NVIDIA GeForce GPUs Do Not Incorporate Peer-To-Peer Support, Redirecting Consumers To Purchase Expensive GPUs

wccftech.com:

Puget Systems, one of the leading custom PC builder companies, was conducting a test on AMD and Intel PC server CPU-based systems that utilized NVIDIA GeForce GPUs from the last generation and the current generation and included the enterprise NVIDIA RTX 6000 Ada generation graphics card.

During the tests, the information revealed that the GeForce RTX 4090 graphics card had a faulty Peer-To-Peer (P2P) functionality. In turn, the company contacted NVIDIA, who explained that the new graphics card generation no longer offers P2P support.

When Puget Systems attempted to process P2P-based workloads in its benchmarking tests, it appeared "corrupted" or would fail upon access. The testing was sparked because a user on the NVIDIA forums had P2P failures on their system using dual GeForce RTX 4090 graphics cards. It was not until several months had passed that an NVIDIA representative responded to the user.

Hi all. Apologies for the delay. Feedback from Engineering is that Peer to Peer is not supported on 4090. The applications/driver should not report this configuration as peer to peer capable. The reporting is being fixed and future drivers will report the following instead.

— NVIDIA representative

P2P allowed for data to be able to transfer from one NVIDIA-supported graphics card's memory to another NVIDIA GPU. The functionality allows the system memory to be focused on other tasks as the P2P process uses the graphics cards' integrated memory, accelerating access and transmissions to the memory. CUDA-based programs benefit from this feature, especially those utilizing NVLink.

NVLink is a high-speed GPU interconnect that offers a "significantly faster alternative for multi-GPU systems than PCIe-based connections." P2P utilizes both