Before its official release, NVIDIA’s GeForce RTX 4090 graphics card’s first unofficial benchmark was discovered in the Geekbench 5 database. While NVIDIA has been supplying performance data for DLSS 3 and Ray Tracing to gamers, there hasn’t been much information available about the rasterization performance of their next-gen components. Now, the Geekbench 5 CUDA database has the first non-DLSS and non-RT benchmark leak.
The NVIDIA GeForce RTX 4090 was evaluated on an AMD Ryzen 9 7950X CPU platform with 32 GB of DDR5-6000 DRAM without knowing whether it was a reference Founders Edition or a special AIB design. The maximum recorded frequency for the graphics card was 2.58 GHz. The graphics card received a CUDA performance score of 417713. In the same benchmark, the RTX 3090 Ti earns a total of 260,346 points, while the RTX 3090 scores a total of 238,123 points. This represents an improvement of 60% and 75% over those graphics cards, respectively.
A total of 16,384 CUDA cores and 128 of the 144 SMs will be used by the NVIDIA GeForce RTX 4090.
The GPU will have 384 ROPs and 96 MB of L2 cache, which is ridiculous, but since the RTX 4090 is a scaled-down version, it might have slightly lower L2 and ROP counts. Although the clock speeds have not been verified, taking into account that the TSMC 4N technology is being used
With regard to memory specifications, the GeForce RTX 4090 will offer 24 GB of GDDR6X memory that is clocked at 21 Gbps across a 384-bit bus interface. This will offer bandwidth of up to 1 TB/s. The TBP is rated for 450W of power consumption, which is the same bandwidth as the current RTX 3090 Ti graphics card. A single 16-pin connector will supply the card with up to 600W of power. Higher TBP targets will be provided by custom models.
On October 12, the public will be able to purchase bespoke cards made by partners of NVIDIA and the GeForce RTX 4090 GPU. Despite the availability of several GPUs, Proshop does not accept preorders.
Also Read:
