4 ms·

800V to each rackmount unit, with hot plugging of rack units? That's scary. The usual setup at this voltage is that you throw a hulking big switch to cut the po

by Animats 7mo ago

800V to each rackmount unit, with hot plugging of rack units? That's scary. The usual setup at this voltage is that you throw a hulking big switch to cut the power, and that mechanically unlocks the cabinet. But that's not what these people have in mind. They want hot-plugging of individual rackmount units. GE has a paper about the power conversion design, but it doesn't mention the unit to rack electrical and mechanical interface. Liteon is working on that, but the animation is rather vague.[2] They hint at hot plugging but hand-wave how the disconnects work. Delta offers a few more hints.[3] There's a complex hot-plugging control unit to avoid inrush currents on plug-in and arcing on disconnect. This requires active management of the switching silicon carbide MOSFETs. There ought to be a mechanical disconnect behind this, so that when someone pulls out a rackmount unit, a shutter drops behind it to protect people from 800V. All these papers are kind of hand-wavey about how the electrical safety works. Plus, all this is liquid-cooled, and that has to hot-plug, too. [1] https://library.grid.gevernova.com/white-papers-case-studies/ai-factory-800vdc-reference-designs https://library.grid.gevernova.com/white-papers-case-studies... [2] https://www.youtube.com/watch?v=CQOreYMhe-M& https://www.youtube.com/watch?v=CQOreYMhe-M& [3] https://filecenter.deltaww.com/Products/download/2510/202510160921259104.pdf https://filecenter.deltaww.com/Products/download/2510/202510...

