Nvidia Geforce RTX 3080 gets 1,408 extra CUDA cores and 10 GB of graphics memory

The graphics giant Nvidia’s architecture Turing entered the consumer market almost two years ago, and recently the successor Ampere was unveiled for the server market. Products for consumers will be released sometime in 2020, but from Nvidia’s point of view, there has been complete silence about the launch date, specifications and cooling solution.

On rumors, however, information has been frequently presented in the latter category, where several images have shown that the radiator solution has a different design with double fans mounted in different directions. Now it is filled with possible specifications for the three models Geforce RTX 3080, RTX 3090 and a new Titan card, data coming from Twitter user Kkatcorgi via Videocardz.

The technology site believes that specifications and naming for graphics cards are details that may be adjusted even near launch, and that the Ampere family is not yet mass-produced. The data should be taken with a pinch of salt, but can still be judged to be relatively credible because they are in line with specifications that Twitter user kopite7kimi shared at the beginning of the month. Prior to the unveiling of the Ampere-based A100, the same users presented accurate details about the Tensor cores of the circuit.

Specifications for Geforce RTX 3000 and Titan “Ampere”

If the specifications are correct, the Ti version of the Geforce RTX 3080 seems to be ironed out in favor of the RTX 3090. In the past, graphics cards with the extension “90” have been synonymous with dual graphics circuits under the hood, a meaning that with the RTX 3000 series may change . The three cards take real steps up in the number of CUDA cores compared to the Turing counterparts, and unexpectedly, the top circuit GA102 is used for the entire trio.

What the name of the next Titan card will be is not revealed in the Twitter posts, but like its predecessor Titan RTX, it is about 24 GB of graphics memory. The number of CUDA cores increases from 4,608 to 5,376, and at the same time there is talk of a slightly higher memory frequency. With the Geforce RTX 3090 as a replacement for the RTX 2080 Ti, 896 extra CUDA cores and 1 GB of extra memory are visible.

The Geforce RTX 3080 is instead said to increase by 2 GB to land on 10 GB and the effective memory frequency also takes a step up here, from 14,000 to 19,000 MHz. The RTX 3090 is said to take the whole thing one step further with an effective frequency of 21,000 MHz. In the previous report, the same frequencies can be seen, but also that the memory type GDDR6X is behind this level. To date, however, no memory manufacturer has announced circuits with that stamp.

Compared to the Geforce RTX 2080, the listed 4,352 CUDA cores mean an increase of as much as 1,408, which indicates a significant step up in performance. What it looks like on the Tensor and RT fronts is not affected for any of the cards, but even details about clock frequencies, memory buses and power consumption are shrouded in obscurity.

Read more about Ampere: