Nvidia is reportedly still planning fabled RTX 50 Super series for 2026, leak claims — lineup could now include a potential 'RTX 5060 Super' with 12GB of VRAM

News
By
Published

Along with the old classics like the 5070 (Ti) and 5080 Supers.

Blackwell silicon
(Image credit: Nvidia)

A potential "Super" refresh for Nvidia's Blackwell 50-series GPUs has been part of the news cycle for almost a year at this point, with the last substantial sighting coming from Seasonic's PSU calculator nine months ago. Since then, the rumor mill has been mostly silent due to an AI boom-sized component crisis that engulfed all production lines. But now, it seems like the RTX 50 Super series is "back on track," with leaker MEGAsizeGPU claiming that there's even a new SKU in the works.

We've already heard the RTX 5070 Super, RTX 5070 Ti Super, and RTX 5080 Super mentioned before, but this time, there's also an RTX 5060 Super in the mix. This rumored GPU could come with 12GB of GDDR7 RAM, according to the leak, possibly using 4x 3GB modules saturated across the same 128-bit bus that the regular RTX 5060 has. This card could also just be called RTX 5060 12GB, as Nvidia has named a few SKUs like that previously.

The leak doesn't mention other GPUs specifically, but through prior rumors, we can infer that the RTX 5070 Super might feature 18GB of VRAM, while both the RTX 5070 Ti Super and the RTX 5080 Super are suggested to rock 24GB pools — all enabled by 3GB GDDR7 chips. One of the replies in the post above claims that specs for these three SKUs apparently remain unchanged.

Latest Videos From
Swipe to scroll horizontally
Rumored * RTX 50 Super details

Graphics Card

RTX 5080 Super*

RTX 5080

RTX 5070 Ti Super*

RTX 5070 Ti

RTX 5070 Super*

RTX 5070

RTX 5060 12GB (Super)*

RTX 5060

Architecture

GB203

GB203

GB203

GB203

GB205

GB205

GB206

GB206

VRAM (GDDR7)

24GB

16GB

24GB

16GB

18GB

12GB

12GB

8GB

VRAM Bus Width

256-bit

256-bit

256-bit

256-bit

192-bit

192-bit

128-bit

128-bit

CUDA Cores

10,752

10,752

8,960

8,960

6,400

6,144

?

3,840

TGP

415W

360W

350W

300W

275W

250W

?

145W

*= unconfirmed models based on leaks and rumors

We don't know when these GPUs will actually be released, but the leaker says he expects them to still launch in 2026. With DRAM being as expensive as it is right now, and GPU prices affected by the AI boom, these new RTX 50 Super SKUs may not exactly be good value. Back in November of last year, there were murmurs of these GPUs even being cancelled due to just how valuable 28Gbps GDDR7 modules had become.

To add fuel to the fire, Nvidia itself quashed rumors of an RTX 50 Super series announcement at CES 2026 earlier this year. Add to that the leaked RTX 5050 with 9GB of VRAM, highlighting the sheer desperation of the moment, and any hope left was killed. An overpriced RTX 50 Super series launching in these circumstances was unlikely. So, a single churn of the rumor mill isn't enough to reignite excitement; we'll have to wait and see if the frequency of these leaks once again picks up.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

TOPICS
Hassam Nasir
Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

11 Comments Comment from the forums
  • DS426
    I'm surprised that nVidia doesn't call a 5070 with 18 GB of VRAM a "5070 AI" or a 5080 with 24 GB of VRAM a "5080 AI". "Super" doesn't have "AI" in it, so c'mon, that doesn't really work anymore from a marketing perspective.

    I guarantee the thought has seriously circled on some minds there.
    Reply
  • usertests
    I wonder if 5050 9GB will ever appear or if 5060 12GB could be the antidote for the low-end.

    DS426 said:
    I'm surprised that nVidia doesn't call a 5070 with 18 GB of VRAM a "5070 AI" or a 5080 with 24 GB of VRAM a "5080 AI". "Super" doesn't have "AI" in it, so c'mon, that doesn't really work anymore from a marketing perspective.
    This isn't official yet, so there's still time.
    Reply
  • bigdragon
    I have a hard time believing that Nvidia would divert any resources to a new lineup of consumer GPUs given the state of the industry right now. They're barely supplying the current lineup of GPUs. There would need to be a serious dent put in the AI bubble to make such a move.
    Reply
  • thestryker
    If this is still happening I suspect it's driven more by 3GB GDDR7 availability than anything else. We're also late enough in the 50 series that nvidia probably has binned GPUs they need to be using which may add to available capacity. There's also no sign of a future generation for consumer level cards despite being almost 18 months in so they'll have plenty of time to sell.
    Reply
  • Notton
    maybe this is related?
    https://videocardz.com/newz/micron-now-lists-24gb-gddr7-memory-joins-samsung-and-sk-hynix-in-3gb-segment
    In terms of DRAM market share by revenue in 2025, Samsung is 36%, SK Hynix is 32%, and Micron is 23%.
    So that probably brings a sizeable fab online for 24gbit GDDR7.
    Reply
  • usertests
    Notton said:
    maybe this is related?
    https://videocardz.com/newz/micron-now-lists-24gb-gddr7-memory-joins-samsung-and-sk-hynix-in-3gb-segment
    In terms of DRAM market share by revenue in 2025, Samsung is 36%, SK Hynix is 32%, and Micron is 23%.
    So that probably brings a sizeable fab online for 24gbit GDDR7.
    There's that, and it's possible Nvidia ended up with more 3GB GDDR7 than it needs for AI.

