Voltage discussion thread.

AleksandarK

/dev/null
Original poster
May 14, 2017
703
774
With the recent comment made in this thread we started to ge off-topic, so i decided to open a dedicated thread about discussion of voltage in PC/Server space.

My own thoughts so far are that voltage standard should be increased. Allowing us to have more efficient electronics.

Interesting thing to note here is that a PEZY SC2(using 48V) is powering one of the most efficient supercomputers on TOP500 list, which explains enough.
The PEZY-SC2 module card is a node for high-speed computation equipped with this PEZY-SC2 processor, and power transmission loss on the board is greatly reduced by 48 V power supply.
 

Kmpkt

Innovation through Miniaturization
KMPKT
Feb 1, 2016
3,382
5,936
On top of power efficiency, which is always awesome sauce, power connectors could become smaller which is great for SFF. Imagine powering a 16 core SLI rig off an 8 pin PCIe power connector.

I think the downside of 48V at the hardware level is that the downconversion to the very small voltages required in a microprocessor is more inefficient when stepping from say 48V to 1V instead of 12V to 1V. Unless I am mistaken that means more heat produced on the motherboard/GPU.
 

ChainedHope

Airflow Optimizer
Jun 5, 2016
306
459
Take my words as a grain of salt as I am not a PI engineer (I work as an SI) and its early in the morning but I don't think we are at the stage that we should be using 48v supplies for consumer electronics.

In the next 5-10 years? Sure. Not right now. Downing 12v to 5v, 3.3v, and 1v is fairly uncomplicated and cheap. The same cannot be said for 48v supplies. Sure its the same idea, but components have less tolerance to fluctuations meaning that things can go very wrong very fast if a consumer lives in an area with a bad power grid or in a house with cheap wiring. Along with this it brings the costs up by a good margin as you need better components to make it work efficiently. Take for example a cheap 500w 80+ (no metal rating) ATX powersupply. There is a good reason those dont cost more than $40-50 USD. They dont require the best components and can be made dirt cheap (those psus only cost ~$20-30 USD to manufacture, the rest makes up R&D, marketing, and profit after those are sorted out) but as we get higher in efficiency the price rises exponentially with 80+Titanium being about 10-20x more expensive for R&D and component costs than the 80+.

Now take this to 48V supplies. You need better over/under current/voltage and power surge protection, you need better down converters, you need better components in general. Now take the heat into consideration. Converting requires a loss of energy which translate into heat (pretty self explanatory) but its more you need to cool meaning a bigger heatsink, faster fans, etc... Sure efficiency is better on a large-scale like a supercomputer setup with multiple XEONs and GPUs in large clusters, but for a consumer device? Its not really all that feasible. You can make the argument that consumer electronics already use 48v supplies (A/V equipment) but its not the same complexity. Those supplies downvolt to a specific amount and that amount only for its use case, we arent splitting a 48v supply into 5v, 3.3v, and 1v. The most reasonable thought would be to split a 48V down into 4x 12V and split accordingly to spec, but you just created something that now requires 4x 48v-2-12v converters along with 4x the converters to go from 12v to 5v, 3.3v, and 1v if you dont plan on making each 12v connection only convert to 1 standard voltage.

Now then I can already tell you some of what I said is incorrect, but im going off some assumptions ive seen in the industry and my work with other PI's [Power Integrity Engineers] as well as my knowledge as an SI [Signal Integrity Engineer].

Also as im thinking about it, higher voltage wouldn't mean smaller wires. Voltage affects wire insulation and Current affects inductance. So sure you save on the copper, but your trading it off for more insulation around the wire. Its less insulation by far than what you would need for copper making the wires "smaller" but it wouldn't be that much noticeable for say 8-pins for example.
 
Last edited:

EdZ

Virtual Realist
May 11, 2015
1,578
2,107
I don't think there would be much increase in insulation thickness, most wiring is already specced for well above 12V (just because the cheapest materials handle more, so why spend more money to develop a less effective insulation?). But more than wiring, raising the voltage to reduce the current and I^2R losses means you can run more power over existing traces. 12V PCIe power over card-edge is 75W, upping to 48V gives you 300W to play with without needing to dramatically beef up traces or connectors. Being able to eliminate some or all of the internal point-to-point wiring in a system reduces part and assembly costs, which could offset some of the cost increase of more robust VRM filtering stages.
 

AleksandarK

/dev/null
Original poster
May 14, 2017
703
774
Last edited:
  • Like
Reactions: EdZ and Kmpkt

ChainedHope

Airflow Optimizer
Jun 5, 2016
306
459
I don't think there would be much increase in insulation thickness, most wiring is already specced for well above 12V (just because the cheapest materials handle more, so why spend more money to develop a less effective insulation?). But more than wiring, raising the voltage to reduce the current and I^2R losses means you can run more power over existing traces. 12V PCIe power over card-edge is 75W, upping to 48V gives you 300W to play with without needing to dramatically beef up traces or connectors. Being able to eliminate some or all of the internal point-to-point wiring in a system reduces part and assembly costs, which could offset some of the cost increase of more robust VRM filtering stages.

Valid point on the wires and the power loss. Like I said, it was early in the morning when I typed that up lol. The only other real issue I can see would be arcing but to be honest thats not much of an issue since the motherboard manufacturer would need to redesign their power systems anyways so changing some trace spacing wouldn't be much to add on top.

Valid points guys.

One thing to address is that the power conversion from 48V to 1V can be done VERY efficiently.
http://epc-co.com/epc/GaNTalk/Post/14229/48V-to-1V-Conversion-the-Rebirth-of-Direct-to-Chip-Power
This means that the whole standard way of building VRMs for processors could change slightly, while boosting overall efficinency to much grrater procentage than now.

It looks good but I'm not seeing pricing details and thats just one part of the puzzle. The slight change you mention could end up costing millions in setup fees for a factory to start churning out a prototype and more in R&D (VERY conservative estimate, it would be much higher). And with how slow the research groups work you wont see a good whitepaper or talk about a new VRM filter using this kind of technology in the commercial space for at least 5 years. Id expect it to get more common in the industrial and research sectors though.