Nvidia Stops RTX 5090 Production to Feed Pro Cards

Leakers say Nvidia has stopped making the GeForce RTX 5090 so the GB202 chip can go into RTX Pro cards instead. A factory notice from around October 10 also puts a 24GB RTX 5080 on top of the consumer stack.

Nvidia has not confirmed any of it, and its GeForce site still presents the 5090 as the most powerful card in the lineup.

Two Leakers, One Factory Notice

The hardware leaker hongxing2020 posted a leaked factory notice on October 10 saying Nvidia would no longer sell any RTX 5090 models, including the D variants built for China. The same message said the 16GB RTX 5080 would be pulled, and that a 24GB RTX 5080 would become the highest-end GeForce card.

We have received a notification from the factory: NVIDIA will no longer sell any RTX 5090 models, including the D variants. Going forward, the highest-end GeForce model will be the RTX 5080 24GB, and the RTX 5080 16GB version will be discontinued. Please take note.

Factory notice, posted by hongxing2020

A second leaker, MEGAsizeGPU, posting as Zed__Wang, backed the silicon part of that claim and wrote that GB202 supplies are RTX Pro exclusive from here. That account did not sign off on the 24GB 5080. Nvidia has not announced an end date and has not issued a public comment.

If the notice holds, GeForce loses the only Blackwell card built on GB202. The 5090 uses a cut-down GB202-300 with 21,760 CUDA cores. RTX Pro 6000 boards use a fuller GB202 with 24,064 CUDA cores. The consumer lineup would then sit on smaller dies, with GB203 in the 5080 class at the top.

The Same Die Lists at $8,900 as a Workstation Card

The RTX 5090 went on sale January 30, 2025, at $1,999 with 32 GB of GDDR7 memory on a 512-bit bus. Nvidia still calls it the most powerful GeForce GPU it has made. The RTX PRO 6000 Blackwell Workstation Edition uses the same die family, the same 1,792 GB/s of bandwidth, and 96 GB of GDDR7 with ECC. Nvidia’s marketplace has that workstation card listed at $8,900 on Nvidia’s store.

That is more than four times the GeForce launch price for three times the memory, on a board rated at 600 W instead of 575 W. Every GB202 that leaves the line as a 5090 is one that did not ship as a Pro card. Pulling the cheaper SKU also stops labs from buying 5090s as cut-price stand-ins for workstation silicon.

GB202 ON THREE CARDS

The 5080 column is the card you can still buy. It is half the cores, half the bus, and half the launch price of the 5090, on a different chip. It is not a substitute for GB202. It is the ceiling GeForce would live under if the big die stays in the Pro stack.

Why GDDR7 Keeps Going to AI Racks

The holdup on a fatter 5080 is memory, not a new GPU design. TrendForce wrote on September 30 that GDDR7 demand driven by AI chips is outrunning the PC market, that GDDR7 output stays limited because advanced capacity is going to server DRAM, and that graphics memory prices are still climbing. Conventional DRAM contract prices were projected to rise 10-15% quarter on quarter in the fourth quarter of 2026.

A 24GB 5080 needs 3GB GDDR7 packages in place of the eight 2GB chips that make 16 GB on the current card. Eight 3GB chips would make 24 GB on the same 256-bit bus. Those denser packages have been scarce enough that Nvidia already sat on a Super refresh. VideoCardz wrote in July 2026 that RTX 50 Super boards had reached at least one add-in partner, then Nvidia held the launch because of 3GB GDDR7 cost.

The same squeeze has shown up outside GeForce. Sony has been linked to a later PlayStation 6 window, Valve has said memory capacity is a limit, and Xbox has warned that memory costs would hit pricing and supply for Project Helix. A halo PC card that pairs a huge die with a lot of GDDR7 is the first GeForce SKU you would starve if those packages pay more in a rack.

A 24GB RTX 5080 Does Not Fill the Hole

Hongxing2020’s notice treats a 24GB RTX 5080 as the new GeForce flagship. Separate claims, passed to board partners only by mouth, put that card in the first quarter of 2027. Partners had not received development files when those claims circulated. MEGAsizeGPU did not confirm this half of the story.

