NVIDIA has officially showcased the new GeForce RTX 40 series graphics cards with the first two SKUs being the usual “90” RTX 4090 and “80” RTX 4080 models.
The new Ada Lovelace architecture takes on a revolutionary step in AI and deep-learning assisted rendering technology due to the fact that while the transistor count is breaking the ceiling at 76 billion, the amount of software work put out by Team Green has fully proven that the way of improving performance by AI is the way to go. Manufactured with TSMC’s 4N process technology, the Ada Lovelace-equipped cards are packed with the 3rd Generation RT Cores and 4th Generation Tensor Core for the most bonkers performance to date.
The spotlight is none other than the usual enthusiast-grade GeForce RTX 4090 that comes with 16384 CUDA cores and boost clocks of up to 2520MHz. Heck, even Jensen said 3000MHz is possible with it but who knows what it cost to do that? But anyway, that amount of CUDA cores and top frequency number means a single-precision performance of 82.6 TFLOPS. Known for the beefy VRAM, the “90s” big brother of the RTX 40 series will come with 24GB of GDDR6X memory with 1TBps of peak bandwidth. All of these will draw a default TDP of 450W and that’s 100W more than the GeForce RTX 3090 Ti. One should really not arm this card within a system with less than a 1000W PSU for real.
It’ll become available on October 12 at a “sweet” price of USD1,599 (~RM7289).
On the other hand, the GeForce RTX 4080 is not that shabby too yet the interesting thing is it will come in 2 different VRAM capacities of 16GB and 12GB with the former running the AD103 GPU with 9728 CUDA cores at 320W while the latter is down to 7680 CUDA cores at 285W. As Jensen only gave us a starting price of USD899 (~RM4100), it is safe to assume that it is for the 12GB model.
All of the announced GPUs will use PCIe 4.0 x16 and no NVLink connectors for curious people.