Leaker claims Nvidia plans to launch RTX 5080 before RTX 5090 — which would make perfect sense for a dual-die monster GPU

Nvidia GeForce RTX 5090 fake mockup joke
(Image credit: Nvidia / Tom's)

Renowned leaker @kopite7kimi believes Nvidia will launch the upcoming Blackwell RTX 50-series GPUs this fall, starting with the RTX 5080 rather than the flagship 5090. As with any unofficial rumor, sprinkle liberally with salt. Still, it could make plenty of sense, depending on the specifications and configurations Nvidia uses for what will inevitably be some of the best graphics cards when Blackwell arrives.

A bit of historical review is in order to set the stage. Two years ago, Nvidia launched it's RTX 40-series Ada Lovelace GPUs by starting with the top RTX 4090 on October 12, 2022. It was a tour de force for the green team, with a massive leap in generational performance — and only a minor $100 increase in price relative to the prior generation RTX 3090. The step down RTX 4080 arrived on November 16, 2022, one month later. The RTX 4070 Ti (after a rebranding) launched on January 5, 2023, with the rest of the GPUs following over the coming months. But while that's what Nvidia did with the RTX 40-series, it's by no means the precedent.

The RTX 30-series Ampere GPUs for example began with the RTX 3080 on September 17, 2020, with the top-tier RTX 3090 coming one week later. RTX 3070 arrived on October 29, 2020, with the RTX 3060 Ti launching on December 2, 2020. So, everything was basically in order from the top down, except the first two GPUs were swapped. The RTX 20-series Turing GPUs also had the RTX 2080 launch one week ahead of the top-tier RTX 2080 Ti, while earlier generations often didn't have the "ultimate" GPU come out until nearly a year after the x80-class card — GTX 1080 Ti launched on March 5, 2017 while the GTX 1080 came out on May 27, 2016.

Kopite7kimi tends to have accurate information about Nvidia's plans, and was making a comment on a different rumor that claimed Nvidia would only release the flagship RTX 5090 this year, with the less expensive parts arriving in 2025. But Kopite7kimi suggests that Nvidia's RTX 5080 is a 2024 product, and that it will come out before the top-of-the-line RTX 5090. That agrees with earlier rumors that indicated both GPUs would launch in 2024.

The range-topping RTX 5090 is projected to be based on the GB202 GPU, while the RTX 5080 will use the GB203 GPU. Both of those are merely codenames, and there's no clear indication of what other specifications might exist. Other rumors say that Nvidia decided not to create a GB204 chip, but instead it will have GB205 this time around. But here's where things get a bit interesting.

Nvidia's next-generation Blackwell GB200 data center and AI GPUs have been announced, and there are some big changes in store. The GB200 processors will be made on TSMC's custom 4NP manufacturing technology, a refined version of the current 4N fabrication process that used for Hopper H100/H200 and Ada Lovelace RTX 50-series GPUs. As a refinement of an existing node, it won't enable significantly higher transistor density, nor will it provide a full node performance uplift. To create a higher performance flagship solution, Nvidia opted to link two full reticle size chips together via a new NV-HBI link (Nvidia High Bandwidth Interface) that provides 10 TB/s of throughput.

Given what we know of the data center Blackwell solutions, it's entirely possible that Nvidia could take a similar approach with the top consumer and professional GPUs. Specifically, GB202 could be two GB203 chips linked via NV-HBI, doubling the potential core counts, memory capacity, and memory bandwidth — without any of the messiness of previous dual-GPU solutions. Note also that rumors of a 512-bit memory interface for RTX 5090 have popped up, with a large gulf between it and the rumored 256-bit RTX 5080.

This is speculation on our part, but if the RTX 5090 does indeed use a dual-die solution like GB200, having RTX 5080 launch first would be completely reasonable. In fact, it would almost be foolhardy to try to push out the dual-die cards before the single-die solution.

Either way, we have credible rumors that Nvidia plans to launch Blackwell consumer GPUs this year. With a smaller die on GB203 and the possibility of linking two chips together for GB202, Nvidia could be trying to kill two birds with one stone — or chip. Smaller chips are easier to debug and manufacture, and having a halo product that doubles the specifications in every area relative to the penultimate GPU would certainly attract some deep pockets. We'll find out more about the RTX 5080, RTX 5090, and other Blackwell GPUs later this year, and it seems likely we'll see at least two of the GPUs launch in 2024.

Anton Shilov
Freelance News Writer

Anton Shilov is a Freelance News Writer at Tom’s Hardware US. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

With contributions from
  • Notton
    Using flawless silicon in RTX5080, and imperfect in RTX5090 is certainly a clever way to improve useful yields.

