Windows 10 will play well with Intel’s new VPUs for offline AI processing
Microsoft’s new machine learning platform will encourage adoption of AI and Intel’s VPUs will ensure the PC doesn’t slow down in the process
Microsoft has decided to use Intel’s new Myriad X vision processor units (VPU) for handling all artificial intelligence (AI)-related processing on Windows 10 devices. The announcement was made at the Windows Developer Day conference, where Microsoft announced its new machine learning (ML) platform for the next Windows 10 update. The new platform will allow AI-based applications to offload most of the workload from cloud to the Windows 10 device.
Its software development kit (SDK) comes with pre-trained machine learning tools which will help developers build AI solutions that will work within the device with minimal involvement of cloud. Most existing AI tools require high bandwidth as a lot of data processing happens on the cloud. Microsoft claims the ML platform will eliminate costs associated with bandwidth used on AI processing and will lead big savings for people working on AI tools on the PC.
The option to run AI tasks on the device itself and reduce reliance on the cloud will reduce the cost of running them and can encourage developers to incorporate more AI-related tools such as personal assistants, search tools and advanced biometric features on Windows 10 PCs. They will also run faster as most of the action will take place on the device now.
The platform can determine how much power an AI task requires and will allocate to the hardware that is better equipped at handling them. Light AI tasks will be handled by CPU, while the more complex will be assigned to the VPU.
Processing this heavy data set on a Windows PC will put a lot of stress on the device’s central processing unit (CPU) and graphics processing unit (GPU) and will slow down other apps and processes. This is where Intel’s new VPUs come into the picture.
Developed by Intel’s subsidiary company Movidus, Myriad X VPU comes with a dedicated hardware accelerator called neural compute engine for fast deep neural network inferences. Based on a 16nm architecture, the new VPU offers 16 high-performance cores optimised for deep learning processes and computer vision systems. Movidus claims the new VPU can carry out 1 trillion operations per second.
Intel’s VPUs are not the only option available when it comes to handling AI workload on PCs. Nvidia’s new Titan V GPU for PCs is powerful enough to handle deep learning applications on the device itself. Titan V GPU for desktop is available at $2,999 (approximately Rs1,95,182). There is no update on the price and the devices which will be powered by the Movidus X VPU.