Microsoft is starting to let GPUs replace NPUs in some AI tasks.
Microsoft is set to expand that GeForce RTX card-based machines can use the Local AI feature on Windows, much like machines that have passed the Copilot + PC standard. Previously, it was limited to machines with NPUs of a certain strength.
.
This change marks Microsoft's new direction in pushing AI on Windows, because from the time many AI features were originally tied to the Copilot + PC concept, which required 40 TOPS or more NPUs. Now the company is starting to open the way for GPUs, especially NVIDIA GeForce RTX, to take on more AI processing roles on board.
.
Machines that use 30 Series or more NVIDIA GeForce RTX cards and have at least 6GB of VRAM will be able to use the power of the GPU to run some Local AI tasks on Windows without always needing a Copilot + PC specification NPU.
.
Simply put, people with gaming PCs, desktops, or creators line notebooks already equipped with RTX cards may have the opportunity to use more AI capabilities on Windows in the future without switching to a new generation of Copilot + PCs alone.
.
This does not mean that all Copilot + PC features are unlocked for RTX to use immediately. Some system-level features, such as Recall, Click to Do, or some Windows Studio Effects, may also require a Copilot + PC.
.
What Microsoft is opening up to now is the Windows AI APIs and Windows ML side that allows developers to create AI apps that run on a machine, opting for the right hardware - CPU, GPU or NPU - instead of being limited to NPU alone.
.
Examples of tasks that may benefit from running through the GPU include text summarization, text rewriting, language translation, running local language models, data analysis tasks, including AI features that require more processing power or memory than a typical NPU.
.
For NVIDIA, this change is considered good news, because it makes the GeForce RTX screen card not only seen as hardware for gaming and graphics work, but more as an integral part of Windows-based AI systems, especially as the RTX GPU already has Tensor Cores and a healthy AI ecosystem.
.
At the same time, NPU has not lost its importance because it is also suitable for light AI tasks that require continuous and energy-efficient tasks, such as camera image adjustment, noise reduction, or behind-the-scenes features. The GPU is better suited to heavier AI tasks, such as language models, visualizations, many document summaries, or AI apps that require a lot of VRAM instead.
.
Source: techspot
