Nvidia RTX 50-series detailed in eye-opening leaks
Nvidia’s next line of GPUs, the RTX 50-series, dubbed Blackwell, has had even more information leaked by renowned leaker Kopite7Kimi.
In what must look like ancient hieroglyphics, Nvidia leaker Kopite7Kimi, has revealed a bit more about Team Green’s next-gen GPUs.
Detailing how the cards are structured, Kopite states that the next-gen Blackwell GPUs from Nvidia are already bigger than the 40-series. Kopite also makes the distinction between the GB100 and GB202 chips that Nvidia is working on.
As it stands, GB100 will be Nvidia’s designated hardware for things like data centers, while GB202 will be what the gaming crowd will be wanting to focus on. The way GPUs are made, Nvidia starts with a larger die and then cuts it down for each product.
Kopite’s tweet details the two clusters that make up part of the GPU and gives us some early insight into how powerful they could be.
Speculation surrounding the Blackwell GPUs seems to indicate the way GB202 is going to be put together could be significantly more powerful than the current gen. With 12 texture processing clusters and eight graphics processing clusters, this equals a potential 192 streaming multiprocessors (SMs).
RTX 50-series Blackwell GPUs could house 512-bit bus
NvidiaThe RTX 4090, the current top dog in the GPU pile, has 128 SMs, for a total of 16,384 CUDA cores. Now, it’s expected that the GB202 will feature an estimated 24,576 CUDA cores. On top of this, it could also house a 512-bit bus, which will greatly enhance the overall speed of the device.
In comparison to the 4090, it appears the next-gen Nvidia flagship GPU could boast a significant performance increase over its predecessor.
Previous 5090 rumors have pegged it as being 50% more powerful, but more recent ones claim an impressive 1.7x improvement. There’s no official word from Nvidia outside of a presentation leak on the next-gen hardware just yet, but it’s expected to land later in 2024.