Chip stocks just suffered one of their ugliest stretches on record, and a political shift in Washington may be to blame. History offers a surprising clue about what happens next, but a November 17 earnings report could rewrite it.
I’m not so sure. I’m involved in power, cooling, HVAC of data centers as an electrician. Some of these server racks are all liquid cooled and not like an AIO you slap in a desktop, it needs to handle tens of thousands of kw of heat per rack. They don’t run on typical home power, often running on high voltage like 700v DC as a source, are loud and power hungry, relying on industrial grade UPSes and power distribution units or diesel generators in the event of loss of power. They’re built for hundreds of gigabits per second networking via fibers.
I’m not involved in the nitty gritty of what every motherboard or GPU is like but these places absolutely design high level industrial enterprise gear that is specific to doing the task they need. It would take some massive redesign to be able to re-use boards but anything’s possible I guess if a company had a method of taking decommissioned gear.
The problem is they say “we will buy out your entire production capability for the next 4-5 years” to RAM and GPU and chips manufacturers, so consumer grade equipment is kind of an afterthought and thus they scale it back and jack up the prices
The key part of your energy dissipation requirement is that it’s per rack. A home user wouldn’t need an entire rack of compute, so a case built for only one or two boards would have nowhere near the same cooling requirements.
Also designing a different power supply is very trivial. The silicon itself is only running at 1-2v, it’s not like they’re passing 700vdc through the actual compute boards.
I’m not so sure. I’m involved in power, cooling, HVAC of data centers as an electrician. Some of these server racks are all liquid cooled and not like an AIO you slap in a desktop, it needs to handle tens of thousands of kw of heat per rack. They don’t run on typical home power, often running on high voltage like 700v DC as a source, are loud and power hungry, relying on industrial grade UPSes and power distribution units or diesel generators in the event of loss of power. They’re built for hundreds of gigabits per second networking via fibers.
I’m not involved in the nitty gritty of what every motherboard or GPU is like but these places absolutely design high level industrial enterprise gear that is specific to doing the task they need. It would take some massive redesign to be able to re-use boards but anything’s possible I guess if a company had a method of taking decommissioned gear.
The problem is they say “we will buy out your entire production capability for the next 4-5 years” to RAM and GPU and chips manufacturers, so consumer grade equipment is kind of an afterthought and thus they scale it back and jack up the prices
The key part of your energy dissipation requirement is that it’s per rack. A home user wouldn’t need an entire rack of compute, so a case built for only one or two boards would have nowhere near the same cooling requirements.
Also designing a different power supply is very trivial. The silicon itself is only running at 1-2v, it’s not like they’re passing 700vdc through the actual compute boards.