Intel is attracting attention by constructing a supercomputer equipped with its own GPU and Nvidia’s own developed CPU.
Each develops chips in one another’s area of expertise, and Nvidia heralds full-scale competition within the CPU market dominated by Intel, and Intel within the GPU market dominated by Nvidia.
On the ISC 2023 conference held in Hamburg, Intel accomplished the shipment of the next-generation GPU ‘Ponte Vecchio’-based Argonne National Laboratory’s ‘Aurora’ supercomputer. )’ based Isambard 3 supercomputer, Reuters reported on the twenty second (local time).
Ponte Vecchio is Intel’s next-generation GPU. It’s the HPC lineup of the graphic architecture Xe graphic. It relies on the Xe architecture optimized for HPC and AI workloads. By tying a whopping 47 tiles into one, the transistor density was raised to 100 billion.
The GPU is built using Intel’s Foveros 3D packaging technology. It integrates multiple design assets (IP), including HBM memory and other mental property, into one package.
Ponte Vecchio is produced in a 7-nm process at an external foundry. As Intel postponed the production schedule for several years, the plan to construct the Aurora supercomputer also suffered setbacks, but evidently the construct was finally accomplished.
As well as, Intel announced that it’ll soon release the ‘Falcon Shores’ XPU, which was unveiled last 12 months. Falcon Shores is a high-performance and supercomputing processor that mixes Intel’s Sapphire Rapids Xeon Scalable CPU and next-generation GPU Ponte Vecchio into one.
Falcon Shores is a CPU-GPU integrated processor introduced for the primary time within the server and supercomputing sectors. On this case, for the reason that high-performance server CPU and the high-performance computational GPU exchange huge amounts of knowledge, data and memory bottlenecks are more likely to occur.
Considered one of the alternatives to beat that is to integrate the CPU, GPU, and memory right into a single package. Intel believes that with Falcon Shores, it may well increase performance per power and memory bandwidth by greater than 5 times.
The second good thing about integrating server-class CPUs and GPUs is space savings. For those who insert several huge server CPUs and GPUs that take up quite loads of space on a server motherboard, the quantity of the server will inevitably increase.
In a situation where the scale of knowledge centers continues to grow, the looks of small-sized servers is welcome. Ultimately, this results in cost savings.
Nevertheless, Intel didn’t reveal a precise release date for Falcon Shores. With a purpose to launch, the Ponte Vecchio GPU desperately must make a step forward within the high-performance GPU market dominated by Nvidia.
Nevertheless, it’s a situation where the industry may change in a number of years because it is concurrently conducting unprecedented levels of research and investment and is disclosing an ambitious product roadmap. I’m interested by the long run whether Intel’s transformation will succeed.
Meanwhile, Nvidia is working with Hewlett-Packard Enterprises (HPE) to construct a supercomputer called the Isambard 3 on the University of Bristol within the UK, based entirely on Grace CPU chips without Nvidia’s GPUs.
The University of Bristol plans to make use of the supercomputer for research in fields corresponding to AI, life sciences, medicine, biotechnology and astrophysics.
Based on Nvidia, Isambard 3 will feature 384 Grace CPU processors. Each processor comprises 144 CPU cores based on an Arm design. The core supports SVE, an Arm instruction set technology that permits CPUs to run AI workloads more efficiently.
Nvidia expects the Isambard 3 to deliver speeds of as much as 2.7 petaflops when performing calculations involving FP64 data. One petaflop is corresponding to 1,000 trillion computational operations per second. FP64 is a knowledge format commonly utilized by supercomputers to arrange the data they process.
Isambard 3 is scheduled to grow to be operational in 2024. Once operational, the system is predicted to devour lower than 270 kW of power, making it probably the most energy-efficient supercomputers in Europe. Nvidia estimates that the Isambard 3 is six times more energy efficient than the University of Bristol’s previous generation supercomputer.
Nvidia emphasized that as climate change becomes an increasingly existential issue, it will be significant for computing to embrace energy-efficient technologies. In other words, a plan was made to create a more energy-efficient supercomputing center based on CPU, excluding GPU.
It appears that evidently Nvidia, which has dominated the GPU market, which is the idea of high-performance computing and supercomputers, has even secured initiatives within the CPU field that consider energy efficiency.
Reporter Park Chan cpark@aitimes.com