NVIDIA RTX 4090 (AD102) FP32 performance up to 100 TFLOPS

by ·

While rumors of AMD’s RDNA3 flagship processor suggest it will offer 4x more single-precision computing power than RDNA2, NVIDIA’s next-gen flagship GPU has also made a considerable leap in performance. The AD102 GPU based on the Ada Lovelace architecture is expected to provide more than 100 TFLOPS performance with FP32, which is 2.5 times the 40 TFLOPS of the RTX 3090 Ti and 2.8 times that of the RTX 3090. However, FP32 (single precision) performance is not guaranteed to be an improvement over gaming performance.

In addition to the improvement of FP32 performance, NVIDIA’s next-generation RTX 40 series will also add more optimizations to current technologies, such as ray tracing, DLDSR, etc., or add new functional technologies.
NVIDIA RTX 4090 FP32 performance

To achieve 100 TFLOPS of performance, the AD102 GPU with 18432 CUDA cores must be clocked at 2.7 GHz, but the cores on the RTX 4090 will almost certainly be partially disabled. So the clock speed will be relatively higher. It is rumored that the clock of the next-generation flagship card AMD and NVIDIA may be very similar. In the case of the AMD Navi 31 GPU, reaching the rumored 92 TFLOPS means a GPU speed close to 3.0GHz is required.

There have been many specs rumored about the next-generation graphics card, but in fact, the specifications may change before the release, but the only certainty is that the next-generation flagship graphics card will be far more power-hungry than the current one.

Via: videocardz