AMD RDNA 3 uses chiplet designs with dual graphics circuits

When AMD unveiled the Zen 2 architecture and processors in the Ryzen 3000 series, the company also presented a different solution to the previous generation and competitor Intel, namely split-silicon (Multi-Chip Module, MCM). Through several circuits, for example, it is possible to use different manufacturing techniques and create a more flexible platform.

Contents hide

1 Read the latest news about graphics cards:

2 Related posts:

Intel is also investing in MCM solutions in data center-oriented Xe graphics devices, and with the “Hopper” architecture, it is rumored that Nvidia is embarking on the same path. At the beginning of the year, it emerged via a patent that AMD also has divided graphics circuits on the menu in the future. In the same vein, Twitter user Kepler_L2 announced that the previously mentioned circuit Navi 31 is a so-called chiplet, which acts as an element in future graphics cards.

However, the post flew under the radar of the media, but Videocardz is now paying attention to this in connection with more AMD-signed patents in the same district appearing on the web. As for the Navi 31, it first appeared in the source code for Mac OS “Big Sur” in June, but since then the lid has been on. Kepler_L2 believes that the circuit, like the Navi 21 from the Radeon RX 6000 family, has 80 Compute Units (CU), which translates to 5,120 full-scale stream processors.

According to the user, the top model with Navi 31 must be equipped with double chiplets, for a total of 10,240 stream processors – double up from the Radeon RX 6900 XT. The future will tell if this really happens, but it may take more than a year before AMD reveals any official information. According to the product plans from last year, RDNA 3 will appear sometime in 2022, with promises of a transition to more advanced manufacturing than TSMC’s 7 nanometers.

Read This Now: Youtube profile tests games in grainy 72p

With the fresh and latest generation AMD graphics card, the company did ray tracingdebut and the new patent mentioned above is about the next generation. These are methods for synchronizing a load between a separate processor that handles ray tracing, with graphics hardware distributed over several circuits. In a serial on Twitter gathers user Underfox links to patents.