Its launch is still around a few years away, but thanks to an alleged leak, we already have the first unofficial details of the GeForce RTX 50 and Radeon RX 8000provisional names under which we know the future graphics generations of NVIDIA and AMD.

We start with the Radeon RX 8000. This new generation of graphics cards will use the RDNA4 architecture and in theory should make the leap to TSMC 3nm node. AMD has plenty of room to make this leap, as it is theoretically expected to hit the market in late 2020, by which time the 3nm node for complex, high-performance chips should be fully mature.

The Radeon RX 8000 will theoretically offer an important thing improved ray tracing performance, Although we don’t have specific details at the moment. We will find one more important news in the cache, an odd link because it’s speaking in a general way and it’s assumed that AMD won’t be limited to L3, also known as infinite cache. This may indicate a possible increase in the L2 cache.

To further improve performance, AMD will use a new data preloading system, will introduce the Wave Matrix Multiply-Accumulate (WMMA) of the second generationwhich could double the performance per computing unit, provided they are used correctly, and other modifications are also expected to improve the ratio of performance to efficiency, in addition to the use of next-generation ray tracing cores and new specialized units for artificial intelligence.

As for the GeForce RTX 50, we can also expect a 2024 release and a jump to the 3nm node. Rumor has it that this new generation it will have a modular design, this means it will be configured with multiple chips following the approach we saw in the Radeon RX 7000, although in theory it would be limited to the high-end. I remind you that there were rumors about a jump to a chiplet with a GeForce RTX 40, and that ultimately did not come true. Personally, I think NVIDIA will choose to stick with the monolithic core design for as long as possible, and I doubt the supposed jump to an MCM design this generation.

The GeForce RTX 50 will be based on the Blackwell architecture, and the source assures that NVIDIA is already testing various “monsters” based on the MCM design. This architecture will use completely redesigned SM units and with the new structure, it will rely on Hyperspeed interconnects to connect to the chips and will integrate a noise reduction accelerator in the pipeline of RT coresfunction for which tensor kernels were used.

We don’t have more details about these two architectures, but in the coming months new rumors and leaks will reveal more information that we will share with you. Fountain highlights that NVIDIA will provide the biggest performance leap in its history with Blackwell, a statement that was also made with Ada Lovelace, and which ultimately did not come true, because although it represents an important leap, it is not the biggest that NVIDIA has made in its entire history.