With keynote speeches from Raja Koduri and Imad Sousou, Intel is focusing more on the software for the Open Source Technology Summit than before. Because software can be comparatively cheap to optimize the performance and sometimes significantly increase, as it is only possible through hardware improvements.
For the 15th Open Source Technology Summit (OSTS), which will be held for the first time a little more public than last, Intel brings developers from Alibaba, Amazon, AT & T, Google, Huawei, JD.com, Microsoft, MontaVista, Red Hat, SUSE and Wind River together to talk about the possibilities of software for the future. Because undoubtedly, Intel's hardware profits very much from the developments, which is why Intel is constantly expanding its efforts in the segment. Many of the companies involved publicly state today that in the past they did not rely too much on Open Source or even laughed at it. But this seems to be changing now.
Intel emphasizes that already today more than 15,000 employees in the company take care of software – almost 15 percent of the entire workforce. Among other things, these are responsible for the fact that software is one of the most important technologies in order to implement the efforts and improvements in the CPU architecture, process technology, memory usage, new interconnects and security – again a major topic at Intel , emphasized among others Imad Sousou as General Manager of Intel's complete software division and Raja Koduri in his keynote Think Exponentialwho is much more at Intel than just the boss for the upcoming graphics card generation Xe.
At the start of Cascade Lake-SP and -AP, Intel had already mentioned the benefits of optimized software, at least in the small print. Because the high performance, which is now achieved by the DL Boost and VNNI for deep learning, is not only realized via hardware. Compared to the original Xeon-SP, the nearly sixfold performance comes about through optimized software, the more of kernels, clock and additional instructions then ensures the jump to the maximum factor of 30x. And in the future, every Intel model will rely on some kind of acceleration in the field, so proper software is essential. At the start of the OSTS, Intel also presents matching current benchmark scores for deep learning applications, in which CPUs with optimized software even beat Nvidia's fastest GPU solution, the Tesla V100.
Today's optimizations start with the BIOS as one of the basic construction sites. Intel wants to partner with partners to modernize the BIOS again and build modular. This starts with the removal of many legacy issues and goes from modernization to improved scalability. This is to be realized through the ModernFW Initiative, in parallel there are further efforts in the container environment with the focus on the cloud, which is to be realized via rust-vmm.
Intel's own Clear Linux * Project is also a very fast solution for independents, which will be expanded to include more kits in the future. In addition to a Clear Linux Developer Edition, there will be a Deep Learning Reference Stack for the use of Advanced Vector Extensions 512 and the new Vector Neural Network Instructions (AVX 512 VNNI). Another Data Analytics Reference Stack is designed to provide new opportunities in enterprise environments.
ComputerBase has received the information from Intel partly under NDA, was also invited to the OSTS after Stevenson, USA. The flights and accommodation were paid by Intel. There was no requirement for reporting.