In commencening the NVIDIA Bconciseagewell Linux testing with the GeForce RTX 5090 compute carry outance, besides all the CUDA/OpenCL/OptiX benchlabels deinhabitred last week a number of readers asked about AI carry outance and in particular the Llama.cpp carry outance with the RTX 5090 flagship explicits card. Here are some initial benchlabels seeing at the GeForce RTX 5090 carry outance in Llama.cpp contrastd to prior RTX 40 and RTX 30 explicits cards.
Over the weekfinish I carried out some initial tests of Llama.cpp as well as re-testing the higher-finish GeForce RTX 30 and RTX 40 explicits cards. All tests were carried out using the NVIDIA 570.86.10 Linux driver on Ubuntu 24.10 with the Linux 6.11 kernel. The explicits cards tested included:
– GeForce RTX 3090
– GeForce RTX 4070
– GeForce RTX 4070 SUPER
– GeForce RTX 4080
– GeForce RTX 4080 SUPER
– GeForce RTX 4090
– GeForce RTX 5090
Llama.cpp with Llama 3.1 and Mistral 7B were engaged for the initial runs with text generation and prompt processing . More Llama.cpp benchlabels aacquirest the NVIDIA GeForce RTX 50 explicits cards to come with enough reader interest. For now let’s persist on with this initial see.