    The big Blackwell workstation cards use 96 GB GDDR7 and may be bought one at a time. Blackwell Ultra B300 has 288 GB HBM3e, and are probably bought hundreds or thousands at a time by the median customer.
    Reply
  • thestryker
    usertests said:
    There's that, and it's possible Nvidia ended up with more 3GB GDDR7 than it needs for AI.
    My thoughts were basically that memory manufacturers may be emphasizing 3GB for current production since nvidia is the only customer right now. It would explain the laptop 5070s with 12GB appearing as well.
    Reply
  • PapiPapi
    DS426 said:
    I'm surprised that nVidia doesn't call a 5070 with 18 GB of VRAM a "5070 AI" or a 5080 with 24 GB of VRAM a "5080 AI". "Super" doesn't have "AI" in it, so c'mon, that doesn't really work anymore from a marketing perspective.

    I guarantee the thought has seriously circled on some minds there.
    GTX
    RTX
    AIX
    Reply
  • usertests
    DS426 said:
    I'm surprised that nVidia doesn't call a 5070 with 18 GB of VRAM a "5070 AI" or a 5080 with 24 GB of VRAM a "5080 AI". "Super" doesn't have "AI" in it, so c'mon, that doesn't really work anymore from a marketing perspective.

    I guarantee the thought has seriously circled on some minds there.
    PapiPapi said:
    GTX
    RTX
    AIX
    I'll argue against it on the basis of Nvidia wanting AI users to head over to Blackwell Pro and RTX Spark instead. I have no doubt some AI users will gravitate towards new 18-24 GB gaming cards and have a good time with them, but Blackwell Pro (especially 96 GB) and RTX Spark are built for those >70B LLMs. And they're a lot more expensive. Chances are Nvidia will just call them Supers again.

    In an RDNA5 leak, MLID proposed that AMD should call their new GPUs "PTX" to emphasize path tracing performance. "PX/PTX" would be the gamer-friendly new frontier of stupid GPU naming.
    Reply
  • dva852
    usertests said:
    I'll argue against it on the basis of Nvidia wanting AI users to head over to Blackwell Pro and RTX Spark instead.
    RTX Spark N1X variant is too expensive to go consumer, probably till '28. N1 variant w/ 128-bit 16/32GB will be consumer in '27, but main use case won't be AI.

    Ditto for Strix/Gorgon Halo. AMD can't really undercut Nvidia because of high RAM pricing.

    Pro/dev/biz audience will pay for Spark. But hobbyists are still left wanting for reasonable ($1K'ish) AI solutions, w/ 32GB or 64GB.

    The only Super that would interest AI fans would be either 5070Ti or 5080 24GB. Nvidia is stingy with VRAM count, because of fear of cannibalizing its Pro line, so 5070Ti Super isn't likely, and 5080 Super will likely hit $2K street. (Used 3090 w/ 24GB is ~$1.1K.)

    There's Intel Arc Pro B70 (32GB GDDR6) for $1.1K ASP. But Intel's software stack (PyTorch XPU) is less functional than CUDA and AMD ROCm, at the hobbyist/enthusiast level. It's OK for LLMs, but is brittle for things like ComfyUI.

    With increased attention on LPDDR5X (256-bit wide) as baseline for AI, I think AI GPU will bifurcate from gaming GPU, as LPDDD5X is cheaper, lower-power, allow higher density, and "good enough." The closest to doing this is Intel, with Crescent Island as a bellwether. Its focus is on edge AI. AMD is too busy chasing Nvidia in data center to devote more effort in client.

    Ideally, we'd have Intel dropping AI GPU w/ 32GB LPDDR5X for ~$500, or 64GB for ~$800 some time next year. (Current contract pricing for 32GB LPDDR5X is $140-160 for PC OEMs.) Timing for any consumer AI product will hinge upon Microsoft releasing an AI-capable Windows. Stars will hopefully align in 2H27.
    Reply