NVIDIA's geforce RTX 4090 graphics card is again detailed in the new rumored specification, and its speed is expected to be twice that of geforce RTX 3090. It is said that NVIDIA geforce RTX 4090 has 16128 cores, 24GB memory, 21gbps speed, 450W TDP and twice the speed of RTX 3090.
Kopite7kimi again posted the latest specifications on twitter. NVIDIA geforce RTX 4090 will also use the same ad102 GPU as the "Ti" variant, but its configuration is rumored to be reduced. The ad102 GPU will use 144 SMSs, which is 71% more than the existing ga102 GPU, and has a huge number of 18432 CUDA cores. It is said that the GPU of RTX 4090 has 16128 CUDA cores, which is 50% higher than that of RTX 3090, but 14% lower than that of the flagship configuration. This configuration may appear on RTX 4090 Ti in the future.
In addition to the number of SM, ADA Lovelace GPU will also increase the size of L2 cache. Starting from the ad102 GPU, the flagship product will be equipped with up to 96MB L2 cache, which is 16 times higher than the 6MB L2 cache of ga102.
The graphics card will also be equipped with 24GB gddr6x memory, which runs at 21gbps through 384 bit bus interface and provides more than 1.0tb/s bandwidth. The latest details also reduce the overall TDP number to 450W, which means that the RTX 4090 may not rush to high frequency like the 600W flagship model.
As we can see on geforce RTX 3090 Ti, some models exceed 450W TBP (reference value), up to 516w (customized by AIB board manufacturer). This will also be the case for graphics cards based on ADA Lovelace. Although this requires strong cooling equipment, it seems that AIBS are ready. Once the 600W TBP limit is exceeded, we may see the use of more hybrid style GPU coolers instead of traditional air cooling, which will increase by 33% compared with RTX 3090 ti. According to the latest rumors, NVIDIA is expected to launch its geforce RTX 40 series in mid July.