NVIDIA GeForce RTX 4090 And 4080 PCB Details Exposed In Schematic Leak

Let's start this story with a bit of context for people who may not have been following along. NVIDIA's extant GeForce RTX 30 series cards currently top out at the RTX 3090, which can draw up to 431 W in our testing. At this year's CES back in January, Team Green announced the RTX 3090 Ti, based on the same Ampere architecture as the RTX 3090, but presumably boasting even higher performance.

The RTX 3090 Ti was expected to be out in January, and then February, yet it never showed. Now, mum's the word officially, although rumors have it launching next Tuesday. There are all kinds of arguments you can make for the delay: it would seem somewhat unnecessary in the current market, since NVIDIA doesn't need an even higher-end halo product, or maybe they couldn't get it "enough faster" compared to the RTX 3090 to warrant the "Ti" name.

Now, we're looking at the likely debut of lady Lovelace in Q3 of this year, possibly as soon as just four months away. In that case, the release of an RTX 3090 Ti could come too close to the release of its successors, potentially causing considerable consternation in those that may buy the final Ampere card.

The idea that the top-end Ada Lovelace cards will be extremely thirsty is not a new one. We heard back in November of last year that NVIDIA's next-gen would double the power consumption of Ampere, and there were rumblings of similar ideas even before that. More recently, we've heard that a supposed "RTX 4090 Ti" could even exceed the power demands of the 12VHPWR connector, which is capable of delivering 600 W on its own.

"But wait Igor," you might think. "We're talking about Lovelace, not Ampere." Indeed, according to Igor, the supposed top-end consumer Ada Lovelace part (presumably AD102) will purportedly be pin-compatible with the GA102 processor used in the RTX 3090 and presumably its "Ti" derivative.

To support his claim, Igor presents a rough schematic that he claims is for the PCB design of the RTX 3090 Ti, and thus a future Lovelace card as well. We've reproduced it above. Folks familiar with such things will take one look at this board and grimace, because we're looking at fully 24 voltage converters that Igor claims will be driven through eight power phases, with an additional four regulators for the GDDR6 memory.

We also can see in his board schematic that there are twelve pads for GDDR6X memory. GDDR6X ICs currently cap out at 2GB per package, so we're looking at a maximum possible memory capacity of 24 GB using this board design, although it seems possible to us that NVIDIA or a board partner could route additional memory onto the back of the card.

Six-hundred watts is an immense amount of heat to remove. Igor says that NVIDIA's Founders Edition card will rely on a triple-slot air cooler, but that most board partners will be using an even larger cooler that extends into the fourth slot. Liquid-cooler options will seemingly be available too, but apparently not as the primary SKUs.