Nvidia has just released its new RTX 3000 series graphics cards today. The RTX 3080 with 10GB gddr6x will be available from 17 September, for 719 euros. On September 24, the RTX 3090 with 24GB gddr6x will follow for 1549 euros. In October, the 519 euro RTX 3070 with 8GB gddr6 will be released.
Nvidia positions the RTX 3080 as the main video card for gamers. The manufacturer claims that the new model is up to twice as fast as the RTX 2080, for the same price. Nvidia bases this on ‘average performance in graphically intense games at 4k resolution’. The manufacturer does not show concrete benchmarks. Nvidia presumably assumes ideal scenarios when using techniques such as RTX ray tracing and dlss enabled.
There will also be an RTX 3090 and it positions Nvidia as a counterpart to the Titan video cards. This top model gets 24GB gddr6x memory and costs 1,549 euros. This version is physically even larger and Nvidia claims that this card is suitable for gaming in 8k resolution.
The cheapest model in the RTX 3000 series will be the RTX 3070. That 519 euro card will have 8GB of gddr6 memory, not the faster gddr6x. According to Nvidia’s marketing slides, this card is also faster than the RTX 2080 Ti.
Nvidia’s Founders Edition editions have a new cooler, with two fans in a push-pull arrangement. The size of the cards is mainly determined by the heatsink and fans, the pcb is smaller than with the previous generation of video cards. Nvidia showed a PCB of the RTX 3080 with 18 phases. The top model takes up three slots, the other two cards have two slots. The video cards are 313, 285 and 242mm long respectively.
It is striking that Nvidia writes in its specsheets that the RTX 3090, 3080 and 3070 have 10,496, 8704 and 5888 Cuda cores respectively. That’s exactly double the amount mentioned in rumors and leaked specs. The number of Cuda cores mentioned is also much higher than with the GA100 GPUs for data centers. Nvidia may use the fp64 units of the Cuda cores with the GeForce variants as fp32 units. This may explain the doubling in the number of FP32 units. Nvidia has not yet released a whitepaper of the GPUs from the new GeForce cards.
Nvidia has the Ampere GPUs for the GeForce RTX cards made by Samsung using an 8nm process. The GPUs have RT cores for second-generation ray tracing and third-generation Tensor cores for deep learning. Nvidia claims that the cards are therefore much faster in RTX games with ray tracing.
In Nvidia’s own ray tracing demo, that difference goes up to four times in favor of Ampere versus Turing, but less in games. A slide shows that the new RT cores in games such as Battlefield V, Control, Metro Exodus, Minecraft RTX and Quake II RTX perform about 1.5x to 2x better. These are tests in 4k resolution and otherwise unknown settings. DLSS may also have been used.
The new RTX cards will have HDMI 2.1 connections and support PCI-e 4.0. The new video cards were announced in a live stream. This can be viewed via YouTube, among others. The full specifications can be seen on a comparison page from Nvidia.