rdtsc 7mo ago
It is a pretty clever design > When it is detected that the PDB starts to detach from the interface, the hot-swap controller quickly turns off the MOSFET to block the discharge path from Cin to the system. After the main power path is completely disconnected, the interface is physically detached, and no current flows at this time > For insertion, long pins (typically for ground and control signals) make contact first to establish a stable reference and enable pre-insertion checks, while short pins (for power or sensitive signals) connect later once conditions are safe; during removal, the sequence is reversed, with short pins disconnecting first to minimize interference.
- Animats 7mo ago
  Yes, I read that. There had better be a backup system. MOSFETs tend to fail ON, and there's a megawatt going into each rack. Somehow this seems the wrong approach to AI.
  - superxpro12 7mo ago
    Fail spectacularly Data center workers are gonna need those big yoink sticks and those thick arc-fault bibs that furnace operators wear.
    - Animats 7mo ago
      McMaster-Carr has the personal protective equipment required for 800V: Flash protection face shields [1] and Class 0 electrical protection gloves.[2] It's not that bad. It's just ordinary industrial protective gear. [1] https://www.mcmaster.com/products/arc-flash-protection-face-shields/ https://www.mcmaster.com/products/arc-flash-protection-face-... [2] https://www.mcmaster.com/products/electrical-protection-gloves/gloves-2~/ https://www.mcmaster.com/products/electrical-protection-glov...
  - rdtsc 7mo ago
    I figure the latch-up failure mode is a known failure mode. It's not something totally unexpected. > Somehow this seems the wrong approach to AI. One a more ironic level, it's seems on par with the whole AI approach!
- contravariant 7mo ago
  The phrase "when it is detected" makes me nervous.
  - rdtsc 7mo ago
    That's why I included the bottom part describing how some pins are longer than others. It's sort of how some hotplugs work in most cases. First ground connects, then other set of pins and so on. So when is detected a physical act of being disconnected.
dist-epoch 7mo ago
Or maybe you can require technicians to be in full-protection electrical suits.
elif 7mo ago
As long as you can control for fire, electrical safety seems like a temporary condition as robots and intelligent machines are cheaper and more available long term solution to hot swap blades in datacenter racks.
- QuantumGood 7mo ago
  I think you're being downvoted for speaking of a complex future possibility ("robots and intelligent machines ... solution") as if it was a proven commodity. There will be many twists and turns in the path to the possible reliability, scalability, and cost effectiveness of robots and intelligent machines.
  - elif 7mo ago
    Yea I literally said long term but HN voters have such a hate boner for anything adjacent to AI
    - kmacdough 7mo ago
      But the discussion is about the safety of systems being actively implemented. For now, people are required to maintain data centers. Right now, these centers are transitioning to high voltage DC. Right now the best attempts at fully automated data centers are in progress with unverified status. So, for now we need to keep the humans safe. Future data centers not needing humans doesn't keep the current humans safe. Thus, optimistic future speculation is off topic and not relevant to the discussion of human safety with high voltages. People are tired of off topic speculation masquerading as relevant to real world problem solving. It's not a hate boner. EDIT: expanded to make it clearer why I believe the speculation to be wildly off topic for this particular thread.
  - elif 7mo ago
    It is a proven commodity already. Just not in our behind-the-curve country. Look at NTT Data or SoftBank. https://www.softbank.jp/en/sbnews/entry/20250917_01 https://www.softbank.jp/en/sbnews/entry/20250917_01
riedel 7mo ago
An EV fast charger can do 1000V, so with a bit of logic that sounds doable.
- t0mas88 7mo ago
  EV chargers take a different approach. There is no power on the connector while you're plugging it in. It then locks in place before the contactor closes and power is delivered. Unplugging is the same, power is removed before the plug is unlocked for release.
  - dotancohen 7mo ago
    Do the EV chargers not have an alternative disconnect sequence for when the two sides are forcibly mechanically separated? Either by accident, malicious intent, or mechanical failure of some component.
    - Animats 7mo ago
      Yes. See UL 2594.[1] There's a lot of monitoring on those things, plus regular circuit breakers and GFCIs. [1] https://www.intertek.com/batteries/automotive-battery-ev-testing/evse/ul-2594/ https://www.intertek.com/batteries/automotive-battery-ev-tes...
  - reportingsjr 7mo ago
    This is effectively how hot swapping is done for may modern systems. It seems counter intuitive to call it hot plugging, but it’s what you’ll find if you look up “hot swap controller”.
hinkley 7mo ago
The power connectors will be on the far side of the rack from the service side so shouldn’t be a problem for humans touching the third rail so to speak. With that sort of voltage you should be able to use a capacitive or inductive sensor to activate a relay.
- mikewarot 7mo ago
  With a power rail that could supply kiloamps of fault current, thus megajoules of energy into an arc, I wouldn't be so sure of that. Copper vapor inhalation is a definite possibility and a horrible way to die.
  - hinkley 7mo ago
    Obviously the contact point in the sled will need to be disconnected while not grounded, as the top level suggested, or shorted to ground across a large resistor. It’s not the amperage that’s going to cause problems it’s the 800 volts, backed by the amperage. Small enough capacitance and large enough ground load should let you tap in without frying anything including air. Wouldn’t the biggest problem will be loading sleds into a hot rack, with all of the sleds powered down? Which probably you shouldn’t do. The parallel loads on a rack that’s half on should sink a lot of the potential.
brucedawson 7mo ago
It's easy to make these 800V systems totally safe. You just have to combine this idea with the extremely sensible plan to put data centers in space and then there will be no worry about when "someone pulls out a rackmount unit" because there not be someone within hundreds of kilometers.
- kmacdough 7mo ago
  The concern isn't keeping people out of the facility altogether. It's keeping people safe when they are hot swapping dead or outdated rack units. You could consider a robotized approach, but for now robots aren't very good at this type of task. Whether the robotized system is in space is irrelevant. It's not a "next five years" type of solution.
chris222 7mo ago
A high powered EV charger is 800 volts. Totally safe and outdoors in thousands of places.

by Animats 7mo ago

rdtsc 7mo ago

It is a pretty clever design > When it is detected that the PDB starts to detach from the interface, the hot-swap controller quickly turns off the MOSFET to block the discharge path from Cin to the system. After the main power path is completely disconnected, the interface is physically detached, and no current flows at this time > For insertion, long pins (typically for ground and control signals) make contact first to establish a stable reference and enable pre-insertion checks, while short pins (for power or sensitive signals) connect later once conditions are safe; during removal, the sequence is reversed, with short pins disconnecting first to minimize interference.

Animats 7mo ago
Yes, I read that. There had better be a backup system. MOSFETs tend to fail ON, and there's a megawatt going into each rack. Somehow this seems the wrong approach to AI.
- superxpro12 7mo ago
  Fail spectacularly Data center workers are gonna need those big yoink sticks and those thick arc-fault bibs that furnace operators wear.
  - Animats 7mo ago
    McMaster-Carr has the personal protective equipment required for 800V: Flash protection face shields [1] and Class 0 electrical protection gloves.[2] It's not that bad. It's just ordinary industrial protective gear. [1] https://www.mcmaster.com/products/arc-flash-protection-face-shields/ https://www.mcmaster.com/products/arc-flash-protection-face-... [2] https://www.mcmaster.com/products/electrical-protection-gloves/gloves-2~/ https://www.mcmaster.com/products/electrical-protection-glov...
- rdtsc 7mo ago
  I figure the latch-up failure mode is a known failure mode. It's not something totally unexpected. > Somehow this seems the wrong approach to AI. One a more ironic level, it's seems on par with the whole AI approach!
contravariant 7mo ago
The phrase "when it is detected" makes me nervous.
- rdtsc 7mo ago
  That's why I included the bottom part describing how some pins are longer than others. It's sort of how some hotplugs work in most cases. First ground connects, then other set of pins and so on. So when is detected a physical act of being disconnected.

