BuildNot yet confirmed elsewhere1 publisher2 min readPublished
Leaked factory notice says Nvidia will no longer make any RTX 5090 model
Nvidia has stopped selling GB202 chips for every GeForce RTX 5090 board, including China's D models, according to a leaked factory notice. If it holds, local AI teams lose their largest-memory consumer option once existing 5090 stock sells through.
The Engineer · Build desk

What happened
- The same notice names a 24GB RTX 5080 as the new top GeForce card and says the 16GB RTX 5080 will be discontinued.
- A second leaker, @Zed__Wang, confirmed the report and said GB202 will now go only to data center and professional graphics cards.
- Nvidia had not responded to Tom's Hardware's request for comment, and the company typically does not comment on rumors.
- Partners, distributors and retailers likely hold enough RTX 5090 stock to last a while, but Tom's Hardware expects availability to fall and prices to follow.
Why it matters
- constraint A model sized to fit one 32GB GeForce card would have to be split across smaller cards or moved to professional hardware once 5090 stock is gone.
- cost The next memory tier up is priced as professional gear: the RTX 6000 Pro has three times the 5090's memory and can sell for $15,000, per Tom's Hardware.
- decision Buyers who need a GB202-class GeForce card must choose between shrinking 5090 inventory and a Rubin-based RTX 60-series rumored for late 2027 or 2028.
GB202 is one die sold at several price points. Its full form, with 24,064 CUDA cores and 96GB of memory, is the RTX 6000 Pro [4]. The RTX 5090 uses 21,760 of those cores and carries 32GB [3]. That is 90.4% of the full core count (21,760 / 24,064) [7]. The RTX Pro 5500 has exactly the 5090's 21,760 cores, more GDDR7, and a target market of professional visualization and AI [3].
So every die that ships on a 5090 board is a die Nvidia could have sold on a professional card. Tom's Hardware argues that ending GeForce GB202 boards would stop relatively cheap consumer cards from competing with Nvidia's own accelerators in AI [5]. It also argues the move would send limited manufacturing allocation to higher-margin parts at little cost in gaming, because neither the RTX 5080 nor the RTX 5090 has a direct rival [5]. We think that reasoning is sound.
For local AI work, the figure that moves is memory per card. The 5090 has 32GB [3]. The notice's new flagship has 24GB [10]. The largest GeForce frame buffer would shrink by 8GB, or 25% (8 / 32) [6]. That card is not yet a shipping product. Tom's Hardware links it to long-rumored RTX 50 Super plans, including a 24GB RTX 5080 Super that was reportedly postponed over memory supply constraints [13]. Its argument is that with no 32GB 5090 boards consuming GDDR7, the 24GB card gets better memory supply [13].
The sourcing is an unnamed factory, one leaker who posted its notice, and a second who confirmed it [9][11]. The notice ends with "Please spread the word" [10], a request the internet has honoured. Tom's Hardware describes the claim as unverified and not an official Nvidia announcement [1].
We think the economic case makes the notice credible enough for teams that depend on 32GB GeForce cards to plan around it before Nvidia confirms anything.
What to watch
- An Nvidia statement or a board-partner notice confirming or denying the end of GB202 supply for GeForce cards.
- Launch details for the 24GB RTX 5080, including its price and whether it ships under RTX 50 Super branding.
- Retail RTX 5090 pricing and stock as partner and distributor inventory sells through.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence22
- Adoption
- Insufficient
- Hype gap+35
- Incentives60
- Confidence28
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The claim is unverified and is not an official Nvidia announcement.
- [2]
Tom's Hardware asked Nvidia for comment and the company had yet to respond; Nvidia typically does not comment on rumors.
- [3]
The GeForce RTX 5090 has 21,760 CUDA cores and 32GB of memory; Nvidia's RTX Pro 5500 has the same 21,760 CUDA cores with more GDDR7 memory and targets professional visualization and AI applications.
- [4]
The fully-fledged RTX 6000 Pro has 24,064 CUDA cores and 96GB of memory and can sell for $15,000.
- [5]
Discontinuing GB202 GeForce products could stop relatively inexpensive consumer cards competing with Nvidia's professional accelerators in AI, redirecting limited manufacturing allocation to higher-margin products without sacrificing its gaming position, as neither the RTX 5080 nor RTX 5090 has a direct rival.
- [6]
Moving the top GeForce card from the 32GB RTX 5090 to the 24GB RTX 5080 cuts the largest GeForce memory pool by 8GB, or 25%.
- [7]
The RTX 5090 uses about 90.4% of the full GB202 core count.
- [8]
The RTX 6000 Pro carries three times the RTX 5090's memory.
- [9]
Nvidia has reportedly ceased selling GB202 graphics processors for consumer GeForce RTX 5090-series add-in boards, including 'D' variants for the Chinese market, according to a leaked notice from an unknown factory cited by @hongxing2020.
- [10]
"Going forward, the highest-end GeForce graphics card will be the RTX 5080 24GB, while the RTX 5080 16GB version will be discontinued. Please spread the word."
ReportedInsufficientSource: Leaked factory message cited by @hongxing2020, via Tom's HardwareView cited source - [11]
Leaker @Zed__Wang confirmed the report and said that from now on GB202 silicon will be used exclusively on data center and professional graphics cards.
- [12]
Add-in-board partners, distributors and retailers likely hold inventories of GPUs and finished cards that could sustain RTX 5090 availability for some time, but availability will decline if GB202 supply for GeForce ends, affecting prices.
- [13]
Nvidia has long been rumored to be developing RTX 50 Super cards, including an RTX 5080 Super with 24GB of GDDR7, reportedly postponed due to memory supply constraints; the rumored RTX 5080 24GB could be a revival of those plans and, with no RTX 5090 32GB on the market, would have better memory supply.
- [14]
Recent rumors suggest Nvidia's Rubin-based GeForce RTX 60-series GPUs will not arrive until late 2027 or even 2028.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.