[ad_1]
Ahead-looking: We’re approaching some extent the place conventional copper interconnections will not be capable of carry sufficient information to maintain GPUs and different specialised chips totally utilized. The AI market is urgently demanding a next-generation resolution to this interconnection bottleneck, and Broadcom seems to be engaged on an optics-based resolution that’s nearer to the chip itself.
Broadcom is growing new silicon photonics know-how aimed toward considerably growing the bandwidth accessible to GPUs and different AI accelerators. By using co-packaged optics (CPOs), the fabless chip producer goals to combine optical connectivity elements straight into GPUs, enabling greater information charges whereas concurrently lowering energy necessities.
The corporate has been engaged on CPO options for a number of years and showcased its newest developments on the latest Scorching Chips conference. Broadcom’s “optical engine” reportedly delivers a complete interconnect bandwidth of 1.6 TB/sec, equal to six.4 Tbit/sec or 800 GB/sec in every path.
This new connection can present “error-free” information switch to a single chiplet, reaching efficiency ranges corresponding to Nvidia’s NVLink and different specialised information heart options. Nonetheless, Broadcom has not but included its optical interconnections right into a market-ready GPU, such because the A100 or MI250X. As a substitute, it used a check chip designed to emulate an actual GPU for demonstration functions.
In line with Manish Mehta, Broadcom’s vice chairman of the optical methods division, copper connections begin to degrade after simply 5 meters. Whereas optical communications have lengthy been considered as the answer to this sign degradation subject, they historically require far more energy than copper-based applied sciences.
For instance, Nvidia estimates that an optics-powered NVL72 system would require a further 20 kilowatts per rack, on prime of the 120 kilowatts the system already consumes.
Broadcom has managed to cut back energy consumption with the usage of co-packaged optics, which locations particular person transceivers in direct contact with the GPU. The corporate utilized TSMC’s chip-on-wafer-on-substrate (CoWoS) packaging know-how to bond a pair of high-bandwidth reminiscence stacks to the compute die. The logic and reminiscence elements of the chip sit on a silicon interposer, whereas Broadcom’s optical engine is situated on the substrate.
Mehta defined that CPO know-how might join as much as 512 particular person GPUs throughout eight racks, permitting your complete setup to perform as a single system. Compared, Nvidia’s NVL72 can obtain related unified computing capabilities with “simply” 72 GPUs, suggesting that Broadcom’s resolution might ultimately supply a aggressive benefit for next-generation AI workloads.
[ad_2]