nvidia built a 24GB RTX 5080 Super. then it vanished over a $60 memory chip

Nvidia designed a 24GB version of the RTX 5080. Board partners reportedly had physical units sitting ready to ship. Then, around December 2025, the entire RTX 50 Super refresh — the 5070 Super, the 5070 Ti Super, and that 24GB 5080 Super — quietly dropped off every public roadmap. Nearly eight months later, per How-To Geek and corroborating reports from TweakTown and AllKeyShop, there's still no launch date. There might never be one.

What Nvidia actually built, then buried

The appeal here wasn't subtle. A 24GB 5080 Super would have hit that capacity by swapping in 3GB memory modules on the same PCB and the same memory bus width the regular 5080 already uses — no redesign, no new die. Fifty percent more VRAM, same card. For gamers that's headroom for the next few years of texture bloat. For anyone running local AI models on a gaming GPU instead of a $2,000+ workstation card, it was the first genuinely affordable path to 24GB from Nvidia since the RTX 3090.

the math that killed it

Here's the number that actually mattered in Nvidia's boardroom: 3GB GDDR7 modules reportedly run $60-70 each, against roughly $20 for the 2GB parts used in every RTX 50-series card sold today. An RTX 5080 Super needs eight of those modules to hit 24GB. Do the arithmetic and you get $480-560 in memory alone, versus about $160 for the 2GB equivalent — before the die, the board, the cooler, or a single cent of margin. That's not a rounding error. It's most of a mid-range GPU's entire bill of materials, in memory chips, on a card Nvidia would still have to price competitively against AMD.

if you were waiting for the 24GB card

• Don't hold your GPU budget hostage to a launch nobody at Nvidia has confirmed and multiple reports say is dead for at least another year.
• If 24GB is specifically what you need for local AI, a used RTX 3090 still beats a new RTX 5060 Ti 16GB on VRAM per dollar today.
• Two RTX 5060 Ti 16GB cards in one system get you 32GB of pooled VRAM for less than a single 4090, if your motherboard has the PCIe lanes.
• For the DeepSeek R1 32B distill specifically, 24GB is still the number to chase — just not from a card that doesn't exist yet.

It's worth saying plainly: this is bad news specifically for anyone hoping Nvidia would hand consumers an affordable 24GB card. The company hasn't shipped one at a reasonable price since the RTX 3090, and the closest recent alternative — Apple cutting Mac Studio RAM while raising its price — tells you the entire high-VRAM market is getting squeezed from every direction at once, not just Nvidia's corner of it.

Nvidia will sell you an RTX 5090 at MSRP if you show up to the right parking lot — it did exactly that at QuakeCon a few weeks ago. What it apparently won't do right now is ship a mid-range card with real VRAM headroom, because the memory inside it is worth more sold to a data center. That's the RTX 50 generation in one sentence, and there's no sign it changes before the next one.