We have Moore’s Law for a while. At least this is what the CEO of Intel has stated in a recent interview, Pat gelsinger. In an online conversation with precisely Gordon Moore, the manager assured that the semiconductor industry has done its homework so that the famous law, which states that the power of processors doubles every two years, will continue to be fulfilled for at least minus the next ten years.
It is not just any promise. Until less than a year ago, the technology industry had assumed that continuing to comply with Moore’s Law was nothing short of a chimera, and that the great advances for the computer industry would in any case come from the software side, which would surpass hardware. in ability to innovate.
However, for Gelsinger, history can be written differently. After 12 years in which the ICT industry has joined forces to go from petaflop to the exaflop As a way of measuring performance in large supercomputing centers, the CEO of Intel is convinced that it will take less than five years to reach the zettaflop as a new order of magnitude in these centers.
The interview, which has been developed within the framework of the Intel Innovation Event, also serves as a declaration of intent for large hyperscalar companies in the public cloud, which are by far the companies that currently acquire the most processors and for whom this type of messages may make more sense.
However, this is not the only interesting message that Gelsinger wanted to deliver at the event. The Intel executive expressed his opinion that after a few years in which computing capacity has been decentralized in the heat of precisely cloud computing, we are heading into a new scenario in which that capacity is going to be increasingly transferred to the extreme, in solutions of edge computing where latency plays a key role, for example to process AI workloads in real time.
Intelligent Processing Unit
Among the more specific announcements, Gelsinger has referenced Intel’s partnership with Google for the development of its Mount Evans intelligent processing unit (IPU) and a development kit to make network and data center infrastructure programmable.
This new capacity will be available not only for large cloud players, but also for large financial or insurance companies that are already able to take advantage of programmable infrastructure technologies (Intel Tofino 3, the P4 programming language), in order to achieve high performance. in applications that, for example, need to do network transactions at very high frequencies.
According to Intel, these technologies will allow organizations modernize the way you manage your networks and data centers, avoiding problems such as packet loss at the millisecond level. In this sense, the company assures that in this development it will be possible to program all the elements of the network and server infrastructure.
At the same time, the company ensures that it is working to make advanced computing capabilities, such as machine learning, accessible to more people, with initiatives such as toolkits. oneAPI or the AI inference engine OpenVINO for Intel processors.
Finally, from the multinational they affirm that they intend to further increase the availability of AI with Sapphire Rapids, the code name for its next-generation Xeon processor, which promises to deliver 30 times the performance.
