The GeForce RTX 4090 is already so big that any PC builder worth their salt knows that you’ll need specialized brackets just to keep it from falling and dismantling your computer. The anticipated RTX 5090, Nvidia’s supposed next large and in-charge GPU, is set to be an absolute monstrosity of a graphics card. Leaked specs show it has more than 5,000 more CUDA cores than the Ada Lovelace GPUs and a default TGP that honestly makes me slightly afraid of plugging it in.
Leaked specs first shown via VideoCardz imply the long-rumored Blackwell GPUs from Nvidia are set to steal hearts and wallets and demand a lot from your local electrical grid. Of course, I’m exaggerating, but the rumored 600W power draw would indeed require a sizable PSU in any PC hoping to run it. Compare that to the RTX 4090, which demanded a 450W base.
The 50-series GPUs include an RTX 5080 with 10,752 CUDA cores and 16 GB of GDDR7 VRAM. The RTX 5090 contains a ludicrous 21,760 FP32 CUDA cores and 32 GB of VRAM on a 512-bit bus. Otherwise, the step down from that—the 5080—is a relatively subdued upgrade from the 4080. It contains 1,024 more CUDA cores than the last generation, though it has the same amount of V-RAM with a 176 GB/s jump in memory bandwidth. The leak suggests the 5080 is pulling 400W rather than the previous 320W.
GeForce RTX 5090 to feature 21760 CUDA cores, 32GB GDDR7 memory and 600W, RTX 5080 gets 16GB VRAM/400W (leak specs) https://t.co/BMAJSGOS5N pic.twitter.com/SNgSsVxnfx
— Wario64 (@Wario64) September 26, 2024
The specs were taken from routine and usually reliable Nvidia leaker Kopite7kimi on Twitter. We’ve previously cited them with past RTX 4090 leaks back in 2022. Still, you shouldn’t run out the door for a new power supply just yet. VideoCardz noted the original leaker didn’t specify whether that default power draw was TBP or TGP, which is essentially the difference between a graphics card’s total power requirement and one from most peripherals.
There’s still so much we don’t know about the card, such as its other core values, clock speeds, or cooling apparatus. With those speeds, it would certainly run hot. Kopite7kimi previously claimed the RTX 5090 Founders Edition has a 2-slot cooler rather than being a 4-slot GPU. This makes some sense, especially if Nvidia hopes customers to set the graphics card in anything other than a massive tower. It also means the card could use some non-standard form of cooling, but at this point, it’s too early to speculate.
Both 5080 and 5090 will require a PCIe 5.0 slot, which debuted earlier this year. That’s expected, but it also reminds consumers that they will need to procure a new motherboard if they hope to upgrade from any PCIe 4.0-based graphics card.
Both GPUs seem like solid upgrades for performance in high-resolution gaming. That being said, the 5090 is practically double the supposed performance of the 5080, making me wonder just how costly the card will be at release. You’ll also need to deal with CPU compute power limiting the actual performance gains of any new card. In today’s current games environment, the 5090 seems more built for enterprise heavy lifting rather than game performance. There’s no modern day demanding game like Crysis forcing everybody to upgrade. Most gaming folk will be just fine sticking with the current gen for now.