The GeForce RTX 5070 will be positioned as the graphics card high range input within the next generation of NVIDIA. Based on this information alone, it is easy to understand why it will be one of the most important models and will attract the most interest from consumers, but unfortunately the information we have about it is still very scarce.

In general, most of the leaks and rumors that have appeared so far have focused on general information about the Blackwell architecture, as well as the GeForce RTX 5080 and GeForce RTX 5090, which will be the first graphics cards to be launched in this new generation. by NVIDIA. Despite everything, I managed to collect enough information to share with you. an estimate of what we can expect from the GeForce RTX 5070.

GeForce RTX 5070 specifications

As I told you before, we still don’t have official information about its possible specifications, but I have outlined a GB204 core estimate that allows me to leave you preview of a possible configuration What this new graphics card could have:

• Graphics core GB204 made on TSMC’s 4nm node.
• 64 SM units.
• 8,192 shaders at speeds between 2.8 GHz and 3.2 GHz.
• 256 texturing units.
• 96 raster units.
• 64 cores of the fourth generation RT.
• 256 tensor kernels of the fifth generation.
• 192-bit bus.
• 12 GB or 16 GB of GDDR7 memory at 28 Gb/s (672 GB/s bandwidth).
• 48 MB L2 cache.

In a quick comparison with the GeForce RTX 4070, we see a significant increase in the number of shades which would increase from 5,888 to 8,192and also a substantial increase in the number of texture and raster units, tensor cores and RT cores. The 192-bit bus would be retained and the memory configuration could be 12GB or 16GB (GDDR7 in both cases).

Ideally, NVIDIA would choose to configure 16 GB of graphics memory, as it would give the GeForce RTX 5070 a much more solid value and make it a much more interesting option. Regarding TGP, I am inclined to believe that it will be preserved between 225 and 265 watts.

Architecture, design and benchmark performance

It will most likely use a GeForce RTX 5070 blackwell architecture, as there is no indication that NVIDIA has a different architecture lined up for its mainstream consumer graphics cards. It will be manufactured at TSMC’s 4nm node and will be assembled core GB204. It’s important to note that the configuration we set up can be trimmed to make room for the GeForce RTX 5070 SUPER.

The GeForce RTX 5070 will have a Founders Edition version that will retain the minimalistic and elegant aesthetic we have already seen with the GeForce RTX 4070. Its GPU will have monolithic core design, which means that all GPU elements will be integrated into a single silicon chip and will include fourth-generation RT cores, fifth-generation tensor cores, next-generation shaders with all Blackwell enhancements, and a new optical accelerator.

The L2 cache will again have a big role in the GeForce RTX 5070, which will have 48 MB of L2. This allows you to reduce the dependency on the memory subsystem and Searching for cached data will increase traffic. With this amount of cache, the optimal resolution level will be 1440p, although we can use it for gaming in 4K without any problems.

In terms of performance, the GeForce RTX 5070 will have maximum peak performance 52.42 TFLOPs in FP32, a figure that would practically place it on par with the GeForce RTX 4080 SUPER. It should be taken into account that the first will have all the advantages of the new NVIDIA architecture and that it could also come with the support of exclusive technologies.

Release date and possible price

The launch of the GeForce RTX 5070 will take place in first quarter of 2025, i.e. between the months of January, February and March. It will arrive after the GeForce RTX 5080 and GeForce RTX 5090 and will be available before the GeForce RTX 5060 Ti and GeForce RTX 5060.

As for the selling price, it would be most logical to place it between 549 and 599 dollarsnumbers that could happen in Spain after currency conversion and tax application 599 or 649 euros. Due to the cost increase that switching nodes and using GDDR7 memory will bring, we may see an even higher selling price, so keep that in mind.

AI generated images.