Nvidia leaks confirm that the RTX 4090 will have 70% more Cuda Cores and 16 times more L2 cache than the RTX 3090 Ti

NVIDIA is not having a good time right now as the hackers who attacked it have started leaking sensitive data. First of all, it was confirmed that the successor to NVIDIA Hopper has the code name Blackwell, and the source code for DLSS.

According to @davideneco25320, who has apparently seen the full specs of the next-gen NVIDIA GPUs in said files, the company is ready to launch up to 5 new desktop GPUs: AD102, AD103, AD104, AD10 and AD107.

The table of specifications published by David presents the families of Turing GPU, Ampere y Yes, the GeForce RTX 40 series will feature up to 144 Streaming Multiprocessors, which would be the specifications of the AD102 GPU. If NVIDIA kept the same CUDA core count per SM, this would mean a maximum of 18432 cores.

The company is also planning the AD103 GPU with 84 SM (10752 CUDA cores) and AD104 with 60 SM (7680 cores). The entry-level and mid-range AD106 and AD107 GPUs should offer 36 SM (4608 cores) and 24 SM (3072 cores) respectively.

The NVIDIA GeForce RTX 40 series is expected to launch in September this year.

With all this said, NVIDIA RTX 40 GPUs will be equipped with up to 96 MB of L2 cache, 90 MB more than RTX Ampere GPUs. The data purportedly confirms that “Ada” GPUs would have 16MB of cache per 64-bit memory bus, significantly more than Ampere at 512KB with 32-bit memory.

The top-end GPU codenamed AD102 would have 96MB of cache, AD103 and AD104 would get 64MB each, while AD106 and AD107 would be equipped with 48MB and 32MB respectively.

With all this, the RTX 40 GPUs will get up to 71% more cores than the RTX 30 GPUs, but that’s assuming that each chip had its full implementation released to the market. Naturally, there are still many things that are not clear, such as the speed and power requirements for the new cards, but it is quite obvious that the RTX 40 GPUs will be a significant improvement over Ampere.

Source: videocardz and videocardz