NVIDIA GeForce RTX 5090 GPU PCB Features Lots of GDDR7 Memory Sites & Massive BGA Package For Next-Gen Blackwell GB202 GPU
The NVIDIA GeForce RTX 5090 will be the flagship gaming graphics card within the RTX 50 "Blackwell" GPU lineup. The specifications of this beast have been discussed over the past few months but now, we have a proper leak that exposes the crowded PCB which seems to be an AIB reference design.
Starting with the details, the first thing that we can notice is the massive BGA package for the next-generation GB202 GPU. The GB202 GPU is going to be the flagship chip within the Blackwell stack & will power graphics cards such as the RTX 5090 and a possible RTX 5090 Ti / RTX 5080 Ti design in the future. Based on previous information, the RTX 5090 is going to feature the GB202-300 GPU & will include a total of 21,760 CUDA cores in 170 SMs. The graphics card is expected to clock close to 3 GHz outside the box.
As for the memory, there are a total of 16 GDDR7 sites which confirm up to 32 GB of VRAM. These GDDR7 modules are arranged exactly as the past leaks had hinted and will feature speeds of up to 28 Gbps across a 512-bit bus interface. This would offer a total bandwidth of 1792 GB/s, a massive uplift over the 1008 GB/s bandwidth of the existing RTX 4090 GPU. The GPU and memory should feature around 30 VRM phases and over 60 capacitors on the main PCB.
That's a lot of power for a compact PCB of this size, and it looks like power is delivered by a single 16-pin connector based on the newest 12V-2x6 standard. A single 16-pin connector can deliver up to 600W of power, which is the maximum TBP of the RTX 5090 GPU.
Further zooming in on the PCB reveals that the board features the model number "PNT5090 v1.0" followed by "100-10145-DFAA-A02". The RTX 5090 will feature two baseboard designs, the PG144 and PG145 SKU30. The former is a reference design for the Founders Edition card while the second is the reference design for the AIB models. The PCB is indeed much shorter, but it looks like NVIDIA is following a more standard design approach versus what had been hinted at before.
The NVIDIA GeForce RTX 5090 will be the flagship GPU and will launch either by the end of January or early February, while the RTX 5080 is currently scheduled to hit shelves by mid-January. Expect more information on the next-gen lineup at CES 2025, which is a few weeks away.