RTX 5090 Founders Edition is equipped with 16 GDDR7 memory

This year, NVIDIA is set to release its next generation of gaming graphics cards, the GeForce RTX 50 series based on the Blackwell architecture, anticipated to arrive as early as the fourth quarter. Like the previous GeForce RTX 30/40 series, the initial focus will be on high-end products. While the appearance of the RTX 5080 is uncertain, the RTX 5090 is virtually assured.

Recently, on Chiphell, a user disclosed some details about the GeForce RTX 5090 Founders Edition, indicating it will feature a triple-PCB design, allowing for dual-sided cooling. The accompanying GB202 memory interface is 512-bit, requiring 16 GDDR7 memory chips, with a memory capacity reaching 32GB.

The multi-PCB design is not new to NVIDIA. For instance, the currently available Hopper architecture H100 uses this design, and even more extreme is the GeForce 9800GX2, which had two independent boards, each with a GPU chip. According to VideoCardz’s summary, with the increase in memory configuration, the layout of memory chips will change from the RTX 4090’s 3-4-1-4 to 4-5-2-5 (clockwise direction), accompanied by a schematic diagram.

As for the structure of the triple-PCB design, it remains unclear. Considering the leaked four-slot thickness cooling module design of the unreleased RTX TITAN ADA, some may speculate about the PCB with the GPU chip being parallel to the motherboard. According to subsequent user comments, the three PCBs are divided into the main board, a board housing the display controller, power supply system, and other components that do not need direct contact with the cooling assembly, and the PCIe component.

Rumors suggest the GB202 will have 192 SM units (24,576 CUDA cores), though the GeForce RTX 5090 will not utilize all SM units, expected to employ around 170 to 180 of them.