Several network sources reported the alleged characteristics of future NVIDIA GeForce RTX gaming cards of the 3000 series, as well as the new TITAN model, built on the basis of Ampere architecture.
Frankly speaking, all this is rumored, since only NVIDIA CEO Jen-Hsun Huang can confirm or deny the information below, when he officially announces the new generation of graphics accelerators. According to the latest data, the production of new consumer cards based on Ampere will begin no earlier than August, and when their presentation will take place is now not clear.
Nevertheless, it should be noted that the Twitter user under the pseudonym KatCorgi serves as the source of the data below. In the months before the official announcement, he “predicted” the correct number of tensor cores of the NVIDIA A100 graphics accelerator.
2nd Gen NVIDIA TITAN
GA102-400-A1 5376 24GB 17Gbps
GeForce RTX 3090
GA102-300-A1 5248 12GB 21Gbps
GeForce RTX 3080
GA102-200-Kx-A1 4352 10GB 19Gbps
– KatCorgi (@KkatCorgi) June 19, 2020
Interestingly, according to KatCorgi, in the hierarchy of the new series, the GeForce RTX 3090 will take the place where the RTX 3080 Ti could be expected. If this is true, then it is completely unclear why NVIDIA should change the name of the older model, especially when you consider that the NVIDIA x90 number suffix is usually received by video cards using two GPUs. In addition, in its data, the source does not name the new TITAN RTX as a video card built on the basis of Ampere architecture. Instead, the model is designated as “TITAN 2nd Generation”. If the name for it has not yet been confirmed, then there is no reason to believe that the GeForce RTX 3090 is the name of the flagship of the new series.
Although NVIDIA is known for its habits of changing everything at the very last moment. The company also changed the names of its products in fact right before their presentation. So, for example, it was with the GeForce GTX 880 graphics cards. Two weeks before their announcement, their name was changed to the GeForce GTX 980.
As for the characteristics, according to published data, the new generation TITAN RTX graphics processor (GA102-400-A1) will receive 5376 CUDA cores and 24 GB of video memory. The source does not indicate which memory standard this video card will use – GDDR6 or GDDR6X. On the other hand, so far none of the memory chip manufacturers has yet announced the GDDR6X. In turn, the production of GDDR6 memory became known long before the announcement of TITAN RTX based on Turing.
The video card GeForce RTX 3090 (GA102-300-A1) is attributed to 5248 CUDA cores and 12 GB of video memory with a speed level of 21 Gbit / s. With this indicator, its throughput could theoretically be 1 TB / s.
For the GeForce RTX 3080 (GA102-200-Kx-A1), which will replace the current GeForce RTX 2080 (Super), an increase in video memory is expected. The new product, according to the source, it will be 10 GB with a speed of 19 Gbit / s. If this information is correct, then the memory can receive a bus of 320 bits, and its throughput will be 760 GB / s. But the most interesting detail here is that the new product can be built on the basis of a “large GPU”, rather than a “stripped down” one. Similar to the NVIDIA cards x80 models have not been observed since the time of Kepler. The last such graphics card was the GeForce GTX 780 Ti.
It is also difficult to believe that the TITAN RTX (Ampere) will receive such slower video memory compared to the RTX 3090. On the other hand, in order to fit the attributable volume of 24 GB, the memory will probably have to be placed on both sides of the board, which will really reduce its speed.
If you notice an error, select it with the mouse and press CTRL + ENTER.