GNA: Smart PCs with AI accelerator.

Intel has been betting on built-in AI processing capability in its processors to meet the immense demand in this industry. With this, the new equipment offers more performance dedicated to neural network processing.

11th Generation Intel® Core™ processors with the GNA feature are energy efficient because they are designed to process and accelerate AI processing while freeing precious CPU resources.

What is Intel GNA?

The Term GNA is an acronym for Gaussian & Neural Accelerator Library attributed to the coprocessor for Neural Acceleration integrated in some processor models to speed up the execution of inference algorithms. GNA is very similar to the math coprocessor of 386 processors from the 90's. In other words, it works through the intel API (library libgda). Its implementation in deep learning models can happen during training, importing models from conventional frameworks (tensor flow, torch and others) or using the Intel Deep Learning SDK Inference Engine (Open VINO).

Equipment with Intel® GNA

Below is the list of processor supporting Intel® GNA:

• Intel® Speech Enabling Developer Kit 
• Amazon Alexa* Premium Far-Field Developer Kit 
• Intel® Pentium® Silver Processors N5xxx, J5xxx and Intel® Celeron® Processors N4xxx, J4xxx:
• Intel® Pentium® Silver J5005 Processor 
• Intel® Pentium® Silver N5000 Processor 
• Intel® Celeron® J4005 Processor 
• Intel® Celeron® J4105 Processor 
• Intel® Celeron® ProcessorN4000 
• Intel® Celeron® ProcessorIntel® 
• Core™ Processors (formerly codenamed Cannon Lake): Intel® Core™ i3-8121U 
• 10th Generation Processor Intel® Core™ Processors (formerly codenamed Ice Lake):
• Intel® Core™ i7-1065G7Processor 
• Intel® Core™ i7-1060G7 Processor 
• Intel® Core™ i5-1035G4 Processor 
• Intel® Core™ i5-1035G7 Processor 
• Intel® Core™ i5-1035G1 Processor 
• Intel® Core™ i5-1030G7 Processor 
• Intel® Core™ i5-1030G4 Processor 
• Intel® Core ™ i3-1005G1 
• Intel® Core™ Processor i3-1000G1 
• Intel® Core™ Processor i3-1000G4 
• All 11th Generation Intel® Core™ Processors (formerly codenamed Tiger Lake). 

When using GNA, the hardware allocates the inference model in memory to later run it parallel to the processor (thus freeing it for other tasks). Another important point is its energy efficiency, which allows its presence even in ultra thin notebooks. Some products without x86 processors, such as Alexa Premium Far-Field Voice also use GNA technology.

So we can understand a greater computational power in edge computing, because with a differentiated. Energy efficiency, thus allowing to process inference without compromising CPU workload. 

The source code for the libgda library can be downloaded from https://github.com/intel/gna. To understand its implementation from kernel module compilation to library abstraction, I have posted an article in the SDB (support database) category of the openSUSE Linux distribution https://en.opensuse.org/SDB:Install_GNA_in_NUC_Beast_Canyon.

The hardware used was the Intel NUC 11 Extreme “Beast Canyon” which is pre-ordered in the US. But with the instructions in the article, it is possible to understand the technology before it hits the mass market. Soon a tutorial in Portuguese on the portal Viva O Linux.

Conclusion: The future has arrived! The desktop, server and notebook market will provide hardware with the ability to process inference without compromising CPU usage (thus avoiding overhead). With this feature, the artificial intelligence used with a hybrid eco system (computer edge and cloud), will bring greater benefits with a lower computational cost.