Last month, the specifications of the first two new-generation Nvidia graphics cards, the GeForce RTX 5090 and the GeForce RTX 5080, were released. However, it looks like Nvidia will announce a third model at the very beginning, which could be the GeForce RTX 5070 card. with or shortly after the first two models – the release dates of the reviews themselves will probably be staggered. However, the actual unveiling should probably take place during CES 2025.
The website WCCFtech brought information about the third model of the Blackwell generation, allegedly based on its own sources. According to his information, this card will be based on the third GPU in the line of Blackwell chips, which is called GB205 (RTX 5090 has GB202 chip, RTX 5080 contains GB203, GB204 chip does not exist).
This is interesting because the full GB203 chip configuration is used in the RTX 5080, and we would expect that there must be some trimmed down version to use up the factory scrap. This could be the RTX 5070 Ti or 5070 Super, but it is possible that such a model will not be released until later, or Nvidia will use scraps only in notebooks.
We don’t yet know how many computing units this card will have. According to earlier information from leaker Kopite7kimi, the GB205 is supposed to contain 50 SM blocks, which would give 6400 shaders, 50 RT cores and 200 tensor cores. There should be five GPC blocks, each containing 5 TPCs (one TPC is two SM blocks). The GEForce RTX 5070 card can have either the full 6400 shaders, or a slightly reduced number.
But what the new information does say is the memory capacity. It will be 12 GB, i.e. without any improvement compared to the previous generation (RTX 4070). It probably also means confirmation that the GPU has a 192-bit memory bus, which will probably have chips with a capacity of 16 Gb. Nvidia should use the new GDDR7 memory technology, the throughput should probably be 672GB/s, because according to WCCFtech, 28.0GHz chips will be used (that’s the speed that the RTX 5090 is supposed to use). In theory, it would be possible to get up to 768 GB/s if chips running at the effective speed of 32.0 GHz, which is supposed to be used by the RTX 5080, were used, but the RTX 5070 will not be so lucky.
Unfortunately, this graphics should have a slightly higher consumption than the previous generation. Nvidia will increase the TDP to 250W, while the RTX 3070 was a 220W card and the RTX 4070 had a TDP of 200W. This figure was confirmed by reliable leaker Kopite7kimi. However, it is not completely excluded that it will change, for example, Nvidia could still move the frequencies higher and with them also increase the consumption limits.
Revealed 1/6/2025?
That being said, this is how the card is supposed to be announced during the Blackwell graphics unveiling in January during CES 2025. Nvidia’s boss is scheduled to give an introductory presentation at the trade show today, January 6th, so Blackwell graphics could theoretically be announced on that day. However, several weeks may pass between this preliminary announcement and when the GeForce RTX 5070 goes on sale. It’s likely that the more expensive models will go on sale first and the RTX 5070 will follow with some delay – this tends to be the usual release policy.
At the same time, the GeForce RTX 5070 could be the first model of the Blackwell generation that will be financially affordable anyway. The GeForce RTX 5090 will probably be an even bigger hit than the current GeForce RTX 4090, but judging by past generations, it’s quite possible that even the RTX 5080 will cost over $1,000 (CZK 28,000 / €1,100). The GeForce RTX 5070 could be much better, the RTX 4070 cost “only” $700 when it was released, i.e. about twenty thousand (€770). This used to be the price for a high-end card back in 2016-2017, but since then Nvidia has raised the prices quite a bit.
Resources: VideoCardz, WCCFtechKopite7kimi (1, 2)
Source: www.cnews.cz
