NVIDIA GeForce RTX 4070/4090 specifications exposed
According to past news, in the Ada Lovelace architecture GPU that NVIDIA will release this year, the GeForce RTX 4090 is equipped with AD102, and the NVIDIA GeForce RTX 4070 is equipped with AD104. Compared with the current Ampere architecture products, there may be some unexpected places in the specifications and performance.
Recently, Twitter user @kopite7kimi revealed some specifications of the GeForce RTX 4070 and GeForce RTX 4090. GeForce RTX 4090 corresponds to AD102-300 GPU, PCB is PG137/139-SKU330, video memory is 24GB GDDR6X, video memory rate is 21Gbps, and TDP is 600W. GeForce RTX 4070 corresponds to AD104-400 GPU, PCB is PG141-SKU341, video memory is 12GB GDDR6, video memory rate is 18Gbps, and TDP is 300W.
It is said that the GeForce RTX 4090 will have a maximum of 144 SMs, that is, 18,432 CUDA cores, which is a 71% increase in size compared to the GA102. At the same time, NVIDIA will also increase the capacity of the L2 cache, from 6MB of the GA102 to 96MB. Since the video memory bit width is 384 bits, plus the rate of 21Gbps, the video memory bandwidth will reach about 1.0 TB/s. Judging from past news, the power consumption of GeForce RTX 4090 is 600W or above, it seems that there is no suspense, and it needs to be equipped with two 16Pin 12VHPWR interfaces. There has been news in the past that the GeForce RTX 4090 has a memory rate of 24Gbps, and Nvidia seems to have lowered the specs.
The GeForce RTX 4070 will be equipped with 60 groups of SM, that is, 7680 CUDA cores, the L2 cache is 48MB, and the video memory bit width is 192 bits. With the rate of 18Gbps, the video memory bandwidth will reach about 400 GB/s. Compared with the current 290W of the GeForce RTX 3070 Ti, the power consumption of the GeForce RTX 4070 has only increased by 10W. It is rumored that the performance of the GeForce RTX 4070 is comparable to that of the GeForce RTX 3080 Ti, and will be equipped with a 16Pin 12VHPWR interface.
NVIDIA will also use the AD102 on the GeForce RTX 4080. It is rumored that the memory width will be reduced to 256 bits, the memory capacity will be 16GB of GDDR6X, and it will be equipped with a 16Pin 12VHPWR interface. The advantage is that the power consumption will be much lower than the GeForce RTX 4090, which may be between 350W and 450W, which is close to the graphics card equipped with GA102.
