Nothing here is announced by Nvidia. This piece separates what is official, what is rumor, and what is the sensible reading of both.
What Nvidia has actually said
Jeff Fisher, Nvidia's senior vice president of personal computing, told Taiwanese media on October 16 that the RTX 5090 has not been discontinued and that Nvidia will continue shipping it. Taiwan's Central News Agency reported his comments from an event in Taipei. He acknowledged that supply chain capacity is constrained, but said the company is working to bring products to market.
That is a narrow denial. It says the product isn't being killed. It doesn't say how many cards will ship, when they'll be in stock, or how Nvidia divides its biggest Blackwell chip between gaming and professional cards. "Not discontinued" and "easy to buy" are very different claims.
What the rumors say
The chain of rumors started with leakers. One leaker, hongxing2020, shared an alleged factory notice saying Nvidia would no longer accept orders for RTX 5090 models, including the China-market D variants. The same notice claimed the RTX 5080 16GB would eventually be discontinued and replaced by a 24GB version. Nobody has authenticated that notice. Nvidia hasn't confirmed it, and the notice itself has not been verified as genuine.
Taiwanese outlet BenchLife then added detail, as relayed by VideoCardz, which tagged its post as a rumor:
- Nvidia is preparing an RTX 5080 with 24GB of GDDR7 for release in Q1 2027, replacing the existing 16GB model rather than selling alongside it.
- Board partners are still awaiting technical details. They have reportedly only been told verbally that development materials are coming. The Q1 2027 window is therefore soft.
- BenchLife also supports earlier claims that Nvidia is redirecting GB202 supply, the chip behind the 5090, toward its RTX PRO series. The outlet suggests this may be an allocation change rather than a formal discontinuation.
That last point fits Fisher's wording. A chip can quietly stop flowing to GeForce partners without anyone issuing an end-of-life notice. TechPowerUp's summary of the BenchLife report says the chips are being shifted to more profitable GB202-based RTX PRO cards such as the RTX PRO 6000 and the recently unveiled RTX PRO 5500.
Where the "half the CUDA cores" math comes from
Nvidia lists the RTX 5090 at 21,760 CUDA cores and 32GB of GDDR7. The RTX 5080 has 10,752 CUDA cores and 16GB. That puts the 5080 at roughly 49% of the 5090's core count, close to half but not exactly.
The headline's framing assumes a 24GB 5080 keeps those 10,752 cores. Earlier leaks suggest it would. The memory would sit on the same 256-bit bus but run at 32Gbps instead of 30Gbps, and the card would use 3GB GDDR7 modules. TechSpot's July report put the core count at an unchanged 10,752. It also said a 415W TDP had appeared in Seasonic's PSU calculator, against 360W for the standard card.
Two caveats apply:
- Those specs belong to the older "RTX 5080 Super" rumors. BenchLife's report doesn't say whether the 24GB card is that same product or a revised model under a different name. The "Super" label is a guess. Treat the 415W figure the same way.
- Nothing is being removed from the 5090. The "lose half its cores" idea only holds if the 5090 vanishes from sale and the 5080 becomes the top card by default.
Would 24GB matter?
Eight extra gigabytes is a 50% bump in memory, but it isn't 50% more rendering speed. The extra VRAM helps in specific cases: 4K with path tracing, heavy texture packs, large AI models, and some creative workloads. If a game isn't hitting the 16GB ceiling, it won't get much from the extra memory. The core count, and presumably the architecture, would be unchanged.
There are no benchmarks, no price, no confirmed US availability, and no Nvidia announcement. Anyone calling this card a 5090 replacement in performance terms is ahead of the evidence.
Why this keeps happening
The memory market and the professional GPU market both pull in the same direction. The cut-down 5080 chip, GB203, doesn't compete for the same wafers as GB202. GB202 is the more valuable part because it also powers workstation cards. Every one that becomes a GeForce card is one that can't become a pricier RTX PRO board. Nvidia could also be constrained on GDDR7 for a 32GB consumer card. TechSpot noted that Super-series plans had reportedly been paused because of the high cost of 3GB GDDR7 chips, amid an AI-driven memory crisis.
Both explanations are plausible, but neither is confirmed by Nvidia. For a buyer, the outcome is the same either way: if stock isn't there, it doesn't matter why.
The three-generation drift
The XDA piece that prompted this story makes a useful historical point using core counts:
| Generation | 80-class cores | Flagship cores | Ratio |
|---|---|---|---|
| RTX 30 series | 8,704 (3080) | 10,496 (3090) | about 83% |
| RTX 40 series | 9,728 (4080) | 16,384 (4090) | about 59% |
| RTX 50 series | 10,752 (5080) | 21,760 (5090) | about 49% |
The 3080 and 3090 shared the GA102 die. Since then, the 80-class card has moved to a smaller chip, and the gap has widened each generation. A 24GB 5080 would extend that trend rather than start it.
Prices and availability: a caution
XDA quotes street prices of about $8,000 for the cheapest RTX 5090 and $1,700 for the cheapest 5080. Those are one writer's snapshots. Nothing establishes the retailer, region, or whether the listings were new, used, or third-party. Other coverage differs: one report says the 5090 reportedly no longer appears in Nvidia's US online marketplace, with many street listings above $5,000. The $1,999 launch MSRP is the only figure here that Nvidia itself set. Check current listings yourself before treating any street price as fact.
What to do now
- If you want a 5090: Fisher's statement is the only official word, and it doesn't promise supply. Don't pay a big premium based on rumor alone.
- If you're on a 16GB card at 4K: Watch for a 24GB 5080, but don't wait on a rumor with no partner documentation. CES 2027 is the likely next checkpoint for official news.
- If you run local AI or content work: VRAM capacity often matters more than gaming-style core comparisons. A 24GB card could be attractive, but price and availability are unknown.
- Watch for signals: An official Nvidia announcement, partner documentation reaching board makers, and what happens to RTX 5090 listings at major retailers. All three would carry more weight than another leaker post.
Bottom line
The rumors are consistent on one point: Nvidia may be putting more of its largest chip into professional products and giving gamers a better-equipped 5080 at the top. Fisher's denial rules out a formal discontinuation, not a tighter squeeze on GeForce supply. Until Nvidia announces something, a 24GB RTX 5080 is a rumor with a plausible rationale, not a flagship.