It was previously reported that Nvidia has saved its best AD102 for the next-tier GeForce RTX 4090 Ti. It has 18,176 CUDA cores, and 96MB of L2 cache, and is equipped with GDDR6X memory at a rate of 24Gbps. The memory bandwidth will be increased from 1TB/s to 1.1TB/s, the acceleration frequency is 2.75 GHz, and the actual game frequency can even reach 2.95 GHz. Compared with the GeForce RTX 4090, the overall performance is improved by 10% to 20%, and the power consumption is only 475W.
The AD102 on the GeForce RTX 4090 Ti is not the core of the complete specification, and there are still two sets of SMs missing. Some people speculate that Nvidia may also launch a complete core “TITAN”, equipped with 48GB of video memory, and the power consumption will be further increased to 600W or more. However, this situation does not seem to happen. Twitter user @kopite7kimi revealed that Nvidia has abandoned the launch of the TITAN with the Ada Lovelace architecture, and instead transferred the work to the GeForce RTX 4090 Ti, reducing the specifications of the video memory.
This news has also sparked new speculation. Some people wonder whether Nvidia will abandon the TITAN series. It has not launched any TITAN based on Ampere architecture GPU before. If the Ada Lovelace architecture GPU is still the case, the last TITAN series product is the TITAN RTX, built on the Turing architecture GPU.
At present, NVIDIA has provided AD102 products for professional workstations. The NVIDIA RTX 6000 workstation graphic card was previously released, equipped with 48GB of GDDR6 video memory with ECC, the graphics card is dual-slot thickness, and the power consumption is 300W. From this high-end professional workstation graphics card, the design of the Ada Lovelace architecture GPU is still very efficient.