Of course, these are not the first independent GPUs manufactured by Intel. Already in the 1990s, there were corresponding chips. If we refer to more modern times, we can certainly talk about the fact that the semiconductor company is innovating here. At the same time, the company is venturing into very specific areas.
Ponte Vecchio chips are produced in a 7 nanometer architecture and must actually be deployed in a relatively short time. Because they will not be presented in the form presented now in the free sale and certainly not as a graphics card in the world of gamers. Instead, the Aurora supercomputer GPUs are used to ensure that for the first time the mark of an Exaflops is broken.
The best of the best
In the individual nodes of the Aurora cluster, two Xeon processors do the basic work, but pass the crucial part of the calculations to six Vecchio Bridge GPUs each. To avoid any bottleneck in memory, Intel relies on a combination of DDR5 memory and Optane memory.
The Compute Express Link (CXL) is another feature of previous architectures in supercomputers. It is not absolutely necessary that all data is always executed via the Xeon – the six graphical processors of the node can also communicate directly with each other.
Intel will drive the HPC market with its own GPUs, rivaling the Radeon Instinct and Tesla AMD and Nvidia series. After all, you are moving on a field that you know well. However, Intel also wants to offer GPUs for PCs and laptops, which are essentially based on the same architecture.
See as well: Aurora: US hits records for first exascale supercomputers