英伟达下代旗舰 RTX 6090 推迟到 2028 年
Nvidia's next-gen flagship RTX 6090 delayed to 2028, still 32GB VRAM unchanged. Waited almost two years, only to get the exact same number. The RTX 6090 leak is out: 32GB VRAM, same as the current 5090. According to Moore's Law Is Dead, this chip is codenamed GR202, built on 3nm, 32GB G
📜 Table of Contents
NVIDIA's Next-Gen Flagship RTX 6090 Pushed to 2028, Still 32GB VRAM Unchanged
Waited almost two years, only to get the exact same number.
The RTX 6090 leaks are out — 32GB of VRAM, same as the current 5090. According to Moore's Law Is Dead, the chip is codenamed GR202, built on 3nm, 32GB GDDR7, 525W power draw — actually a bit lower than the 5090's 575W — and still a single 16-pin power connector.
Power draw went down, VRAM capacity didn't budge.
The VRAM didn't increase not because NVIDIA didn't want it to. It's because they can't afford to.
Asking for 8GB More Would Have Killed the Card
NVIDIA had planned a version of the RTX 5080 with 24GB of VRAM, originally slated for a January launch next year. The approach was uncharacteristically simple — swap eight 2GB GDDR7 modules for eight 3GB ones. No core changes, no PCB redesign.
Around November 10th, Board Channels got word from channel sources that NVIDIA had notified its domestic AIC partners the model was cancelled, with no new schedule given. NVIDIA never publicly acknowledged this 24GB version existed — all specs and timelines came from the supply chain.
What's more convoluted is how it died. Back in October, also per supply chain sources, the card was supposedly being accelerated — launching as early as December this year, with some claims it would succeed the discontinued 5090 as the consumer flagship. Less than a month later, it went from accelerated to indefinitely shelved.
There's only one reason: VRAM prices are absurdly high. Per channel sources, a single 3GB GDDR7 module costs roughly three times as much as a 2GB one. Swapping all eight would triple the VRAM cost outright. Early leaked pricing was even more ridiculous — reportedly $2,999, half again the 5090's $2,000 MSRP, for just 8GB more VRAM and half the CUDA cores.
Why are modules so expensive? AI took the capacity. Memory makers are shifting production lines toward HBM, leaving less room for GDDR. In the same batch of reports, the entire RTX 50 Super series was also cancelled — same reason.
So the 6090 staying at 32GB makes sense. Add another 16GB and the cost would blow through the ceiling.
The 5090 Already Went First
In early November, @hongxing2020 on X posted a factory notice saying NVIDIA would stop producing all RTX 5090 models, including the 5090D for the Chinese market. @Zed__Wang added that all future GB202 chips would go to data center and professional cards. Tom's Hardware reached out to NVIDIA for comment and got nothing. The official line is that it's not discontinued, but board partners can no longer get cores or VRAM kits — everything on shelves is inventory.
The math is easy. The same GB202 die in an RTX Pro 6000 with 96GB of VRAM now sells for $16,000 — 87% higher than at launch. Two thousand dollars on one side, sixteen thousand on the other — with only so many wafers to go around, the choice is obvious.
This cut isn't just at the high end. Channel reports say NVIDIA has sent GPU kit price increase notices to AICs worldwide, covering the entire GDDR7 Blackwell line and GeForce's GDDR6 cards. According to Germany's 3D Center, RTX 50 series average prices rose 66% over the past 12 months, with the 5090 up 132% in a year.
This isn't price gouging anymore. This card barely qualifies as a consumer product.
The Calendar Changed Three Times, Each Time Later
January reports said H2 2027. February changed it to end of 2027 for mass production. On September 15, Science and Technology Innovation Board Daily reported H1 2027 was locked in. Late September, it slipped to 2028, citing the mass production pace of high-density GDDR7. On October 28, Kepler_L2 said the mainstream 6060 might slip to 2029.
Three revisions in a year, landing a full year later than the January estimate. Both kopite7kimi and Kepler_L2 point the entire 60 series at 2028. From the RTX 50's January 2025 launch, that's a three-year gap. The old two-year cadence was what the wait-and-see crowd based their math on — at three years, that math completely falls apart.
I'd take pricing with a grain of salt. The $2,500–3,000 range is foreign media extrapolating from the 5090's MSRP, not an NVIDIA quote. Some commentators flat-out said that if VRAM stays at 32GB, $3,000 will be a hard sell to high-end users.
Throughout this whole mess, NVIDIA has said exactly one thing about gaming cards. Someone on X asked Jacob Freeman, GeForce's brand marketing lead, whether GeForce was still happening. He replied with one word: Nope. That denial was about killing the brand, not about making cards. No timeline, no pricing, no roadmap.
I get the silence. Gaming hasn't been NVIDIA's main act for a while, and now they're in a management transition — Jensen Huang stepped down as CEO on November 12, with CFO Cress taking over temporarily. Whether these product decisions are related to the personnel changes, I can't say, and I don't really believe they are. But when management changes, the line that gets sacrificed first is usually the one that doesn't matter.
In China, the Ceiling Is Dropping
The 5090 never had an official channel in China — the 5090D was what sold. Now the 5090D is discontinued too. If the 24GB version is really cancelled, the strongest GeForce available in China will be the 5080 16GB. Current e-commerce pricing: the 5090D at just over 20,000 yuan, the 5080 16GB ranging from 13,000 to 20,000.
The high-end tier went from three models to one, and that one has identical core specs to what launched in January last year.
The VRAM crunch isn't just hitting graphics cards. Domestic 16GB DDR5 memory sticks went from 300–400 yuan to nearly 1,500. The PS6 is rumored to slip to 2028 or even 2029, Valve says new hardware is constrained by VRAM capacity, and Xbox has warned that Project Helix pricing and supply will be affected.
The only thing that's certain is time. From now until the 6090 hits shelves, at least two to three years, and high-end gaming card supply will only get tighter than it is now.
I've followed this product line for years, and this is the first time I've seen the high-end card problem shift from "can't afford it" to "can't buy it." With the 50 series, players could still complain about the price. With this generation, even the thing to complain about has shifted to time.
If you want to wait for a cheap card with lots of VRAM, you'll probably have to wait for AI to loosen its grip first.
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only — no paid placements.