Teardown and Circuit Board Analysis

As of this writing, we are not allowed to post teardown pictures of our GeForce RTX 5090 samples, however, the pictures you see here were taken earlier this month at the NVIDIA 2025 International CES showcase in Las Vegas. We already posted a news article with these pictures. These pictures are not of our review sample.


This is what the PCB of the GeForce RTX 5090 Founders Edition looks like. If you notice, it has no PCIe gold fingers, or display I/O, those spun off to breakaway PCBs over ribbon cables that are routed along the edges of the cards, to not impede airflow from the fans. The PCB has only one external connector, which is the 12V-2x6 power connector.

The PCB clearly looks like it's been designed by the same team that makes the SXM (scalable matrix) variants of NVIDIA's AI GPUs—you can tell from the efficient use of both sides of the PCB. Dominating the topside of the PCB is the GB202 GPU, on which the RTX 5090 is based. This is a massive package. It has sixteen Samsung GDDR7 memory chips surrounding it, making use of its lavish 512-bit wide memory bus. Almost concentric to this is the card's VRM solution, which consists of 19 phases for the VGPU, and 8 phases for the memory. The VRM solution uses high-density ferrite core chokes, and DrMOS supplied by Monolithic Power Systems (MPS).


The pre- and post-filter capacitors of the card's VRM solution are all tantalum SMDs, arranged on the reverse side of the PCB. You will notice two low insertion force FPC (flexible printed circuit) connectors, into which two crucial ribbon cables go in. The one closer to the center of the PCB is where the FPC cable from the PCI-Express 5.0 x16 riser goes in. The other one, closer to the edge of the PCB, is where the FPC cable leads to the display I/O breakout PCB, which has the card's three DisplayPort 2.1b ports and sole HDMI 2.1b port. There's also a tiny white side-facing FPC connector, which leads to the illuminated GeForce RTX logo on the card.


This is how the PCB is located in the center of the card, it is sandwiched by the cooling solution—the main vapor chamber for the hot components on the top side, and a heatsink to pull some heat from the reverse side.


The GB202 silicon is built on a TSMC 5 nm process node, specifically TSMC 4N. This is possibly the biggest gaming GPU by NVIDIA in terms of PCB footprint. It has a large central die, surrounded by a massive package substrate that's levelled by a metal reinforcement brace. The large package indicates a high pin-count. The GPU needs this for its 512-bit wide memory bus, although most pins could be power-related.