Is NVLink BETTER than SLI??

The video opens by contrasting Nvidia's traditional SLI multi GPU setup with the newer NVLink inter-GPU communication protocol. It explains that SLI has long been limited by a master/slave arrangement and relatively modest bandwidth, which constrains how much combined memory and CUDA cores can be effectively shared. The presenter then introduces NVLink as a mesh network that eliminates a single master card and enables higher total bandwidth, potentially pooling memory and compute resources across all GPUs as if they were one card. The discussion highlights that NVLink’s bi-directional mesh can deliver substantially more bandwidth than PCIe, with the promise of improved performance and the ability to handle much larger data sets. However, the video also notes practical limits: consumer RTX cards have limited NVLink links and do not support full resource pooling like professional GPUs, so the expected gaming and compute gains depend on card type and driver support. The testing section uses Quadro and GV100 configurations to compare NVLink and SLI in compute and gaming modes, showing that NVLink can scale differently across workloads, offering clear advantages in some scenarios while being more modest in others. The narrator concludes that NVLink shows real potential and could improve over time as drivers mature, but SLI is not necessarily dead yet, especially for certain high-end configurations. The video closes with an acknowledgement of real world constraints and a light plug for the channel and related products, emphasizing that NVLink’s future in consumer GPUs will depend on driver support and game/developer adoption.

Topics · graphics · hardware-benchmarks · gpu-architecture · gaming-performance · compute-performance