Nvidia Gefore RTX 4070 will have 10GB video memory
Although the day of Nvidia’s release of GPUs based on the Ada Lovelace architecture is getting closer, news about the GeForce RTX 40 series graphics cards is still full of uncertainty. With the introduction of the 16Pin PCIe 5.0 power supply interface, it seems an indisputable fact that GeForce RTX 40 series graphics cards have higher power consumption, but the GPU specifications correspond to each model are still controversial.
Recently, Twitter user @kopite7kimi revealed that the specifications of the GeForce RTX 40 series GPUs have changed again. GeForce RTX 4090 will be equipped with 16,384 CUDA cores instead of 16,128 as previously reported, which means that the number of SMs has been increased from 126 to 128, the TDP is 450W, and it has 24GB of GDDR6X memory at a rate of 21Gbps. The GeForce RTX 4080 will be equipped with 10,240 CUDA cores, a TDP of 420W, 16GB of GDDR6 memory, and a rate of 18Gbps. It may be that a larger-capacity L2 cache is configured, and the bit width and speed of the video memory are lower than those of the previous generation. The GeForce RTX 4070 will be equipped with 7168 CUDA cores and a TDP of 300W. It is rumored that the memory capacity will be reduced from 12GB to 10GB, which means that the bit width will be reduced from 192 bits to 160 bits, and the rate of GDDR6 will be 18Gbps.