With the lifting of the evaluation ban, the NVIDIA GeForce RTX 4090 has also been squeezed out of 100 TFLOPs of computing power, making it the most powerful gaming graphics card to date. For reference, the default performance of the Founders Edition graphics card is 83 TFLOPs (17 TFLOPs lower), while the previous generation GeForce RTX 3090 Ti was only 40 TFLOPs.
In order to reach 100 TFLOPs, WCCFTech tried to unlock power and temperature limits and increased the GPU core/memory frequency by +275 / +1100 MHz respectively.
Then pressurize the GPU core by +55% through Afterburner, otherwise, it may cause a performance slump at 100% occupancy.
Then the AD102 GPU core of this RTX 4090 graphics card soared to 3150 MHz, with a maximum power consumption of 547W and a maximum temperature of 69°C, and the test was carried out under air-cooled conditions, without water cooling or liquid nitrogen cooling.
In the end we saw the magic data of nearly 101 TFLOPs, a 22% performance improvement over the default state and 2.5 times the RTX 3090 Ti.
In addition, the FP32 performance of this AD102 GPU also exceeds 50%, leaving the Hopper H100 GPU for data centers behind.
It is believed that when the Quadro professional card product line also ushered in the Ada Lovelace architecture update, we are expected to see the RTX 6000 / L60 become a popular computing / AI accelerator card in the industry.
In terms of specifications, the RTX 4090 enables 128 groups / 16384 CUDA cores in 144 groups of SM units, supplemented by 72MB L2 cache / 176 ROP raster units.
The video memory is 24GB @ 384-bit 21Gbps GDDR6X, the bandwidth is up to 1 TB/s (same as the RTX 3090 Ti), the TBP power consumption is 450W, and it is powered by a single 16-pin 12VHPWR interface (theoretical power limit is 600W).
