'Could this be the RTX 6090': Tech analyst reckons Nvidia's just-announced AI GPU hides a tantalizing secret that could throw down a massive gauntlet to AMD (www.techradar.com)

Nvidia quietly unveiled the Rubin CPX — an AI-focused compute GPU shown inside the Vera Rubin NVL144 CPX system — boasting 128 GB of GDDR7 memory and advanced PCIe 6.0 support. Nvidia framed Rubin CPX as an inference engine for long‑context workloads such as high‑definition media generation and research/development, emphasizing large on‑card memory and higher I/O bandwidth to serve massive model contexts and high‑throughput inference pipelines. What’s provoked wider interest is that the Rubin render appears to include full graphics hardware (raster units, display engines and many ROPs), prompting speculation it could be repurposed as a future GeForce flagship — the so‑called “RTX 6090.” Early analysis suggests Rubin CPX could package up to 16 GPCs with 6 TPCs each (≈192 SMs) and, with a Blackwell-style 8‑TPC layout, potentially 256 SMs, ~256 ROPs, a 512‑bit GDDR7 interface and close to 28,000 CUDA cores — roughly a 28% core jump over the GB202‑based RTX 5090. Rubin was shown as an artistic render and Nvidia insists it’s AI‑only, so gaming variants remain speculative, but the hardware choices (big memory, PCIe 6.0, and full GPU blocks) have clear implications for both large‑scale model inference and competitive pressure on AMD in the high‑end GPU market.