Skip to content Skip to footer

Intel announces its first exascale GPU for AI applications in data centres

Nvidia and AMD arguably have a runaway lead in the AI space thanks to their development of high-powered GPUs that have been installed in data centres around the world. Intel doesn’t want to be left out of the party, however, and has finally announced its first discrete GPU. 

Launched at Supercomputing 2019, the new GPU is codenamed ‘Ponte Vecchio’ and is based on Intel’s Xe architecture. This architecture will eventually cover the full spectrum of computing, including integrated graphics on Intel’s cheaper chipsets, but for now the company is targeting the high performance computing market. That’s because it doesn’t want to be left behind during the ongoing AI revolution that is taking place in data centres around the world. 

In order to not be left behind, Intel is throwing all it has into the development of Ponte Vecchio. The company has confirmed that it will be the first of its products to be manufactured on 7nm technology, beating out the company’s flagship line of CPUs, while it also utilises the Foveros 3D and EMIB packing processes. 

Technically, the Ponte Vecchio won’t be Intel’s first discrete GPU, as the company plans on launching another GPU in 2020 that will be manufactured on 10nm technology. However, that GPU has still yet to be officially detailed, so for now, it gets the honour of being the first. 

An all-new architecture

You’ll be hearing quite a lot about the Intel Xe architecture in the coming months, and while we don’t know all the details yet, we do know that it’s set to be the architecture underlying all of Intel’s forthcoming chipsets. That means those destined for mobiles, gaming PCs, and even in data centres powering cloud gaming services. 

“The Xe is the one architecture with all the features and capabilities that trickle up and down between all the microarchitectures,” Ari Rauch, VP & GM Visual and Parallel Compute Group at Intel, noted.

“Yes, you will migration of capabilities between microarchitectures. That’s the whole idea of having one Xe architecture.”

A GPU fit for exascale computing

While Intel won’t detail the exact technologies that will be packed into the Ponte Vecchio GPU, it did note that it will benefit from a flexible data-parallel vector matrix engine and high double precision (FP64) throughput, as well as a very high cache and DRAM memory bandwidth. 

All that technology, as well as the tech that hasn’t yet been disclosed, will be put to good use in the first computer to benefit from Intel’s new GPUs. That will be located at the Argonne National Laboratory and dubbed the Aurora supercomputer. Set to be installed in 2021, the compute node architecture of Aurora will feature two 10nm-based Intel Xeon Scalable processors (code-named “Sapphire Rapids”) and six Ponte Vecchio GPUs.

You may also like

Stay In The Know

Get the Data Centre Review Newsletter direct to your inbox.