FinalWire announces a new version of its diagnostic and benchmarking tool AIDA64and among the notable new features is preliminary support for the NVIDIA GeForce RTX 4090, the next high-end GPU from the green company. Unfortunately, the beta update to version 6.70.6033 does not reveal any other details beyond confirming the model name and the fact that it will be based on the graphics chip. NVIDIA AD102.
This is basically to know that the first RTX 40 series GPUs would already be tested, so it’s logical to expect that in a few days, or weeks maximum, we’ll start to see performance leaks.
From what we know so far, the NVIDIA GeForce RTX 4090 will specifically use the GA102-300 graphics chip, which would hide inside nothing less than 16384 CUDA coresbeing secret for the moment the frequencies that the chip could reach, although there are rumors that it will move around the 2.75GHznothing unreasonable if you take into account that its main rival, AMD, will use the same manufacturing process as 5nm from TSMCand it is said to reach frequencies near the 3.00GHz.
The graphics chip will be accompanied by 24 GB of 21 Gbps GDDR6X memory alongside a 384-bit memory interface, while the TDP would be 450Walthough there are custom models which notoriously increase the TDP with configurations of two PCI-Express 5.0 slots of 16 pins. As a result, we will have even more oversized cooling systems, while it is rumored that the reference graphics card, the Founders Edition, will incorporate a third fan to cope with rising temperatures.
NVIDIA GeForce RTX 4080 and GeForce RTX 4070 Specifications
Along with the RTX 4090 we also know the specifications of these two models, being the RTX4060 the only one that has yet to succumb to leaks, perhaps because this model was still listed for early 2023.
The NVIDIA GeForce RTX 4080 would arrive with the AD103-300-A1 graphics chip with 10240 CUDA cores (256 less vs RTX 3090) with 16 GB of GDDR6 memory @ 21 Gbps with a TDP to be fixed at 420W.
Finally, the NVIDIA GeForce RTX 4070 would happen with 7168 CUDA cores (1535 less vs RTX 3080), but it would come with significant cuts compared to its predecessor, such as being accompanied by 10 GB of memory GDDR6 instead of 12 GB, a memory interface of 160 bit instead of 192 bits, and your TDP would be 300W.