AI ambition is pushing copper to its breaking point

Datacenters have been trending toward denser, more power-hungry systems for years. In case you missed it, 19-inch racks are now pushing power demands beyond 120 kilowatts in high-density configurations, with many making the switch to direct liquid cooling to tame the heat. Much of this trend has been driven by a need to …

[Eclectic Man]

Alternatives to Cu

If an optical chiplet fails, there goes your $40,000 accelerator.

I wonder whether silver (Ag) is a viable alternative to copper with the improvement in conductivity, but probably somewhat less expensive than $40,000. When the Allies were building the first atomic bombs the Manhattan Project's boffins realised they would need an enormous amount of coper for the windings necessary for the electromagnets to separate Uranium isotopes, so they 'borrowed' the requisite amount of silver from the US Treasury.* The rest. as they say, is history.

* https://www.americanscientist.org/article/from-treasury-vault-to-the-manhattan-project#:~:text=That%20silver%20was%20shaped%20into,%2C%20Japan%2C%20in%20August%201945.

[DS999]

Re: Alternatives to Cu

Any silver exposed to the air (as in a connector which is not airtight) will tarnish, and that reduces the conductivity well below that of copper. At higher voltages (50 or 100v IIRC) it can get through the layer of tarnish but communications happens with far lower voltage.

[Eclectic Man]

Re: Alternatives to Cu

Yup, I realsie that silver tarnishes, however teh Hi-Fi industry advertises silver connectors (at incredible prices, I accept) with gold connectors. The issue is whether replacing g the copper with silver, albeit with untarnishing connectors would be enough of a gain on copper wires at a reduced cost compared to photons.

https://hosatech.com/press-release/the-advantage-of-silver-cable/#:~:text=This%20makes%20it%20easier%20to,sound%20could%20be%20too%20bright.

[DS999]

Re: Alternatives to Cu

They advertise them as snake oil for audiophiles with more money than brains, not because they actually perform better in double blind testing.

[Locomotion69]

A few challenges remain...

This is an interesing proof of concept, IMO not ready for "the masses" yet.

And next, there is the need for hardware capable in processing these high loads, which on their side require power, cooling, etc.

[HuBo]

Re: A few challenges remain...

Yes, and yet it wasn't so long ago that fiber-to-the-cabinet (FttC), and then fiber-to-the-home (FttH), seemed like distant pipe dreams, forever pushed back to future horizons, or Chattanooga ...

It seems to me that the main drawback of CPO today (Co-Packaged Optics) is perceived to be cost, as it was with FttH. But at least one recent analysis suggests 20x better profitability and 3-4x better interactivity with the tech, as opposed to copper!

I'm with Ayar, Celestial, Eliyan, Lightmatter, Ranovus and the likes on this, hoping for the tech to get deployed sooner rather than later, hopefully with reconfigurable capabilities as well (eg. NextSilicon).

The CXL 3.0 and IBM's POWER 9,10,11 OpenCAPI OMI (Spinal Tap CPU coming in 2025, with Differential DIMM -- DDIMM -- @ DDR5) stand to benefit greatly from CPO IMHO!