During a keynote at CES 2025 in Las Vegas, which took place tonight and saw the announcement of interesting PlayStation-themed projects, NVIDIA CEO Jensen Huang announced the new desktop GPUs of the GeForce RTX 5000 family, based on Blackwell architecture, from which they take their name.
Let’s see one by one the characteristics of the models presented, the price and the market launch period.
GeForce RTX 5090
21760 CUDA core
bus a 512 bit
32 GB of GDDR7 memory at 28 Gbps for a bandwidth of 1792 GB/s
4th generation RT cores (170)
5th Generation Tensor Core (680)
GeForce RTX 5080
GB203 GPU set at 2.3 / 2.62 GHz
10752 CUDA core
84 RT core e 336 Tensor core
bus a 256 bit
16 GB of GDDR7 memory at 30 Gbps
bandwidth that approaches terabyte, stopping at 960 GB/s
GeForce RTX 5070 Ti
8960 CUDA core
70 RT core e 280 Tensor core
the bus and amount of memory remain unchanged (256 bit, 16 GB GDDR7)
VRAM is set to 28 Gbps
bandwidth di 896 GB/s. La GPU opera a 2,3 / 2,45 GHz
The TGP is equal to 300 Watts, once again supplied via the 16-pin connector.
GeForce RTX 5070
GPU GB205 con 6144 CUDA core (2,16 / 2,51 GHz)
50 RT core e 200 Tensor core
192-bit bus and 12 GB of 28 Gbps GDDR7 memory
bandwidth 672 GB/s, while the TGP is equal to 250 Watts.
Together with the GeForce RTX 5000 GPUs, NVIDIA introduces RTX Neural Shaders, i.e. systems for bringing AI into shaders. NVIDIA describes them as “little AI networks in programmable shaders.” Additionally, NVIDIA announced NVIDIA RTX Kit, a suite of neural rendering technologies for ray tracing games with artificial intelligence.
The prices of the individual models were also presented:
As announced by the company itself, the models presented are now close to being put on the market. Further information is available on the official NVIDIA website.
