Many people found that NVIDIA’s GeForce RTX 4090 based on the Ada Lovelace architecture no longer supports NVLink bridges like the previous generation GeForce RTX 3090. The NVIDIA RTX 6000 Ada Generation, which was subsequently released by NVIDIA, also did not provide support.
In the end, Mr. Jensen Huang, the founder, and CEO of NVIDIA
confirmed in the communication with the media that the Ada Lovelace architecture GPU will not support NVLink, ending any possibility of multiple GPUs, marking the complete demise of SLI through dedicated physical interfaces.
Although Jensen Huang’s words seem to be the final word,
Der8auer, a well-known overclocking enthusiast from Germany,
found that the PCB of the GIGABYTE AORUS series GeForce RTX 4090 graphics card is still reserved for NVLink-related circuit design.
Gigabyte RTX4090 Aorus Master PCB, Source: TechPowerUP
Judging from the current situation, it seems that only Gigabyte’s GeForce RTX 4090 graphics card will do this, and similar products from other manufacturers have not reserved NVLink-related designs. It is not clear whether it is the shielding that Nvidia made on the GeForce RTX 4090 at the last moment or a design problem with Gigabyte. Some people even speculate that Nvidia may have retained the relevant design and will only turn it on when necessary.
Although the multi-GPU rendering at this stage has limited improvement in efficiency for games, and the construction cost and power supply of dual GPUs are very big problems, it is still effective for application scenarios. According to Jensen Huang, support for the NVLink bridge was canceled because the I/O would be used for “other purposes”, so it was decided not to waste resources and leave the NVLink interface.