Sources with RedGamingTech are claiming that NVIDIA has resurrected its plans for a GeForce RTX 4090 SUPER/Ti and a new version of the TITAN RTX due to delays with Blackwell, the next-generation architecture that is expected to power the GeForce RTX 50 Series. The GeForce RTX 4090 SUPER/Ti could feature 1,024 more CUDA cores and deliver 10% faster performance than the current Ada flagship, according to specs that RGT shared in a video this week, while the new TITAN RTX could pack twice the VRAM of the 4090. Blackwell GPUs have reportedly been delayed to 2025.
GeForce RTX 4090 SUPER/RTX 4090 Ti rumors:
- 136 SMs (vs. 128)
- 2,520 MHz~ boost clock
- 96 MB L2 Cache (vs. 72 MB)
- 384-bit
- 1.1 TB/s bandwidth (vs. 1 TB/s)
- 24 GB GDDR6X running at 23 Gbps (vs. 21 Gbps)
- 450 W
TITAN RTX (Ada) rumors:
- 142 SMs (1,792 cores more than 4090)
- 2,520 MHz~ boost clock
- 96 MB L2 Cache
- 384-bit
- 1.1 TB/s bandwidth
- 48 GB GDDR6X running at 23 Gbps
- 550 w
NVIDIA on the TITAN RTX (2018):
NVIDIA TITAN RTX is the ultimate PC GPU for the world’s most demanding users—AI researchers, data scientists, and content creators. Powered by NVIDIA Turing, NVIDIA’s next-generation GPU architecture designed for AI and ray tracing, TITAN RTX delivers the best PC performance for training neural networks, processing large datasets, and creating ultra-resolution video and 3D content.
TITAN RTX features 576 multi-precision Turing Tensor Cores that deliver up to 130 teraFLOPS (TFLOPS) for deep learning training; 72 Turing RT Cores that provide up to 11 GigaRays per second for maximum real-time ray tracing performance; and 24 gigabytes (GB) of GDDR6 memory for training with higher batch sizes, processing larger datasets and animation models, and managing the most demanding creative workflows. Pair two TITANs together with NVIDIA NVLink and double your memory and performance.
Source