GDDR7, and a 420W+ TDP". The source continued: "on that last bullet point regarding TDP, literally '420w' was listed as the current plan, but we were also wanted they may increase it to '450W' if they felt it necessary. Again, they want a solid uplift over the 5080..."

In the leaks from MLID earlier this month, his sources said: "We've (AIB) had 24GB 5080 and 18GB 5070 configurations in labs since 2024. In fact, those leaked 5080 boxes from a few months ago with"24GB"on them were connected to those projects. Basically, NVIDIA's been telling us to"put those variants on hold"while they wait for high capacity GDDR7 to become cheap enough to be worth it for gamers".

MLID's source continued: "If this plan moves forward (it's NOT 100%), I'd estimate SUPER cards will be 7-15% faster than their non-SUPER brothers. Pricing is still undetermined, but it kinda sounds like they'll either keep the same prices or marginally increase them. RDNA 4 has them spooked".