3 ms·

Seems like an odd decision to move power regulation to the main board when it seems like an obvious task for a PSU. Anyone have onsite into why they are doing

by Jemm 7y ago

Seems like an odd decision to move power regulation to the main board when it seems like an obvious task for a PSU. Anyone have onsite into why they are doing this?

NovemberWhiskey 7y ago
If you take a look at a modern PC motherboard, you'll see there's already a considerable amount of board space allocated to local voltage regulation: modern CPUs can require multiple different voltages none of which will be supplied by an ATX power supply. Often it will be desirable to have fine control in software over the regulated voltages to maximize system performance or stability. RAM and other devices on the board, as well as high-powered devices like GPUs probably need their own voltages too. From my perspective, it's just the continuation of the trend to its logical conclusion: if you're going to be generating eight different voltages on the motherboard, GPU etc. why not just add an extra two?
myself248 7y ago
Because logic voltages change over time. When the PC was new, the AT PSU produced a ton of 5v, since most of the logic was 5v TTL circuitry, or CMOS running at 5v. The various 12v and -12v rails were mostly for RS232 ports and drive spindle motors. By the time ATX came along, most of the logic was 3v3, but there was still a fair bit of 5v load, and people were putting more drives into PCs so those needed a lot of 12v. Look at the wires in an ATX connector, there are 2 for 12v, 5 for 5v, and 3 for 3v3. Since all the wires are the same gauge, this hints at how much current the designers thought the motherboard would need on each rail. But pretty shortly after ATX came out, two things happened: First, CPUs stopped being 3.3v, and a big chunk of the load on the 3v3 rail went away. CPUs got VRMs, which sometimes drew from the 5v rail but mostly drew from the 12v rail, to produce the 1.8v or 1.5v or 1.2v or whatever the CPU Vcore wanted. This shifted a lot of the load to those rails, and 12v was preferable since it would suffer less loss coming from the PSU, and could use thinner traces in the motherboard. Second, GPUs were invented. Remember that graphics cards in the ISA/PCI era were mostly just a framebuffer and a ramdac. 3D accelerators became a thing, and started throwing heaps of silicon at the problem, which needed heaps of power. Graphics cards started getting heatsinks, of all things! And since the workload tends to be very parallelizable, the only bound on how much silicon you could recruit, was how much power you could supply. The motherboard's traces and slot contacts could only move so much current, so graphics cards sprouted additional connectors to suckle straight from the PSU. As with CPUs, GPUs ran at core voltages which were A) constantly evolving, B) user-tweakable for performance, and C) incredibly low and impractical to run over long wires because of voltage drop. All these factors made it nonsensical to ask the PSU to produce GPU Vcore. So they all had local regulation that just ran from 12v, for the same reasons as CPU VRMs. Ever since then, the majority of load has been on 12v. A modern PC draws basically nothing on 3v3 since all the logic has moved to local regulation, and the 5v rail is used more for USB than anything else. The negative rails are vestigial (used for some 80s-era logic and RS232 transceivers) anyway. And the strategy of local regulation has proven both efficient and practical. So why not use it for everything? 12v is a sane voltage to standardize on. It's a little lower than I'd like for efficiency reasons (24 would be better), but a ton of proven hardware is already in the field that runs from 12v and it's quite well understood. Going higher than 30 requires different silicon in the power converters, and lower than 12 starts to really hurt efficiency. It's a good compromise. And a single rail will massively simplify PSU design, and improve regulation because there won't be cross-loading concerns anymore.

by Jemm 7y ago

Seems like an odd decision to move power regulation to the main board when it seems like an obvious task for a PSU. Anyone have onsite into why they are doing this?

NovemberWhiskey 7y ago

If you take a look at a modern PC motherboard, you'll see there's already a considerable amount of board space allocated to local voltage regulation: modern CPUs can require multiple different voltages none of which will be supplied by an ATX power supply. Often it will be desirable to have fine control in software over the regulated voltages to maximize system performance or stability. RAM and other devices on the board, as well as high-powered devices like GPUs probably need their own voltages too. From my perspective, it's just the continuation of the trend to its logical conclusion: if you're going to be generating eight different voltages on the motherboard, GPU etc. why not just add an extra two?

myself248 7y ago

Because logic voltages change over time. When the PC was new, the AT PSU produced a ton of 5v, since most of the logic was 5v TTL circuitry, or CMOS running at 5v. The various 12v and -12v rails were mostly for RS232 ports and drive spindle motors. By the time ATX came along, most of the logic was 3v3, but there was still a fair bit of 5v load, and people were putting more drives into PCs so those needed a lot of 12v. Look at the wires in an ATX connector, there are 2 for 12v, 5 for 5v, and 3 for 3v3. Since all the wires are the same gauge, this hints at how much current the designers thought the motherboard would need on each rail. But pretty shortly after ATX came out, two things happened: First, CPUs stopped being 3.3v, and a big chunk of the load on the 3v3 rail went away. CPUs got VRMs, which sometimes drew from the 5v rail but mostly drew from the 12v rail, to produce the 1.8v or 1.5v or 1.2v or whatever the CPU Vcore wanted. This shifted a lot of the load to those rails, and 12v was preferable since it would suffer less loss coming from the PSU, and could use thinner traces in the motherboard. Second, GPUs were invented. Remember that graphics cards in the ISA/PCI era were mostly just a framebuffer and a ramdac. 3D accelerators became a thing, and started throwing heaps of silicon at the problem, which needed heaps of power. Graphics cards started getting heatsinks, of all things! And since the workload tends to be very parallelizable, the only bound on how much silicon you could recruit, was how much power you could supply. The motherboard's traces and slot contacts could only move so much current, so graphics cards sprouted additional connectors to suckle straight from the PSU. As with CPUs, GPUs ran at core voltages which were A) constantly evolving, B) user-tweakable for performance, and C) incredibly low and impractical to run over long wires because of voltage drop. All these factors made it nonsensical to ask the PSU to produce GPU Vcore. So they all had local regulation that just ran from 12v, for the same reasons as CPU VRMs. Ever since then, the majority of load has been on 12v. A modern PC draws basically nothing on 3v3 since all the logic has moved to local regulation, and the 5v rail is used more for USB than anything else. The negative rails are vestigial (used for some 80s-era logic and RS232 transceivers) anyway. And the strategy of local regulation has proven both efficient and practical. So why not use it for everything? 12v is a sane voltage to standardize on. It's a little lower than I'd like for efficiency reasons (24 would be better), but a ton of proven hardware is already in the field that runs from 12v and it's quite well understood. Going higher than 30 requires different silicon in the power converters, and lower than 12 starts to really hurt efficiency. It's a good compromise. And a single rail will massively simplify PSU design, and improve regulation because there won't be cross-loading concerns anymore.