GeForce RTX 4080 may be equipped with 9728 CUDA cores

Although the release time of the Ada Lovelace GPU is getting closer, Nvidia still seems to have some flags, and the specifications of the GeForce RTX 40 series have not been fully confirmed. Nvidia is rumored to have revised the GeForce RTX 40-series specs several times during the product pre-release cycle, and at times, the changes have been relatively large.

Recently, Twitter user @kopite7kimi revealed that the GeForce RTX 4080 is equipped with AD103-300-A1, has 9728 CUDA cores, the corresponding number of SMs is 76 and has 16GB of GDDR6X memory with a rate of 21Gbps. The PCB design is PG136/139-SKU360, the power consumption of the entire graphics card is about 420W, and the 3DMark Time Spy Extreme score is about 15,000 points, which is about 35% higher than the GeForce RTX 3090 Ti.

Compared to the rumored 10,240 CUDA cores or 80 SMs in the past, the specs are down. Considering that Nvidia switched from Samsung to TSMC this time, coupled with the advanced process, the core frequency may be greatly improved. Samsung’s 8nm is strictly an optimized version of 10nm, which lags behind TSMC’s 7nm process node, while Nvidia skips the latter this time and advances to the 5nm process node, which theoretically will have a qualitative leap.

In addition, Nvidia has revised the GeForce RTX 4070’s specifications again, increasing the number of SMs from 56 to 60, and the number of CUDA cores from 7168 to 7680 accordingly. The equipped video memory is changed from 10GB GDDR6 to 12GB GDDR6X, and the bit width is also increased from 160 bits to 192 bits, the speed has also increased to the same 21Gbps as the GeForce RTX 4080, and the memory bandwidth has been directly increased from 360GB/s to 504GB/s. The change is relatively large. The score of 3DMark Time Spy Extreme exceeds 11,000 points.