Alleged Price and TGP of RTX 4070 Revealed by Leak

From Tom's Hardware: VideoCardz has published a leaked slide presumably from Nvidia's final presentation of its GeForce RTX 4070 graphics card. The slide reveals a final price of the add-in-board as well as some of its specifications. In particular, it says that the product will cost $599 and will have a total graphics power of 200W.

Nvidia is expected to release its GeForce RTX 4070 graphics card in mid-April, but most of its specifications have been known for some time, albeit not from the company itself. The upcoming vanilla version of Nvidia's GeForce RTX 4070 shares the AD104 graphics processor with the RTX 4070 Ti, but has only 5888 CUDA cores operating at 1920 MHz – 2475 MHz. Just like the 'Titanium' version, the GeForce RTX 4070 features a 12GB GDDR6X memory subsystem with a 192-bit interface.

The new graphics card is said to have a total graphics power of 200W, which is significantly lower than the TGP of Nvidia's GeForce RTX 4070 Ti (285W) and GeForce RTX 3070 Ti (290W). This would enable graphics cards makers to miniaturize cooling systems of the GeForce RTX 4070 and therefore maximize compatibility with more compact computer cases, which will be another reason why this board will likely enter the ranks of one of the best graphics cards available. Idle power and video playback power of the new board will also be lower compared to that of its predecessors, according to the slide.

Despite having a heavily cut-down AD104 GPU, the GeForce RTX 4070 is expected to produce approximately 29 FP32 TFLOPS of computational power, which is comparable to that of the GeForce RTX 3080. However, the RTX 3080 possesses a 320-bit memory bus and an impressive peak bandwidth of 760 GB/s, which is substantially greater than the 504 GB/s bandwidth provided by AD104's 21 GT/s GDDR6X memory. In fact, even a GeForce RTX 3070 Ti has higher memory bandwidth (608 GB/s) than the newcomer. Meanwhile, the new graphics card is supposed to have a large L2 cache, which will likely compensate for a slower memory subsystem.

View: Full Article