Rumors are swirling about Nvidia's highly anticipated next-generation graphics card, the GeForce RTX 5090. According to a prominent leaker, panzerlied, the RTX 5090 is expected to bring significant improvements across the board and boast a massive 32GB of GDDR7 memory.
Compared to its predecessor, the RTX 4090, the RTX 5090 is rumored to feature a 50 percent increase in scale, with cores clocked 15 percent higher. Additionally, it is said to have a 78 percent larger cache and 52 percent higher memory bandwidth, resulting in a substantial 1.7x performance uplift.
If these rumors hold true, the RTX 5090 could potentially house an impressive 192 Streaming Multiprocessors (SMs) and 24,576 CUDA cores, clocked at 2.9GHz. With 128MB of cache and 24GB or 32GB of GDDR7 memory on either a 384-bit or 512-bit bus, it could deliver a remarkable 1.54TB/s of memory bandwidth.
While details regarding power consumption are currently unknown, there are expectations of increased power requirements. Gigabyte has even anticipated a rise in power consumption for its Nvidia-based high-performance computing (HPC) systems, indicating a possibility that the RTX 5090 could push the limits of the 12V2x6 / 12VHPWR connector's 600W rating.
Both consumer and data-center GPU products will be based on Nvidia's next-generation Blackwell architecture. Additionally, rumors suggest that the 50-series consumer GPUs may utilize a Multi Chip Module (MCM) design, which would separate GPU components into separate dies, similar to AMD's RDNA 3 chiplets. This design would offer greater flexibility and customization options.
Despite the excitement surrounding the potential advancements of the RTX 5090, it is unlikely to hit the market until at least 2025. Nvidia's current flagship, the RTX 4090, continues to dominate the high-end space, leaving ample time for speculation and anticipation to build.
– William Daigneault, source article