GeForce RTX 4070 Ti is a high-end graphics adapter that provides computing performance up to 40 TFLOPs. This card is based on an AD104 processor with 7680 CUDA cores, which is the full configuration of this GPU. The card is equipped with 12 GB of GDDR6X memory with an effective clock frequency of 21,000 MHz. Since the GPU is attached to a 192-bit memory bus, the maximum bandwidth will be limited to 504 GB/s, which is actually 104 GB/s slower than the predecessor GeForce RTX 3070 Ti.
GeForce RTX 4070 Ti should only come to the market in variants from partners, which means no Founder Edition, although the render above may indicate otherwise. All of the previously announced original RTX 4070 Ti models that have been shown so far have three fans and a 2-3.5-slot cooling design. That should be more than enough for a 285W TDP.
This is already the third desktop card based on the Ada Lovelace architecture and the third Ada GPU, after the AD102 (RTX 4090) and AD103 (RTX 4080). All of them are based on the same TSMC 4N process. The AD104 GPU will have about 10 billion fewer transistors than the AD103 (a total of 35.8 billion) and a crystal size of about 295 mm².
As for performance, the developers ambitiously point to the gaming advantage of the GeForce RTX 4070 Ti 12 GB over the flagship of the previous line – GeForce RTX 3090 Ti 24 GB. Very soon, independent reviews will be available, which will allow you to make a more balanced assessment of the capabilities of the new graphics card.
GeForce RTX 4070 Ti will go on sale on January 5. The recommended price is $799. However, this is the starting value for basic models, while forced versions with enhanced VRM and efficient coolers will probably cost more.