Intel launches optical compute interconnect chiplet: Adding 4 Tbps optical connectivity to CPUs or GPUs

[erek]

"The current optical I/O chiplet is largely a prototype, and Intel is collaborating with select customers to further develop and integrate this device with next-generation systems-on-chips (SoCs) and system-in-packages (SiPs).


"The ever-increasing movement of data from server to server is straining the capabilities of today’s data center infrastructure, and current solutions are rapidly approaching the practical limits of electrical I/O performance," said Thomas Liljeberg, senior director of product Management and Strategy, Integrated Photonics Solutions (IPS) Group. "However, Intel’s groundbreaking achievement empowers customers to seamlessly integrate co-packaged silicon photonics interconnect solutions into next-generation compute systems. Our OCI chiplet boosts bandwidth, reduces power consumption and increases reach, enabling ML workload acceleration that promises to revolutionize high-performance AI infrastructure."

Intel's silicon photonics initiative is backed by over 25 years of research and extensive deployment in data centers. The company's hybrid laser-on-wafer technology and direct integration approach offer high reliability and cost efficiency, which Intel says sets it apart from competitors.

So far, Intel has shipped over 8 million photonic integrated circuits (PICs) with more than 32 million integrated on-chip lasers. These PICs were integrated into pluggable transceiver modules and used in large data centers at major cloud providers for 100, 200, and 400 Gbps applications. Next-generation PICs, supporting 200G per lane, are being developed for 800 Gbps and 1.6 Tbps applications."

Source: https://www.tomshardware.com/deskto...g-4-tbps-optical-connectivity-to-cpus-or-gpus

 

[Lakados]

This is some funny shit right here because of the following series of events...

Nvidia released this monstrosity here.
https://www.semianalysis.com/p/nvidias-optical-boogeyman-nvl72-infiniband

To which Jim Keller formerly of Intel responded with.
" Nvidia should have used the Ethernet protocol chip-to-chip connectivity in Blackwell-based GB200 GPUs for AI and HPC"
https://www.tomshardware.com/tech-i...ed-ethernet-to-stitch-together-blackwell-gpus

For Intel to come out with the same conclusions, that copper just doesn't cut it.

This is the sort of comedy I follow this site for.

 

[erek]

This is some funny shit right here because of the following series of events...

Nvidia released this monstrosity here.
https://www.semianalysis.com/p/nvidias-optical-boogeyman-nvl72-infiniband
View attachment 662234

To which Jim Keller formerly of Intel responded with.
" Nvidia should have used the Ethernet protocol chip-to-chip connectivity in Blackwell-based GB200 GPUs for AI and HPC"
https://www.tomshardware.com/tech-i...ed-ethernet-to-stitch-together-blackwell-gpus

For Intel to come out with the same conclusions, that copper just doesn't cut it.

This is the sort of comedy I follow this site for.

really awesome to see this

thanks for sharing

 

[Balkroth]

This is some funny shit right here because of the following series of events...

Nvidia released this monstrosity here.
https://www.semianalysis.com/p/nvidias-optical-boogeyman-nvl72-infiniband

To which Jim Keller formerly of Intel responded with.
" Nvidia should have used the Ethernet protocol chip-to-chip connectivity in Blackwell-based GB200 GPUs for AI and HPC"
https://www.tomshardware.com/tech-i...ed-ethernet-to-stitch-together-blackwell-gpus

For Intel to come out with the same conclusions, that copper just doesn't cut it.

This is the sort of comedy I follow this site for.

I mean, you can use ethernet over optical networks, it doesn't have to be over copper, so his point still stands.