dist-epoch 7mo ago

Or maybe you can require technicians to be in full-protection electrical suits.

elif 7mo ago

As long as you can control for fire, electrical safety seems like a temporary condition as robots and intelligent machines are cheaper and more available long term solution to hot swap blades in datacenter racks.

QuantumGood 7mo ago
I think you're being downvoted for speaking of a complex future possibility ("robots and intelligent machines ... solution") as if it was a proven commodity. There will be many twists and turns in the path to the possible reliability, scalability, and cost effectiveness of robots and intelligent machines.
- elif 7mo ago
  Yea I literally said long term but HN voters have such a hate boner for anything adjacent to AI
  - kmacdough 7mo ago
    But the discussion is about the safety of systems being actively implemented. For now, people are required to maintain data centers. Right now, these centers are transitioning to high voltage DC. Right now the best attempts at fully automated data centers are in progress with unverified status. So, for now we need to keep the humans safe. Future data centers not needing humans doesn't keep the current humans safe. Thus, optimistic future speculation is off topic and not relevant to the discussion of human safety with high voltages. People are tired of off topic speculation masquerading as relevant to real world problem solving. It's not a hate boner. EDIT: expanded to make it clearer why I believe the speculation to be wildly off topic for this particular thread.
- elif 7mo ago
  It is a proven commodity already. Just not in our behind-the-curve country. Look at NTT Data or SoftBank. https://www.softbank.jp/en/sbnews/entry/20250917_01 https://www.softbank.jp/en/sbnews/entry/20250917_01

riedel 7mo ago

An EV fast charger can do 1000V, so with a bit of logic that sounds doable.

t0mas88 7mo ago
EV chargers take a different approach. There is no power on the connector while you're plugging it in. It then locks in place before the contactor closes and power is delivered. Unplugging is the same, power is removed before the plug is unlocked for release.
- dotancohen 7mo ago
  Do the EV chargers not have an alternative disconnect sequence for when the two sides are forcibly mechanically separated? Either by accident, malicious intent, or mechanical failure of some component.
  - Animats 7mo ago
    Yes. See UL 2594.[1] There's a lot of monitoring on those things, plus regular circuit breakers and GFCIs. [1] https://www.intertek.com/batteries/automotive-battery-ev-testing/evse/ul-2594/ https://www.intertek.com/batteries/automotive-battery-ev-tes...
- reportingsjr 7mo ago
  This is effectively how hot swapping is done for may modern systems. It seems counter intuitive to call it hot plugging, but it’s what you’ll find if you look up “hot swap controller”.

hinkley 7mo ago

The power connectors will be on the far side of the rack from the service side so shouldn’t be a problem for humans touching the third rail so to speak. With that sort of voltage you should be able to use a capacitive or inductive sensor to activate a relay.

mikewarot 7mo ago
With a power rail that could supply kiloamps of fault current, thus megajoules of energy into an arc, I wouldn't be so sure of that. Copper vapor inhalation is a definite possibility and a horrible way to die.
- hinkley 7mo ago
  Obviously the contact point in the sled will need to be disconnected while not grounded, as the top level suggested, or shorted to ground across a large resistor. It’s not the amperage that’s going to cause problems it’s the 800 volts, backed by the amperage. Small enough capacitance and large enough ground load should let you tap in without frying anything including air. Wouldn’t the biggest problem will be loading sleds into a hot rack, with all of the sleds powered down? Which probably you shouldn’t do. The parallel loads on a rack that’s half on should sink a lot of the potential.

brucedawson 7mo ago

It's easy to make these 800V systems totally safe. You just have to combine this idea with the extremely sensible plan to put data centers in space and then there will be no worry about when "someone pulls out a rackmount unit" because there not be someone within hundreds of kilometers.

kmacdough 7mo ago
The concern isn't keeping people out of the facility altogether. It's keeping people safe when they are hot swapping dead or outdated rack units. You could consider a robotized approach, but for now robots aren't very good at this type of task. Whether the robotized system is in space is irrelevant. It's not a "next five years" type of solution.

chris222 7mo ago

A high powered EV charger is 800 volts. Totally safe and outdoors in thousands of places.