An early look at the printed circuit board (PCB) design for Nvidia's next-generation GeForce RTX 4080 and RTX 4090 reveals that the forthcoming graphics boards will continue to be huge and power-hungry. A new report claims that makers of graphics cards will be able to re-use PCBs developed for the GeForce RTX 3090 Ti for their next-generation products.
Reference design of graphics cards based on Nvidia's codenamed AD102 graphics processing units can accommodate up to 12 GDDR memory chips, so the GPU will support memory interfaces of up to 384-bits, according to Igor's Lab. The information was indirectly confirmed by @kopite7kimi, who tends to have some accurate details about upcoming products. With up to 12 memory chips, Nvidia's GeForce RTX 4080/4090 graphics cards will be able to house up to 24GB of memory using 16Gb (2GB) DRAM ICs, so expect memory configurations of next-generation boards to be like those available now.
Nvidia's GeForce RTX 4080/4090 reference design for add-in-board manufacturers reportedly has a complicated multi-phase voltage regulating module that uses uPI Semi's UP9512 multiple phase buck controllers. Igor's Lab, which schematically reproduced the said PCB design, claims that the card will use a 12VHPWR (12+4-pin) auxiliary PCIe 5 power connector and consume up to 600W. For those who do not have a modern ATX 3.0-compliant PSU, Nvidia and its partners are expected to bundle 4x8-pin to 12VHPWR adapters with the boards.
While graphics cards are getting larger and more power-hungry, 600W of heat in a client system is not easy to reliably take away even using liquid cooling. Meanwhile, Nvidia's Founders Edition and reference design boards are expected to retain a triple-slot air cooler. However, some custom-designed products are projected to employ a 3.5-wide cooling system, possibly to improve reliability and/or provide some additional overclocking headroom.
Nvidia is expected to use TSMC's 4N fabrication process (a custom Nvidia-specific node) to build its next-generation Ada Lovelace GPUs. So, it is somewhat illogical for these parts to be considerably more power-hungry than Ampere GPUs produced using Samsung's outdated N8 manufacturing technology, even keeping in mind an increase in transistor count. Still, we are somewhat skeptical about a 600W thermal board power (TDP) for next-gen graphics products.
Igor's Lab further indicates that Nvidia's GA102 and AD102 GPUs are pin-to-pin compatible, so makers of graphics cards can re-use their GeForce RTX 3090 Ti PCB designs (probably with some upgrades) for their next-generation products. If the report is accurate, it may be the first time in history that high-end GPUs of completely different generations using different fabrication processes are pin-to-pin compatible and can use the same PCB designs.
With entry-level, mainstream, and notebook parts, companies tend to maintain pin-to-pin compatibility since notebook makers are not very eager to redesign their motherboards every year. However, with higher-end desktop GPUs, developers tend to rework PCB design with each generation to adjust it to s new GPU design, maximize performance, and optimize cooling.
For now, any information about Nvidia's GeForce RTX 4080/4090 graphics cards based on Ada Lovelace GPUs should be taken with a grain of salt as they are still months away from introduction. Nonetheless, some details that leak today may well be correct.