    I totally expect a Titan-class next year featuring the full 512-bit bus with 3GB density VRAM. (48GB total)
    Reply
  • Eximo
    Notton said:
    Using flawless silicon in RTX5080, and imperfect in RTX5090 is certainly a clever way to improve useful yields.

    I totally expect a Titan-class next year featuring the full 512-bit bus with 3GB density VRAM. (48GB total)

    Well, not quite what they have done in the past. 4080 is not a flawless chip, that is where the 4080 Super came from.

    3080 was not the full chip, 3080 12GB, and 3080 Ti, 3090, and 3090 Ti were all variants with only the 3090 Ti being the fully enabled chip.

    2080 was also not fully enabled, that came in the from of the 2080 Super or the RTX 5000. With the 2080 Ti being on a different GPU.

    Last 80 series that came fully enabled at launch was the GTX1080, I believe.

    Sounds like a complete departure here with the 5080 being a monolithic chip and the 5090 being a multi chip module card.
    Reply
  • valthuer
    Notton said:
    I totally expect a Titan-class next year featuring the full 512-bit bus with 3GB density VRAM. (48GB total)

    If 5090 ends up being a beast of a card, i can definitely see Nvidia cancelling any plans for a Titan GPU, just like they did with 4090 Ti. Especially considering the fact that no decent high-end competitor can be expected from AMD's side, any time soon.
    Reply
  • vanadiel007
    Nvidia has to be careful not to go the Intel way, and end up with unreasonable high power consumption just to squeeze out extra performance.
    Reply
  • YouFilthyHippo
    Maybe they'll be smart enough to switch back to the standard 8-pin power connectors
    Reply
  • oofdragon
    Of course it makes sense to launch the 5080 months before, it will be faster than the 4090 so every YouTuber and hipster will buy it even it costs more than 4090. And why wouldn it, it's faster after all! Now vida pushing 80 tier to around $2k and 90 tier to 2 and a half or even 3
    Reply
  • Giroro
    I could see the 5080 being the lowest tier chip that Nvidia launches next gen, and increasing it's price to $1500, if not $2000. But I also wouldn't be surprised if Nvidia launches their Workstation 5080 equivalent cards first at $4k-$5k and pushed back the gaming cards by a year or two.
    Who's going to stop them? AMD/Intel probably are not even going to try to beat the 4090. Even if they could match it in gaming, the 4090 is not primarily being bought as a gaming card, people are buying it for work (And I bet that makes Nvidia unhappy).
    Nvidia doesn't even want to waste their precious luxury silicon on gamers in the first place. I'm sure our crumpled dollars would sully their elite brand if us commoners could ever manage to scrape together $2,000 for their lowest tier, entry level product.

    Or maybe Nvidia's executives have become so rich, so quickly, they'll just assume $2000 is a really good value and everybody can afford it. Sort of a "I've gotten 20x richer in the last 3 years, so surely even the poorest of people could afford to pay at least twice as much as they could last gen", kind of mentality.

    Either way, I'm fully prepared for the next gen to be the worst value and least competitive generation of gaming graphics cards, ever. It's an easy prediction considering their pattern of consistently redefining "worst value" for the current gen, and the last gen, and the gen before that...
    Reply
  • Geef
    Don't worry about yields!They should start with a 5050 and each month go up a tiny bit from there! 5050ti...5060...5060ti...
    Torture Everyone!!They would get tons of rich guys who just can't wait and they buy a slower card for the several months until the fastest card is available!
    Reply
  • Phaaze88
    Geef said:
    Don't worry about yields!They should start with a 5050 and each month go up a tiny bit from there! 5050ti...5060...5060ti...
    Torture Everyone!!They would get tons of rich guys who just can't wait and they buy a slower card for the several months until the fastest card is available!
    Nothing really wrong with continuing with the current schedule, with the early adopters getting bitten by the later release of Super/Ti refreshes.
    Reply
  • 35below0
    oofdragon said:
    Of course it makes sense to launch the 5080 months before, it will be faster than the 4090 so every YouTuber and hipster will buy it even it costs more than 4090. And why wouldn it, it's faster after all! Now vida pushing 80 tier to around $2k and 90 tier to 2 and a half or even 3
    Yeah. The 5080 will compete with and lose to a 5090 so sell it first. Then release 5090 and take the money from those silly enough to upgrade from a 5080 to a 5090.

    Probably it will be priced at the 4090 level or even slightly lower. Only slightly mind! Enough for the headlines.
    Reply