Nvidia has officially launched the much-anticipated GeForce RTX 5090, the flagship graphics card of its latest GeForce 50 series, at CES 2025. This powerhouse has already begun making waves in the world of gaming and creative professionals. However, just before its release, rumors have emerged regarding a possible RTX 5090 Ti variant that could feature cutting-edge upgrades, including a fully enabled GB202-200-A1 GPU and dual 12V-2×6 power connectors, which could theoretically push the power requirements to an astonishing 1,200 watts.
These rumors began circulating after a prototype image surfaced on the Chinese tech forum, Chiphell. According to a report from ComputerBase, this prototype could have a GB202-200-A1 GPU with 24,576 shaders, an impressive increase in performance. This chip would feature 192 active streaming multiprocessors, suggesting it is a fully unlocked version of Nvidia’s GB202 architecture. Along with the enhanced GPU, speculation points to memory upgrades as well: the RTX 5090 Ti could sport 32GB of GDDR7 memory, clocked at a remarkable 32Gbps instead of the RTX 5090’s 28Gbps, pushing the bandwidth beyond 2TB/s.
Adding fuel to the fire, ComputerBase uncovered shipping documentation from NBD Data listing a mysterious graphics card with 96GB of GDDR7 memory marked as “for testing.” This suggests the presence of a high-end professional model, which might well be the RTX 6000 Blackwell — Nvidia’s next-gen workstation graphics card. The large memory capacity and performance indicate this model is tailored more for specialized tasks like AI processing and other data-intensive applications, rather than standard gaming or creative workloads.
>>>VB18 Replacement Battery for Godox V860II V850 V860C V860N Speedlite Flash
As for the GeForce RTX 5090, it is already a beast in terms of performance. With 32GB of GDDR7 memory arranged in sixteen 2GB modules and a 512-bit memory interface, the RTX 5090 delivers substantial bandwidth, making it a top choice for gaming, video editing, and professional design. However, if Nvidia were to use 16GB 3GB chips instead of the 2GB variants, the 48GB memory configuration would be entirely feasible. This could push the RTX 5090 even further, but the true standout would be the 96GB memory configuration mentioned in the shipping records.
The 96GB setup suggests a dual-sided memory layout, with chips arranged in a “clamshell” configuration across the front and back of the card. This design could be a feature of the professional-grade RTX 6000 Blackwell, doubling the memory capacity of the current RTX 6000 Ada, the world’s most expensive consumer GPU, while leveraging a 512-bit memory bus for optimal bandwidth.
What’s Next? The Blackwell Series and AI-Optimized Cards
The Nvidia Blackwell series, slated for an official debut at the GPU Technology Conference (GTC) 2025, is expected to take the graphics card market by storm. This lineup will likely include both consumer and workstation models, with the RTX 6000 Blackwell emerging as the flagship professional card.
While 96GB of GDDR7 memory might seem excessive for typical gaming or creative workloads, it opens up new possibilities for AI tasks and machine learning. Nvidia is likely to introduce an AI-optimized version of the RTX 6000 Blackwell, catering specifically to enterprises and research institutions that require massive memory and computational power. As we approach March 2025, more details about this groundbreaking technology will undoubtedly surface, offering a glimpse into the future of high-performance computing.
>>>ATS065T-P120 for TT Electronics ATS065T-P120