6
\$\begingroup\$

A x16 PCI Express slot can deliver 75W for PCI Express Graphics Card. Some graphics card today also use external PCI Express power to increase above this limit, the most basic of these is the 6-pin PCI Express graphics power connector. This is specified to deliver 75W, bringing the total allowable power of the graphics card to 150W.

What I am unsure of is how does the graphics card balance it's load from these two sources of 12V power? Does it use a current-limiter on both sources or is the set up more complicated than that?

\$\endgroup\$
  • 2
    \$\begingroup\$ I wouldn't be surprised if most of this "trickery" was simply achieved by driving different parts of the board from the two different supplies, especially in multi-GPU cards. \$\endgroup\$ – Polynomial Nov 19 '12 at 10:49
  • 3
    \$\begingroup\$ @Polynomial Spot on. I've just traced through an nVidea GeForce 8800GTX: There's a power switching controller, but most of the card is divided into two distinct power districts. Almost none of the power other than data interface appears to be using the PCIe bus power, it's all running off the SMPS connector. \$\endgroup\$ – Anindo Ghosh Nov 19 '12 at 11:07
  • \$\begingroup\$ @AnindoGhosh Interesting! I've got a spare GTX260 at home which takes two separate 75W power connectors - I'll have to take a look at how that does it. \$\endgroup\$ – Polynomial Nov 19 '12 at 11:09
  • \$\begingroup\$ @AnindoGhosh Looks like the GTX260 does something similar. There are actually four power districts and three ground planes. Looks like they have some sort of current limiter between the two PSU cables which drives the fourth district, which looks as if it goes directly into the GPU. \$\endgroup\$ – Polynomial Nov 19 '12 at 18:35
6
\$\begingroup\$

Most of this, as you've probably seen from the comments, isn't actually electronic trickery. They run multiple power districts across the board and drive them from the PCI-e slot and the cables separately. They're all running on a common ground from the same PSU, so it works quite nicely. Usually you'll see the VRAM and peripheral logic running from the PCI-e slot supply, and the GPU itself running from the PSU.

For boards like the GTX260 OC2, which are explicitly designed for overclocking, you'll see cases where two 75W 4-pin PSU connectors go into the same card. When more than one is used, they do use some form of current limiting (can't tell what on my card - it's under the housing) to "combine" the maximum current of the two inputs. This apparently results in a maximum consumption of 225W rather than 150W, which can be useful for people insane hardcore enough to cool their hardware with liquid nitrogen.

Rather surprisingly, a little research indicates that computer power supplies are designed in a way that allows you to directly wire two +12V rails together to get a boost in maximum current, without any damage to the supply. I'm skeptical, but it looks like some nutcases on various forums have done it, and one company even sells a special connector that combines them. If it's a legitimate thing that the ATX spec says they should allow, then I guess certain cards may just join the rails together in a similar way. I wouldn't recommend trying it yourself, though.

\$\endgroup\$
1
\$\begingroup\$

It might be simpler than that given that both 12V rails emanate from the same source - the power supply. In which case it's a matter of parallel rails to take the current. However, yes, I'd also expect current limiting on both sources.

Edit: AnindoGhosh's comment is a better answer than this! I had just wondered if different parts of the board might be powered by the different supplies and the answer seems to be yes. Interesting.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.