Nvidia Reportedly Readies 800W RTX 4090 Ti With 18,176 Cores, 48GB GDDR6X

Nvidia GeForce RTX 3080 Founders Edition
(Image credit: Nvidia)

Nvidia is projected to release three high-end GeForce RTX 40-series graphics cards based on the Ada Lovelace architecture later this year targeting demanding gamers. But apparently Team Green is also prepping a rather monstrous graphics board based on its top-of-the-range AD102 GPU that will carry 48GB of memory with a typical board power of 800W, clearly gunning for the top of the GPU benchmarks hierarchy and the fastest of the best graphics cards.

Hardware leaker @Kopite7kimi dubbed Nvidia's flagship GeForce RTX 40-series graphics card 'The Beast' without attributing a model name to it, but normally Nvidia would call it GeForce RTX 4090 Ti or Titan. This board is said to carry a GPU with 18,176 CUDA cores (up from 16,384 CUDA cores in case of the GeForce RTX 4090) as well as 48GB of GDDR6X memory featuring a 24 GT/s data transfer rate (up from 24GB of memory at 21 GT/s in case of the RTX 4090).

Swipe to scroll horizontally
Header Cell - Column 0 GPUFP32 CUDA CoresMemory ConfigurationTBP
GeForce RTX 4090 Ti (aka, The Beast)AD102-4501817648GB 384-bit 24GT/s GDDR6X800W
GeForce RTX 4090AD102-3001638424GB 384-bit 21GT/s GDDR6X450W
GeForce RTX 4080AD103-3001024016GB 256-bit 18GT/s GDDR6?420W
GeForce RTX 4070AD104-275716810GB 160-bit 18GT/s GDDR5300W

Considering the higher number of CUDA cores, additional memory, and substantially higher TBP, the alleged GeForce RTX 4090 Ti should be significantly faster than the alleged non-Ti variant, particularly in high resolutions with high detail settings. Meanwhile, the beasty card is said to feature a TBP of 800W and will need an extremely capable power supply unit as well as a sophisticated cooling system.

The rather extreme thermals of the monstrous Ada Lovelace card might cause consternation among ProViz professionals that work with highly detailed 3D models that can actually benefit from 48GB of GDDR6X memory onboard. Cooling an 800W card would inevitably mean more fans and higher fan speeds, neither of which are pleasant in a professional environment (outside of the data center).

What remains to be seen is how much will Nvidia charge for its 'Beast' graphics board with 48GB of memory. A fully-fledged AD102 chip is projected to cost a lot, and 48GB of GDDR6X memory should carry an extreme price tag too. Given the rumored specs, we expect Nvidia's alleged GeForce RTX 4090 Ti board will feature an MSRP above $2000, the price of the outgoing GeForce RTX 3090 Ti.

By way of comparison, even if Nvidia keeps clock speeds relatively consistent between generations, RTX 4090 Ti could have over 70% more computational power than the RTX 3090 Ti. Memory bandwidth meanwhile may only be 14% higher, so Nvidia would need improved caching or memory compression technologies to scale performance. Regardless, it looks like a massive bump in performance — and power use.

Nvidia has not yet formally announced its GeForce RTX 40-series graphics boards, so consider everything detailed here as rumors and speculation. This is certainly not the first nor the last time we'll hear some juicy rumors about Nvidia's upcoming Ada Lovelace solutions. The real question: What will it punish more, your wallet or your home circuits?

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.

  • peachpuff
    These should be bundled with a psu.
  • Giroro
    I could write something snide about the 4090 Ti popping circuit breakers, but realistically those cards only exist to give away to YouTubers in the "have over 10 people on staff category". All of those people are running commercial power, so it really doesn't matter.

    I'm a lot more upset about the idea their "4070" could be a 300W card. That's significantly higher than even a 2080Ti. I live in an area with this rare event called "Summer", which happens on a semi-regular basis. . .

    A sub-200 Watt card might not even happen from Nvidia next generation, which is wild to me. If these are the results Nvidia is getting, then maybe they should finally remove the useless Ray Tracing software. Game devs don't want to support it. It's time to move on. Stop adding all that cost and power to gaming cards for tech that is really only useful to Hollywood render farms.
  • Roland Of Gilead
    Wow, this is getting insane!

    On average a 80-100w increase on most cards, with the 3090ti to 4090ti even more!

    A 650w PSU with midrange Ada cards may not be sufficient! Lotta folks will have to consider better PSU's to upgrade.
  • Giroro
    peachpuff said:
    These should be bundled with a psu.

    Don't worry, it will be.
    But it probably will be a severely underspecced PSU, which might catch fire.
  • peachpuff
    Giroro said:
    Don't worry, it will be.
    But it probably will be a severely underspecced PSU, which might catch fire.
    Gigabyte psu? 😂
  • King_V
    Just wait 'til the RTX 5090 Ti comes out....

  • Phaaze88
    This... this is really going to be a BFGeforce and not a Titan? I mean, 48GBs of Vram...
  • husker
    Seems to be more for just bragging rights than a practical commercial product and to see how far they can push the limits of the technology. Kind of reminds me of a concept car, but in this case they can sell a few production units as well.
  • spongiemaster
    Everything about this "leak" sounds like Nvidia just trolling people or trying to figure out who is leaking stuff on twitter. If the 4090 has a 450W TDP, how does the 4090Ti have an 800W TDP? Even Nvidia's upcoming enterprise H100 SXM boards top out at 700W. Based on specs alone, there should not be a 350W difference. I can't see Nvidia releasing a 48GB RTX card either. If this card exists in any form, it will be back to the Titan branding.
  • 10tacle
    Interesting specs of the 4080 and 4090 between CUDA cores and memory bus compared to the 3080 and 3090 (if this 4-series info is true) and then seeing where the 3080 TI is. I'm still scratching my head why Nvidia even created the 12GB version of the 3080 which just made it effectively a 3080 Ti at the same price point (and I'm a 3080 Ti owner). But what I find most interesting is Nvidia dropping the 4080 to a 256-bit memory bandwidth bus leaving a large 50% gap to the 384-bit bus for the 4090. One can only speculate what their plans are for a fill.


    CUDA Cores

    Memory Bus

    RTX 409016384384-bit24GBRTX 408010240256-bit16GBRTX 309010469384-bit24GBRTX 30808704/8960320-bit/384-bit10GB/12GB
    RTX 3080 Ti