Recently, hackers from South America said that they hacked into Nvidia's database and obtained more than 1TB of confidential data, including Nvidia's algorithm on DLSS and algorithms that limit computing power, in addition to NVIDIA's future graphics card specifications, such as the GPU based on the Ada architecture that everyone is most concerned about, and its specifications have been exposed by these hackers, and it seems that the specifications are still quite powerful.
According to the information provided by the hackers, Nvidia RTX 40 series graphics cards have a total of five models, including AD102, AD103, AD104, AD106 and AD107, the most flagship version of which is naturally the AD102, with 144 SM units, if calculated according to 128 CUDA per SM unit, that is to say, the AD102 has 18432 stream processors, while the RTX 3080 has 84 SM units. That's 10,752 stream processors. That is to say, the specifications of the AD102 are 71% higher than that of the RTX 3080, of course, the actual performance is limited by the GDDR6 video memory, data transmission bandwidth, etc., and it is not a linear growth, but it is estimated that the AD102 should be more than 35% compared to the RTX 3080.
Of course, higher performance means higher power consumption, it is reported that there are already power supply manufacturers specially designed PCIe 5.0 power supply interface, can provide up to 600W of power, it seems that RTX 40 series graphics card electric tiger is no longer a suspense. It is not known whether the existing power supply can meet the power supply needs of future graphics cards.