The configuration matches older Super rumors: same GB203 die, same 10,752 CUDA cores, more memory. Gaming speed would move little. VRAM would rise 50% if the 3GB modules show up. The 5090’s 512-bit bus and 21,760 cores would stay gone.

WHAT THE 24GB CARD CHANGES

• The die: GB203 stays; this is not a new GB202 GeForce.
• The cores: 10,752 CUDA cores, unchanged from the 16GB RTX 5080.
• The memory: 24 GB on 3GB GDDR7 chips, if supply allows, versus 16 GB now.
• The bus: still 256-bit, so bandwidth does not jump with the extra gigabytes.
• The timing: first quarter of 2027 in the partner briefings, with no public Nvidia date.

Calling that card a replacement for the 5090 only works as catalog language. It is a 5080 with more memory, sold into a hole the Pro line created.

The 32GB Card Local AI Relied On

Gamers feel the missing halo SKU first. The quieter loss is the only 32GB card in the GeForce stack. People running local models bought 5090s because 32 GB was the cheap way to keep large weights on one board, without paying workstation money. Pulling that SKU does not kill home inference. It pushes it onto used 5090s, onto 16GB 5080s that thrash sooner, or onto the $8,900 Pro card.

That demand is why 5090s were already hard to find at $1,999 long before any factory notice. Once new supply stops, leftover partner kits and the second-hand market become the whole channel. Some 5090 owners now say they will skip the next GeForce if the halo card is the one Nvidia gives up. Board partners lose a high-margin SKU they could still sell above list. China loses the D variants in the same notice.

Nvidia can do this because AMD has no direct answer at this tier. GeForce still has 5080, 5070 Ti, and 5070 cards on shelves. The company keeps the gaming brand without spending GB202 wafers on a $1,999 board.

Rubin Still Looks Like 2027 or 2028

The 5090 has been on sale about 20 months, from January 30, 2025, to the October 10 notice. A mid-cycle Super was the usual way to stretch that generation. Memory cost took that option off the table. If GB202 stays Pro-only, GeForce waits on Rubin, the RTX 60 series, which rumor still places in late 2027 at the earliest, and possibly 2028. Rival next-gen PC GPUs are being discussed on a similar clock.

THE BLACKWELL FLAGSHIP CLOCK

1. January 6, 2025: Nvidia announces the GeForce RTX 5090 and RTX 5080 at CES.
2. January 30, 2025: Both cards go on sale at $1,999 and $999.
3. July 2026: RTX 50 Super boards reach at least one partner; Nvidia holds the launch over 3GB GDDR7 cost.
4. October 10, 2026: hongxing2020 and MEGAsizeGPU say new GB202 GeForce supply has ended.
5. First quarter of 2027: Board partners are told, verbally, to expect a 24GB RTX 5080.
6. Late 2027 or 2028: Rumored window for Rubin-based RTX 60-series cards.

Until Nvidia speaks, the consumer page and the factory chatter disagree. The GeForce site still sells a 32GB flagship. The notice says that flagship is done.

WHAT WE KNOW

• The claim: Two regular hardware leakers say new GB202 supply is reserved for RTX Pro, not GeForce.
• The products: RTX 5090 launched at $1,999 with 32 GB; the Pro 6000 workstation lists at $8,900 with 96 GB.
• The memory: TrendForce still has GDDR7 demand led by AI, with graphics DRAM prices rising.

WHAT IS UNCONFIRMED

• Official EOL: Nvidia has not said the 5090 is out of production.
• The 24GB 5080: No public spec sheet, price, or ship date from Nvidia.
• Rubin: Late 2027 or 2028 remains rumor, not a launch plan.

If the notice is right, the next year of high-end GeForce is a 5080-class card waiting on denser memory, while GB202 keeps shipping in workstation boxes. That split was already visible in the price list. The leak is the moment Nvidia would stop pretending the same die still belongs to gamers.