NVIDIA GeForce RTX 5090 PCB Leaked: Specs, Design, and Power Details

The Chiphell Forums have leaked the PCB design for NVIDIA‘s forthcoming GeForce RTX 5090 graphics card, the flagship SKU of the upcoming RTX 50 series, based on the new “Blackwell” microarchitecture. We’ve been hearing about the RTX 5090 for months, but this leak provides an unambiguous view of the AIB reference design and crowded PCB.

NVIDIA GeForce RTX 5090 PCB Leak Reveals GB202 GPU, 32GB GDDR7 Memory & Single 16-Pin Connector

The first notice is the big BGA package for next-gen GB202 GPU for RTX 5090 and perhaps follow-up designs down the line like the RTX 5090 Ti and RTX 5080 Ti. In addition, it will have a new GB202-300 GPU manufactured by TSMC with 21,760 CUDA cores and 170 SMs. Out of the box, the card should (in theory) hit clock speeds approaching 3 GHz. The PCB has 16 GDDR7 slots for memory, which has now been confirmed as 32GB of VRAM.

These GDDR7 modules are binned in the same arrangement as previously thought, achieving up to 28 Gbps speeds over the 512-bit bus interface. With this setup, the card has a whopping 1792 GB/s of bandwidth—well above the already impressive 1008 GB/s of the current RTX 4090. The PCB that will be handling all this power will have some 30 VRM phases and north of 60 capacitors. It will likely power the board with a single 16-pin connector, following the 12V-2×6 specification as up to 600W (combined TBP) is to be provided to the RTX 5090.

The PCB is marked with “PNT5090 v1. 0” and “100-10145-DFAA-A02” which will have two baseboard designs, PG144 and PG145 SKU30. One for the Founders Edition (PG144) and another for AIB models (PG145). Although shorter, the design seems to be more in line with previous leaks than what we’ve seen ahead of the launch. The RTX 5080 is due by mid-January, while the RTX 5090 could arrive by late January or early February. Expect further details to emerge during CES 2025.

FAQs