3 ms·

Intel tried to do the exact opposite, move the VRM to the processor die itself, in Haswell. It did not go well because of the extra heat, but those chips make f

by bserge 6y ago

Intel tried to do the exact opposite, move the VRM to the processor die itself, in Haswell. It did not go well because of the extra heat, but those chips make for great overclock testbeds haha. Other than that, why would you move it to a separate part? Where? Either you push that extra cost onto the motherboard manufacturers or you need a whole new standard and slot for the VRM? What's the advantage? These days, everything is moving towards tighter integration (fuck soldered CPUs, RAM and SSDs), so no one is going to do that.

sp332 6y ago
The advantage is being able to replace the GPU, RAM, etc without having to throw away the VRMs.
- zbrozek 6y ago
  Those VRMs are not that expensive, and you can't realistically move them far away from the load without constantly under- or over-voltaging the transistors you're trying to power. Your power company does not deliver 5 volts to your house for your phone charger, nor does it transport 240 VAC more than a few parcels.
sokoloff 6y ago
We already have 6 and 8 pin additional connectors on graphics cards, so the connector/slot aspect is easily solved. I think the underlying premise is if you upgrade graphics cards more often or otherwise buy 2-4 graphics cards for every motherboard you buy that moving the VRM to the motherboard (and add costs there), you can still save money and reduce waste.
- xxs 6y ago
  >We already have 6 and 8 pin additional connectors on graphics cards, so the connector/slot aspect is easily solved. Those 8 pins are 12V, not 1V. They are spec'd to carry like 12.5A each - make it 15A - bit warmer wires. So around 15W per connector at 1V... and the need is over 300W. So there you have it 20 8-pin connectors. There is a reason power lanes of the grid are higher voltage - they need a lot less current, so they are efficient. Lower voltage should remain on the PCB with multiple power planes.
  - sokoloff 6y ago
    To argue alongside your excellent points: it's not just bit warmer wires, but also the underlying voltage drop that would be harmful in such a design, especially in the context of a 1V DC supply.

by bserge 6y ago

sp332 6y ago

The advantage is being able to replace the GPU, RAM, etc without having to throw away the VRMs.

zbrozek 6y ago
Those VRMs are not that expensive, and you can't realistically move them far away from the load without constantly under- or over-voltaging the transistors you're trying to power. Your power company does not deliver 5 volts to your house for your phone charger, nor does it transport 240 VAC more than a few parcels.

sokoloff 6y ago

We already have 6 and 8 pin additional connectors on graphics cards, so the connector/slot aspect is easily solved. I think the underlying premise is if you upgrade graphics cards more often or otherwise buy 2-4 graphics cards for every motherboard you buy that moving the VRM to the motherboard (and add costs there), you can still save money and reduce waste.

xxs 6y ago
>We already have 6 and 8 pin additional connectors on graphics cards, so the connector/slot aspect is easily solved. Those 8 pins are 12V, not 1V. They are spec'd to carry like 12.5A each - make it 15A - bit warmer wires. So around 15W per connector at 1V... and the need is over 300W. So there you have it 20 8-pin connectors. There is a reason power lanes of the grid are higher voltage - they need a lot less current, so they are efficient. Lower voltage should remain on the PCB with multiple power planes.
- sokoloff 6y ago
  To argue alongside your excellent points: it's not just bit warmer wires, but also the underlying voltage drop that would be harmful in such a design, especially in the context of a 1V DC supply.