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?
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.
Sounds like it's going to take up a bunch more space on motherboards for sata ports, as each sata port (say 6 for example) would need either it's own power cable port coming from the motherboard, or some form of breakout cable from a (likely) single connector.
My current PSU provides cabling to power 4 SATA devices off of a single 6 pin port (of which I don't think it uses all the pins). I don't think it'll eat a bunch of space at all.
Of course if eSATAp is any indicator I would not be surprised if no additional space was used, just a thicker cable.
This seems like an annoying decision for DIY, but the upside seems to be less cable routing. That will surely save costs. Cables and connectors are more expensive than silicon these days.
Old system: Each device needs to connect to PSU and to the motherboard
New system: Each device needs to connect to just the motherboard. Right now, parallel cables, and in a year or two, a common cable carries power+data.
I predict the natural conclusion to this trend will be:
* 2-wire connections, with DC power and AC data on the same lines, a la microphone phantom power. Connectors and cables will be simpler. Electronics will be slightly more complex.
* I predict things like power ready will be communicated over that data line, as will metadata. Instead of 12V sense, we'll have an ADC of voltage on the motherboard communicated back digitally.
* 12V will be on all the time. 12V standby will be achieved by switching power off to some components on the motherboard, and communicating back to the PSU that we don't need a lot of power for a while.
I predict all of this will make computers less hackable for DIY.
Did you RTFA? Pin 7 is 12vSB, which will be on all the time, not the main 12V rails. It's likely to work exactly like 5vSB right now, just at a different voltage.
Personally I predict this will make it much easier to DIY things like vehicle PSUs, since you'll only need a single buckboost converter and some ignition-sense logic. All other factors will be relatively unchanged, just the red and orange wires are going away which is fine because basically nothing used them anyway.
I was wondering what it had to do with the Raspberry Pi considering the hostname, and was secretly hoping this could be a way to use a bigger power supply to power in a Raspberry Pi and also supply 12V to standard 3.5" HDD. That would simplify the wiring for those who wants to make a NAS.
10 comments
[ 2.8 ms ] story [ 35.3 ms ] threadAnyone have onsite into why they are doing this?
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?
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.
Of course if eSATAp is any indicator I would not be surprised if no additional space was used, just a thicker cable.
Guess you're right then. :)
*
Separately, hadn't heard of eSATAp before:
https://en.m.wikipedia.org/wiki/ESATAp
Thanks for pointing it out. :)
Old system: Each device needs to connect to PSU and to the motherboard
New system: Each device needs to connect to just the motherboard. Right now, parallel cables, and in a year or two, a common cable carries power+data.
I predict the natural conclusion to this trend will be:
* 2-wire connections, with DC power and AC data on the same lines, a la microphone phantom power. Connectors and cables will be simpler. Electronics will be slightly more complex.
* I predict things like power ready will be communicated over that data line, as will metadata. Instead of 12V sense, we'll have an ADC of voltage on the motherboard communicated back digitally.
* 12V will be on all the time. 12V standby will be achieved by switching power off to some components on the motherboard, and communicating back to the PSU that we don't need a lot of power for a while.
I predict all of this will make computers less hackable for DIY.
Personally I predict this will make it much easier to DIY things like vehicle PSUs, since you'll only need a single buckboost converter and some ignition-sense logic. All other factors will be relatively unchanged, just the red and orange wires are going away which is fine because basically nothing used them anyway.
Connect 5v to RPi (with a fuse and potentially a zener diode), connect 12v to HDD via standard molex, job done.