Renowned YouTube channels, RedGamingTech and Moore’s Law Is Dead, have provided additional details regarding the highly anticipated GeForce RTX series codenamed Blackwell. According to their sources, the release of these graphics cards is scheduled for Q4 2024 or Q1 2025, depending on various factors such as the sales performance of the GeForce RTX 40 series during that time and the launch of AMD’s rival GPU, RDNA 4. This flexibility allows NVIDIA to adjust the release date accordingly.
However, it is worth noting that the performance improvement in terms of rasterization, specifically the primary CUDA cores, may not be as significant as previously speculated. Initial reports suggested a remarkable 2.6x enhancement compared to the RTX 40 series, with the 3090 model being upgraded to the 4090 model, resulting in approximately a 1.9x improvement. This news indicates that the performance gain may be more modest than anticipated.
As for the flagship RTX 5090 (GB202) chip, it is expected to feature 192 sets of Streaming Multiprocessors (SM), 36GB of memory with a bandwidth of 36 Gbps, a 384-bit bus, and will still employ a monolithic architecture rather than the rumored modular design. This decision to stick with a single monolithic structure has been a topic of discussion for quite some time. Additionally, the manufacturing process for these chips will utilize TSMC N3E technology.
TL;DR: The highly anticipated GeForce RTX series, codenamed Blackwell, is set to be released in either Q4 2024 or Q1 2025, depending on various factors. The performance improvement in rasterization may not be as significant as previously speculated, and the flagship RTX 5090 chip is expected to feature 192 SM sets, 36GB memory, a 384-bit bus, and a monolithic architecture, while utilizing TSMC N3E technology.
Leave a Comment