3.3 & 5 Volt Power - What we really need?

[ejwinters]

Considering today's systems, starting with Core2 Duo and Quads, and moving on to i5 and i7, what kind of power requirements are necessary for th 3.3 and 5 volt rail. I noticed the bigger the supplies get, the more amps are available on each of those lines, but how much is really necessary? What things use those 2 voltages and what is the "limit" of what is really needed?
Do we really need 22A each on 3.3 and 5 volt rails (recent PS i looked at).

Secondly, How much amps does JUST a video card draw, like a gtx280 or above? They tell you to have say 45Amps on the 12 volt rail, is that just for the card or for everything? If its just the card then in systems with 4 12 volt rails @ 25A, that 25A on one rail should be enough for 1 video card right?

How many amps do the drives, both HD and DVD use? I'm considering a raid 1+0 setup (4X1TB) and would like to have 4 DVD burners as well.

[Hondacity]

45a on the 12v...sounds like the total current capability on the 12v rail..

12v devices - video cards..mobo(eps and 24pin) hard drives

5v devices - ssd...some go wild on this now...

3.3v...used on some signalling stuff

c2d c2q going to core7...etc...you just need more 12v power

[jonnyGURU]

+5V is used on any hard drive and not just SSD. The logic boards on HDD's use +5V.

And +3.3V is typically used for memory.

[blivit4]

And +3.3V is typically used for memory.

Older types of memory. DDR2 & 3 are well under 3.3v and are powered from VRM's that probably use 12v as their input.

[370forlife]

Older types of memory. DDR2 & 3 are well under 3.3v and are powered from VRM's that probably use 12v as their input.

Don't know why they would use 12v, that would just put more stress and create more heat on the vrm for the memory.

[jonnyGURU]

Yeah. As far as I know, they're still using VR's that regulate from 3.3V to obtain 1.8V, etc.

[Stefan Payne]

On the Core I7/9xx Series, the 3,3V rail is used to power the northbridge inside the processor.

Besides that, you don't need 10A on the minor rails in a normal (AMD based) computer.
I don't know it's on a LGA1366 Rig...

here the figures (http://www.forum-3dcenter.org/vbulletin/showpost.php?p=7467888&postcount=14), beware: It's in german ;)

[blivit4]

Don't know why they would use 12v, that would just put more stress and create more heat on the vrm for the memory.

By that logic the CPU power would also be derived from 3.3v, but it isn't.

DC-DC converters aren't resistor divider networks, or basic 3 lead regulators that dissipate a lot of heat.
Power is still power, so as voltage goes up the current goes down for a given number of watts.

[ejwinters]

Ok then,so how many AMPS are really required on the 3.3 and 5 volt rails? That was the original main question, which seems to have digressed, lol.:wall:
I mean really if only say 15 amps on each is required, and these PSU companies keep diverting more 12V amperage to those lines (20-30 amps each), thats quite a waste when they could save it for the 12V line!

[Makalu]

well it'll vary from system to system but I figure the typical single card system only uses about 5A each of 5V and 3.3V power. My system is an Asus P5Q E8500 and HD4870 with three hard drives and two optical drives and the Gigabyte Odin PSU measures about 4.5A draw off both...it doesn't change much regardless of idle or running Prime 95 and Furmark.

I had an 8800Ultra card in it previously and it drew about a half amp more off the 3.3V rail.

You can see some measurements of various systems done here:

http://www.xbitlabs.com/articles/coolers/display/system-wattage.html

you can see in the tables they are ~5A or less.

These are all single card systems and somewhere I saw an SLI motherboard being measured and the 5V and 3.3V usage nearly doubled with the second card installed (due to some 3.3V current used by the card but mostly components on the board apparently)...so maybe figure as much as 15A each in a tri-SLI setup? not sure...

[Travis]

By that logic the CPU power would also be derived from 3.3v, but it isn't.

DC-DC converters aren't resistor divider networks, or basic 3 lead regulators that dissipate a lot of heat.
Power is still power, so as voltage goes up the current goes down for a given number of watts.

The lower the difference between input/output, the higher the efficiency of DC/DC conversion.

[Travis]

If you don't use many HDDs you don't have to care about those two rails.

[370forlife]

By that logic the CPU power would also be derived from 3.3v, but it isn't.

DC-DC converters aren't resistor divider networks, or basic 3 lead regulators that dissipate a lot of heat.
Power is still power, so as voltage goes up the current goes down for a given number of watts.

No, cpu gets the majority of it's power from the 4/8pin cpu power connector, memory gets it's power from the 20(+4) connector.

I'm pretty sure the cpu and memory have different VRM's. At least on my old ECS RS482-M754 it did when a MOSFET for the ram exploded when attempting a overclock on the ram. CPU was stock 2.2ghz.

[Stefan Payne]

Yes but you can toast CPUs with an integrated memory controller with too high memory voltages.

[Zero82z]

I'm pretty sure the cpu and memory have different VRM's. At least on my old ECS RS482-M754 it did when a MOSFET for the ram exploded when attempting a overclock on the ram. CPU was stock 2.2ghz.
Of course the CPU and RAM have their own VRMs. They run on completely different (and independent) voltages.

[blivit4]

No, cpu gets the majority of it's power from the 4/8pin cpu power connector, memory gets it's power from the 20(+4) connector.


The CPU gets its power from the VRMs. The CPU VRMs get their 12v from the 4/8 pin power connector. The memory VRMs get their power from the 20/24 pin connector. It doesn't matter which connector is used if it's all 12v. The fact that something blew on the memory VRM could have no effect on the CPU VRMs even if they both use 12v.

The most probable reason for using 12v to derive the CPU voltage is that fewer pins/lighter gauge wire is needed to go between the PS and the MB. If 5v were used, the the current on the cables would be more than double. This matters because there is resistance in the wires between the PS and the MB and on the traces on the MB running from the connector to the VRMs. The result is that the voltage drop due to resistance would more than double.

For example, assume .01 ohm of resistance and the CPU is using 120w.

From W = V x I, I = W/V. Therefore: 120/12=10A
From V=I x R, 10 x .01 = 0.1V drop

With 5V, the voltage drop due to conductor resistance is 0.024V.

A 0.01V drop is 0.08% of 12v, but a 0.024 V drop on 5v is 0.48%.

Whatever efficiencies are gained by keeping the input and output voltages closer are counterbalanced by the higher cost of heavier gauge wire/larger traces on the MB.

[brian c]

Using +12V line for power hungry circuit can also reduce the loss across PSU cables due to I2R. When powered from higher voltage, it requires lower current.