The first glimpses of next-gen Nvidia GeForce GPU models are now starting to surface, with rumors circulating about the forthcoming GeForce RTX 5090, which is said to be based on the forthcoming Blackwell GPU architecture and will be based on the GB202 GPU. According to the latest rumors, GeForce RTX 5000 graphics cards based on the new GPU will be 70 percent (1.7x) faster than the current GeForce RTX 4090.
It’s looking as though Nvidia has learned lessons in terms of memory and bandwidth from the current Ada generation, with some serious improvements to the memory and cache setup rumored to be in line for the GeForce RTX 5090.
These include increasing the memory bandwidth by 52 percent compared with the RTX 4090 (resulting in a figure of 1,532 GB/s), a feat that could come from the use of faster GDDR7 memory on the card, or from expanding the memory interface to 512 bits wide, or perhaps even both. Clock speed is also said to be increasing by 15 percent, taking the RTX 4090’s 2520MHz boost clock to 2898MHz.
The above rumors come courtesy of tech leaker panzerlied on the Chiphell forums, who also supplied a few more details. Other rumors from panzerlied concern a 78 percent increase in the amount of cache, which would give you 128 MB if that figure refers to the L2 cache.
Finally the ‘scale’ is rumored to be 50 percent higher than RTX 4090, which could refer to the number of streaming multiprocessors (SMs) in the GPU. If true, this would give the RTX 5090 a huge total of 24,567 CUDA cores – a feat that would be hard to achieve (along with the extra cache) without a die shrink.
However, regular tech leaker kopite7kimi has also waded into the discussion, saying on Twitter/X that the number of units, such as GPCs, in Blackwell GPUs will ‘not increase significantly’, but that the ‘significant changes’ in its structure will result in improved performance. This tech leaker also stated that this generation will see Nvidia move to a multiple-chip model (MCM), as AMD uses in its current AMD RDNA 3 GPUs, such as the Radeon RX 7900 XTX.
Nvidia GeForce RTX 5090 specs
The rumored GeForce RTX 5090 specs list is:
| GPU | GB202 |
| Streaming multiprocessors (SM) | 192 |
| CUDA cores | 24,567 |
| RT cores | 192 |
| Memory | 24-32 GB GDDR7 |
| Memory interface | 512-bit |
| L2 cache | 128 MB |
| Memory bandwidth | 1,532 GB/s |
| Boost clock speed | 2898 MHz |
Nvidia GeForce RTX 5090 specs
Our best estimate for the GeForce RTX 5090 price is $1,699, based on previous releases, as no official price has been announced.
Nvidia GeForce RTX 5090 release date
Our best estimate for the GeForce RTX 5090 release date is 2025, based on leaks and previous releases, as no official release date has been announced.
While it’s good to see some potential big GPU performance improvements coming our way with the next generation, Nvidia also would also need to make sure the changes to cache and memory bandwidth scale down the range. Currently, cards such as the GeForce RTX 4060 Ti and GeForce RTX 4070 Ti suffer from a lack of memory and bandwidth for their prices.
In the meantime, if you’re looking for the fastest gaming graphics card available, you’ll want to pick up the GeForce RTX 4090, which is likely to be the king of the castle for a good while yet. It’s a serious powerhouse, and is our current favorite for 4K gaming in our best graphics card guide.
Are you holding out for the GeForce RTX 5000 lineup instead of upgrading to Ada? Let us know your thoughts the Custom PC Facebook page, via Twitter, or join our Custom PC and Gaming Setup Facebook group and tap into the knowledge of our 420,000+ members.