Vitor Pamplona on Nostr: Copper is too slow for AI. The era of silicon photonics (pre-packaged fiber optic ...
Copper is too slow for AI. The era of silicon photonics (pre-packaged fiber optic between chips) is finally here.
Optical switches on an AI data center consume a staggering 10% of the total GPU compute power.
“An AI supercomputer with 400,000 GPUs is actually a 24-megawatt laser”
https://spectrum.ieee.org/co-packaged-opticsPublished at
2025-04-02 18:35:59Event JSON
{
"id": "ad4f1b70780fd308b9a308e0439e7c57c5b064e19a3213cbd5fe932e810f1884",
"pubkey": "460c25e682fda7832b52d1f22d3d22b3176d972f60dcdc3212ed8c92ef85065c",
"created_at": 1743618959,
"kind": 1,
"tags": [
[
"alt",
"A short note: Copper is too slow for AI. The era of silicon phot..."
],
[
"r",
"https://spectrum.ieee.org/co-packaged-optics"
]
],
"content": "Copper is too slow for AI. The era of silicon photonics (pre-packaged fiber optic between chips) is finally here.\n\nOptical switches on an AI data center consume a staggering 10% of the total GPU compute power. \n\n“An AI supercomputer with 400,000 GPUs is actually a 24-megawatt laser”\nhttps://spectrum.ieee.org/co-packaged-optics",
"sig": "178aceea956fdb75055332ea64ab21ab149a04af5439acfca630981e01f7290edee18fc3c12432c27fffaa04ad97a7fb09181ace6a5714e4506b2196b273d6d7"
}
