NVIDIA, support for P2P functionality disappears from GeForce

An employee of Nvidiaa few months late, replied to a thread opened on the company’s forum by a user who had encountered a problem with his GeForce RTX 4090: the test P2P (Peer to Peer) failed. Effectively, NVIDIA has confirmed that P2P technology is not supported by the card in question.

“Hi everyone. Sorry if I’m late. Asking the engineering team, I can confirm that Peer to Peer is not supported by the 4090. Applications/drivers should not report this configuration as Peer to Peer enabled. The problem is being fixed and future drivers will return the following result” reads the response where the message obtained is then indicated.

This news comes following the recent tests carried out by Puget Systems on several NVIDIA and AMD systems, which revealed a partial malfunction of P2P on the GeForce. Essentially, workloads done with P2P fail or become corrupted.

For the uninitiated, technology Peer to peer it’s used in the professional sector for the large CUDA-based workloads and allows you to speed up communication between two video cards by bypassing system memory. This is a useful feature in the context of content creation, artificial intelligence or scientific calculations.

The system works both through PCIe is NVLink of NVIDIA, in the latter case with the advantage of a higher bandwidth. The technology on video cards from the house of Santa Clara was introduced with the series GeForce GTX 1000but it seems that the RTX 2000 were the last to fully support it.

The reason is actually quite simple. With the RTX 3090 some users noticed that the P2P functions were enabled, so many workstation and professional machine vendors began to offer configurations equipped with the aforementioned card so as to guarantee lower costs. RTX 3090 cards began to disappear from the market, while sales of RTX A-series (former Framework).

In essence, NVIDIA has managed to separate the mainstream sector even more clearly from the professional one, so as to balance sales. However, there is a rather curious aspect to the choice of NVIDIA. The professional counterpart of the GeForce RTX 4090 is the RTX A6000 which uses the same AD102 GPU. Naturally, part of the configuration changes starting from the amount of VRAM memory to the clear advantage of the Pro proposal.

However, while installing the same GPU and offering comparable performance, only the RTX A6000 is able to take advantage of P2Pat the modest sum of $6,800 against the 1599 dollars (list) of the RTX 4090. It is clear that it is a commercial choice, since, although different, just like in the past, mainstream solutions exploit similar hardware, so similar are the performances. What changes, and also in a rather incisive way, is more the price of the two proposals.