Numenta Startup Achieves Breakthrough in Artificial Intelligence with Intel Xeon Processors
Numenta, a startup in the field of artificial intelligence, has made a significant breakthrough by demonstrating that Intel Xeon processors can outperform the best processors when it comes to working with AI tasks. Using their innovative approach, called the Numenta Platform for Intelligent Computing (Nupic), the company has achieved new levels of performance on traditional processors in the AI field, as reported by Serv the Home portal.
What makes this breakthrough even more remarkable is that Numenta’s approach allows them to surpass the performance of CPUs and GPUs specifically designed for AI work. In fact, Numenta compared the performance of their solution with the NVIDIA A100 graphic processor using an improved 48-core processor of the 4th generation Sapphire Rapids from Intel. In all tests, the improved Intel processor demonstrated higher performance in terms of total throughput compared to the NVIDIA chip. The results showed that the new approach is 64 times faster than the 3rd generation Intel Xeon processor and 10 times faster than the A100 graphic processor.
Numenta is known for their neurobiological approach to AI workloads, which is largely based on the concept of sparse computing – the way the brain forms connections between neurons.
While most modern central and graphic processors are designed for dense computing, which takes a more “gross” approach compared to the brain’s contextual data processing, Numenta’s use of sparse computing can significantly increase productivity. However, traditional processors cannot effectively handle this mode without specialized optimizations.
The Numenta solution is based on the utilization of Advanced Vector Extensions (AVX) – 512 and Advanced Matrix Extensions (AMX) for Intel Xeon processors. These technologies are extensions of the X86 architecture, providing additional sets of instructions that enable processors to perform more demanding functions.
The Nupic service is supplied in the form of Docker containers and can be launched on dedicated servers. This technology proves to be a promising solution for repurposing existing processors in data processing centers for AI tasks, especially given the ongoing supply