Nvidia RTX 5090 prototype surfaces with four 16-pin power connectors and additional VRMs

An alleged prototype of Nvidia’s RTX 5090 has surfaced, featuring four 16-pin power connectors, courtesy Twitter/X user @yuuki_ans. Presumably this particular design is an early engineering sample which was used by Nvidia or one of its board partners for testing purposes. Although the unit pictured has been destroyed by cutting it in half, it does give us an insight in to what Nvidia was working with prior to the final design.

Having four power connectors could be for various reasons. A simple one could be just to provide a fail-safe. In the event of one of the connectors failing, which has been a common sight, then others could be used to continue delivering power to the GPU. Another possible use case could be to distribute power evenly in order to avoid overheating.

Theoretically, four 16-pin connectors are capable of drawing power up to 2,400W, which is roughly the same as running two space heaters at the same time. That is a lot of power for a single component, which is probably why we can see two identical rows of VRMs positioned on the right side of the GPU die mount. A single GPU drawing 2,400W is far beyond any mainstream gaming PC, instead, this class of hardware would probably target niche markets like AI (Artificial Intelligence), HPC (High-Performance Computing), or CG and VFX simulation.

There are additional connectors around the edges of the PCB including fan and USB headers as well as some diagnostic pins. It also has five video out ports, instead of the usual four, which could again be for some specific pre-production testing. If we look at the memory modules it is similar to the RTX 5090, however, we don’t know for sure if there is more memory soldered at the back of the board. In that case, it could be an early production unit for something even more powerful like a 5090 Ti or even the RTX PRO 6000 Blackwell series.

Previous prototype RTX 5090 leaks we have spotted

Back in January, we reported of another RTX 5090 prototype leak featuring dual 16-pin (12V-2×6) power connectors. It offered support for a reported TDP (Thermal Design Power) of 800W, which is about 39% higher than the RTX 5090's 575W. The prototype also included the GB202 GPU but with a higher 24,576 CUDA core count than the RTX 5090's 21,760, as well as higher clock speeds, and 32GB of GDDR7 memory on a 512-bit interface.

Similarly, in a behind-the-scenes video shared by Nvidia, the company briefly showcased an early prototype of the RTX 5090 Founders Edition. It featured a giant quad-slot design with a vertical PCB and three cooling fans. The ‘three-thirds flow-through’ cooling architecture made the GPU so big that if Nvidia went ahead, it would have been the largest Founders Edition graphics card ever. Thankfully, Nvidia chose not to bring that concept to the market, although that hasn’t stopped board partners from making beefy RTX 5090 models for its consumers.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.