NVIDIA deprecated SLI support for gaming in 2020. AMD CrossFire followed. Multi-GPU gaming is effectively dead for consumer applications. But the story isn't over — NVLink still exists for professional and compute applications. Here's the current state.
Why SLI Failed
SLI required developers to write explicit multi-GPU rendering paths. As game engines moved toward more complex, sequential rendering pipelines (deferred rendering, ray tracing, complex post-processing), the multi-GPU scaling that worked in older forward-rendering games became impossible to implement cleanly. Most games never saw meaningful benefits beyond 1.3–1.5x from two GPUs, while rendering software expected 2x. NVIDIA killed it because the effort wasn't worth the outcome.
NVLink: Alive for Professional Use
NVLink on professional-grade NVIDIA GPUs (RTX 4000 Ada, RTX 5000 Ada) allows memory pooling — two 48GB GPUs present as a single 96GB memory space. For ML training with massive models, scientific simulation, and 3D rendering of scenes larger than single-GPU VRAM, this remains valuable. But it requires professional GPUs and matching NVLink bridges.
For Gaming: Single GPU Always
One powerful GPU will always outperform two weaker GPUs for gaming in 2026. An RTX 5080 delivers better real-world gaming performance than two RTX 4070s. Save the budget for the best single card you can afford.
For AI/ML in Nigeria
Consumer NVLink (RTX 4090/5090 bridge) was discontinued. Multi-GPU ML training setups in Nigeria require professional GPU hardware. For most Nigerian ML practitioners, a single RTX 4090 or RTX 5090 is the practical answer — with cloud compute for jobs that exceed local capacity.