The GeForce RTX 5090 will be NVIDIA’s next top-of-the-line graphics card. It will only be launched in end of 2024, according to the latest information, although there are some sources that assure that NVIDIA is in no hurry with the launch and that it could even enter the market in 2025. Everything will depend on the market situation and what AMD does.

Its specs were a real mystery, albeit a purely logical one, and when we saw the path NVIDIA took from the GeForce RTX 20 to the GeForce RTX 40, I was clear about the three keys that I decided to share with you at that moment and these are certainly very important, and we may consider them almost confirmed.

Now, and with the latest rumors coming out, I can now share with you list of possible specifications Which seems pretty plausible to me. Keep in mind that since this is early information and not confirmed, it’s possible that some things won’t come true, but it reminds me a lot of the first rumors that surfaced about the GeForce RTX 40, which eventually died down becomes a reality.

Possible specifications of GeForce RTX 5090

• Graphics core GB202, Blackwell architecture, produced in a 3nm node.
• 24,576 shaders at 2.9 GHz-3 GHz, normal and turbo mode.
• 192 SM units.
• 192 fourth generation RT cores.
• 768 texturing units.
• 256 raster units.
• 768 fifth generation tensor cores.
• Fourth generation optical flow accelerator.
• 128 MB L2 cache.
• 384-bit bus. Some rumors talk about a 512-bit bus.
• 24 GB if the bus is 384 bits or 32 GB if the bus is 512 bits.
• GDDR7 memory at 32 GHz.
• TGP: 450 watts or more.

These specs were calculated based on my knowledge of NVIDIA GPU architecture and design, so you won’t find them as extensive anywhere else. Personally, I think it’s probably NVIDIA because of cost and performance bet on a 384-bit busespecially considering that switching to GDDR7 memory will achieve an enormous increase in bandwidth.

New architecture improve shader performance, and fourth generation RT cores should also have a Increased ability to calculate gross intersections and collisions. To this we must add the possible technical improvements and optimizations introduced by NVIDIA, as well as the new use of AI by both the fourth generation tensor cores and the Optical Flow Accelerator.

I daresay we will probably see a a new technology associated with DLSS that will be exclusive to this new graphics generation, and that NVIDIA could use the GeForce RTX 5090 presentation to show it. This technology will increase the DLSS version number to 4.X.

As I mentioned at the beginning of the article, the introduction and release of the GeForce RTX 5090 will take place between the end of 2024 and the middle of 2025. Personally, I am inclined to think that we will most likely see it between the months October and December 2024because that was the initial window that NVIDIA was tracking, and that represents a steady two-year refresh cycle.

In terms of price, it is almost certain that the use of a 3nm node and the huge complexity that the GB202 will have, in addition to the use of GDDR7 memories, will ultimately affect production costs GeForce RTX 5090, so it is quite likely that its retail price will eventually be a little bit older than the one the GeForce RTX 4090 had at launch.