Rumors are swirling around Nvidia's upcoming GeForce RTX 50-series graphics cards, with leaked specifications painting a picture of impressive performance and some perplexing design decisions. These cards, based on the new "Blackwell" architecture, are expected to push the boundaries of gaming graphics but also hint at a potential shift in Nvidia's focus towards artificial intelligence (AI) applications.
Leaked specifications from reputable hardware leaker @kopite7kimi suggest the flagship GeForce RTX 5090 will be a powerhouse, boasting a staggering 21,760 CUDA cores and 32GB of GDDR7 memory. This monstrous configuration comes with a hefty 600W total graphics power rating (TGP), highlighting the sheer power of this upcoming card.
The GeForce RTX 5080, while still impressive, appears to take a more conservative approach. This card is rumored to be powered by the GB203 GPU, featuring 10,752 CUDA cores and 16GB of GDDR7 memory, with a 400W TGP. Despite a significantly smaller core count and memory capacity, the RTX 5080 is expected to offer significant performance gains over its predecessor due to the leap to GDDR7 memory.
However, the large performance gap between the RTX 5090 and RTX 5080 raises eyebrows. The RTX 5090 boasts nearly twice the CUDA cores and memory capacity of the RTX 5080, making it seem like a more powerful variant of the 4090 rather than a natural progression from the 4080. This drastic difference suggests Nvidia may be aiming to create a new performance tier or perhaps prioritize AI computing over traditional gaming performance.
Several explanations for this unusual gap have emerged. One possibility is that the GB202 processor used in the RTX 5090 might be built using two GB203 dies. This multi-chiplet design has been rumored for some time and has already been implemented in Nvidia's datacenter GPUs. While this approach offers scalability, using CoWoS-L packaging to enable high-speed interconnects for a consumer product could add significant cost.
Alternatively, Nvidia could have chosen to build a monolithic graphics processor with over 21,760 CUDA cores on TSMC's 4nm-class process technology. This would lead to a massive die size, presenting challenges in yield and potentially making it expensive. However, Nvidia has a history of employing redundancies in its GPUs, suggesting this approach is not entirely out of the question.
Given Nvidia's use of similar chips across different product lines, including desktop, mobile, professional, and data center GPUs, it's plausible that the RTX 5090 might be a repurposed datacenter GPU aimed at AI workloads. This strategy could explain the significant performance disparity and potentially lead to higher pricing than even the exorbitantly priced RTX 4090.
Despite the growing number of rumors and leaks, it's essential to remain cautious. Until Nvidia officially confirms its plans, there's room for speculation and potential changes. The rumored release date of early 2025 also provides ample time for adjustments and tweaks before the actual launch.
The GeForce RTX 50-series holds immense promise for pushing graphical performance to new heights. However, the apparent focus on AI applications and the unusual performance gap between the RTX 5090 and RTX 5080 raise intriguing questions about Nvidia's future direction. As the release date draws closer, we can expect the rumor mill to continue churning out theories and specifications, keeping the gaming community on the edge of their seats.
0 comments:
Post a Comment