The RTX 5090 is Nvidia's flagship consumer GPU with 32GB of VRAM, used for local AI inference as well as gaming. Recent coverage has centered on supply and pricing: a report claimed Nvidia stopped taking orders from board partners, while Nvidia's Jeff Fisher denied the card is discontinued. Other stories describe AI firms buying pallets and pushing prices to $5,000, and the theft of two pallets of ZOTAC cards worth about $470,000 at a Brazilian airport.
Rumors say the 16GB RTX 5080 has been removed from NVIDIA's marketplace and will be discontinued alongside the RTX 5090. A 24GB model using the same GPU die is reportedly planned for early 2027, with VRAM the main change. Added AI appeal could tighten stock and pricing.
A report shared on Reddit says Nvidia is reportedly no longer accepting RTX 5090 orders from board partners, with even backorders unavailable. The claim is unconfirmed; if true, it could worsen supply constraints for the flagship GPU.
A Reddit post says RTX 5090 cards are not gone but appear in Chinese labs, where technicians remove their dies and 32GB VRAM to build AI servers. The practice highlights how high-end GPUs are being dismantled and reused amid AI compute demand.
A LocalLLaMA user reported melting a wall plug and socket while performance-tuning a 4x RTX 5090 rig, saying a standard 13A fuse was insufficient and recommending a 16/20A socket. No power figures were provided.
According to a Reddit post, two pallets of ZOTAC RTX 5090 graphics cards worth about $470,000 disappeared at a Brazilian airport, highlighting theft risks for high-value AI hardware in logistics.
Nvidia SVP of Personal Computing Jeff Fisher explicitly denied recent rumors that the RTX 5090 is being discontinued, saying at a post-event interview in Taipei that the GPU will continue shipping.
A Reddit post says the RTX 5090, once a consumer gaming card, is now appearing on smugglers' menus alongside cigarettes, reflecting gray-market demand for high-end GPUs under controls.
Open-source Project Maya runs the 321B-parameter GLM-5.3-Flash locally on one NVIDIA GPU by tiering experts across GPU, RAM and SSD, reaching 30+ tokens/s and up to 118 tokens/s on four RTX 4090s while retaining 97.7-99.2% of FP8 zero-shot accuracy.
A Reddit poster says AI firms are buying gaming GPUs by the pallet, driving the RTX 5090 to $5,000. It signals AI compute demand spilling into the consumer GPU market and inflating prices.
NVIDIA is reportedly discontinuing the RTX 5090, reserving GB202 GPUs for its RTX PRO series.
A Reddit user reports that buying an RTX 5090 at Micro Center now requires paperwork, including signing a no-export declaration, suggesting high-end GPU export controls are reaching retail sales.
AMD's new 256-core EPYC features 16-channel DDR5-12800 memory, reaching 91% of an RTX 5090's memory bandwidth. The setup targets local large-model inference, though the memory kit is expected to be extremely expensive.
GameGPU reports that AMD's next-gen Radeon RX 10800 XT may outperform Nvidia's RTX 5090 by 15-25% in 4K gaming and local AI workloads, per a Reddit share. No official confirmation yet.
A Reddit post argues that instead of paying about $9,000 locally for an RTX 5090, it's cheaper to fly to Taiwan and buy it at a retail price around NT$129,990 (~$4,093), with the poster listing round-trip airfare and a cost comparison.
Nvidia's RTX 5090 disappeared from US online retail listings, with third-party sellers asking up to about $9,500 for the top GPU.
A Reddit user is weighing selling an RTX 5090 for $5k to buy a Mac Studio M5 Ultra priced $5,499, mainly for coding. They note the 5090 has 1.8 TB/s memory bandwidth versus the M5 Ultra's 1.2 TB/s and ask if the upgrade makes sense.
That is everything