RTX 5090 vs. RTX 6090: The Next Generation of AI and 3D Rendering from NVIDIA

The announced confrontation between the GeForce RTX 5090 and the upcoming RTX 6090 embodies NVIDIA's meteoric rise in the field of high-performance computing and professional workflows. These two graphics cards symbolize both the culmination of the current architecture and the expected technological leap for AI and 3D.

AI and Graphics Power: The Blackwell Era (RTX 5090)

Since its launch in early 2025, the RTX 5090 has established itself as the benchmark solution for local AI computing and high-end graphics creation. Based on the 5nm Blackwell architecture, it offers:

21,760 to 24,576 CUDA cores and 680 5th generation Tensor cores.

32 GB of GDDR7 on a 384/512-bit bus, bandwidth up to 1.8 TB/s.

A solid TDP of 575-600W and a boost frequency of 2.4 GHz.

Benchmarks place the RTX 5090 far ahead of the RTX 4090 (+37% in Vulkan, +35% in Blender rendering, +44% in Unreal Engine 5 path tracing). For AI creation and computing, this card delivers:

Up to 3,352 AI TOPS in AI, 5,841 tokens/second on the LLM Qwen2.5-Coder-7B model, and exceptionally low latency on local inference tasks.

Optimized handling of creative workloads: 9:19 for Maya Arnold 4K rendering versus 14:49 for the RTX 4090, +28% gain for V-Ray RTX in architectural ray tracing (10,486 vpaths vs. 8,191).

NVIDIA Studio certification guarantees stability, compatibility, and specific features for major professional software.

Towards the Rubin Revolution: What the RTX 6090 promises

Designed for 2027, the RTX 6090 will represent a turning point thanks to the new Rubin architecture engraved in 3 nm TSMC. Based on the consumer-grade Rubin CPX datacenter GPU, it promises:

From 25,000 to 28,000 CUDA cores (+28% vs. 5090), 225 improved 4th-generation RT cores, 32 to 48 GB of GDDR7 ECC, 512-bit memory bus, and up to 2 TB/s of bandwidth.

Major innovations in AI thanks to 6th-generation Tensor Cores (>5,000 AI TOPS), compatible with local AI model inference up to 100B parameters.

New GDDR7 memory with on-die ECC, real-time data poisoning and scrubbing, for stable, secure, and fast workloads.

A leap in high-resolution rasterization with 256 ROPs and native path tracing support. Real-time 4K/8K and DLSS 5 for multi-frame generation.

Initial projections suggest an 80% performance improvement over the RTX 5090, boost frequencies of 3,100-3,300 MHz, and a unique positioning for generative AI in 3D creation: textures, animations, and multimodal content adapted in real time according to the needs of studios and researchers.