External PCI-E slots or cables...

[Stu L Tissimus]

The other day, I had a pretty silly idea - While brainstorming on what useless things I could add on to my video card to help it dissipate heat faster, I started thinking, "why should it stay inside the case?" The air inside of the computer is already warmed up by the CPU and bridges and such.

Now, theoretically - not that I'd actually do it - let's say I took some strips of copper wire and soldered them to the pins in the PCI-E slot, and then soldered them to the corresponding ends on a video card. But the video card would be located outside of the case, where the air would be cooler. Would such a set-up work, or do you think the resistance in the copper wire would slow down data transfer?

(I have next to no clue about electronics. Does the distance of one thing to another make much of a difference when you're using something with as low resistance as copper?)

(Distance, difference, resistance.... Cool, I can make rhymes.)

 

[Matthias99]

Would such a set-up work, or do you think the resistance in the copper wire would slow down data transfer?

(I have next to no clue about electronics. Does the distance of one thing to another make much of a difference when you're using something with as low resistance as copper?)

Short answer: yes. This is a major problem in IC design as clock speeds start to up into the megahertz/gigahertz range.

It is unlikely that you would get fast/reliable enough signal transfers for the PCIe device to work reliably in such a configuration. Not only would doing this significantly increase the length of the wires between the card and CPU/northbridge, but the extra connections and soldering would make the connection much noisier, and it would be much more vulnerable to EMI unless you shielded the cables. You could also have clock skew problems if the wires are even slightly different in length.

But hey, go nuts. Let us know if it works.

 

[sao123]

Originally posted by: Stu L Tissimus
The other day, I had a pretty silly idea - While brainstorming on what useless things I could add on to my video card to help it dissipate heat faster, I started thinking, "why should it stay inside the case?" The air inside of the computer is already warmed up by the CPU and bridges and such.

Now, theoretically - not that I'd actually do it - let's say I took some strips of copper wire and soldered them to the pins in the PCI-E slot, and then soldered them to the corresponding ends on a video card. But the video card would be located outside of the case, where the air would be cooler. Would such a set-up work, or do you think the resistance in the copper wire would slow down data transfer?

(I have next to no clue about electronics. Does the distance of one thing to another make much of a difference when you're using something with as low resistance as copper?)

(Distance, difference, resistance.... Cool, I can make rhymes.)

Why not just build your computer in an open case?

 

[Bassyhead]

Originally posted by: sao123

Originally posted by: Stu L Tissimus
The other day, I had a pretty silly idea - While brainstorming on what useless things I could add on to my video card to help it dissipate heat faster, I started thinking, "why should it stay inside the case?" The air inside of the computer is already warmed up by the CPU and bridges and such.

Now, theoretically - not that I'd actually do it - let's say I took some strips of copper wire and soldered them to the pins in the PCI-E slot, and then soldered them to the corresponding ends on a video card. But the video card would be located outside of the case, where the air would be cooler. Would such a set-up work, or do you think the resistance in the copper wire would slow down data transfer?

(I have next to no clue about electronics. Does the distance of one thing to another make much of a difference when you're using something with as low resistance as copper?)

(Distance, difference, resistance.... Cool, I can make rhymes.)

Why not just build your computer in an open case?

Yeah, fix the simplest problems first. That or increase airflow within the case.

 

[harrkev]

In theory, what you state CAN be done (within limits), but it simply cannot be done "ghetto style."

You would have to match the imedances of the traces, and there are also things to watch out for, like cross-talk, as well as the increased impedances of the power traces. But, if you are careful, you could make an extender card that could move your AGP card. I would not trust any sort of cable, though.

A general rule of thumb is that a signal travels on a board 8 inches per nanosecond (actual value depends on the stack-up and dielectric constant of the board material, but 8 inches is a pretty good rule of tumb). So, you can extend everything 8 inches and increase delay by two nanoseconds (one there, one back). A 66MHz clock (typical of an AGP port has only 16.6 nanoseconds between clock edges. So, that means that the clock edge is generated by the mobo, the clock has to travel to the graphics card, which will cause the outputs to the AGP interface to change, and those changed signals have to travel back to the nothbridge in 16.6 ns. The question is: how much extra delay can exist in this path before you start violating your setup times at the northbridge? My GUESS is that you might have a nanosecond of margin. So, assuming 1ns of margin, you can extend your AGP slot by four inches.

So, the short answer is that YES, it can be done. But it will be expensive, and you cannot move it much.

**EDIT**
On the other hand, Cypress has a line of "Roboclock" chips that can change the phase of a clock using a PLL. These could be used to get back some of the delay in forwarding the clock, but then you have to carefully look at the setup time at the vid card to make sure that you do not violate timing there.

This discussion assumes that you are using AGP. If you are using PCI Express, that is a serial based protocol. I would guess that you could extend that by feet without problems. Since it is a serial protocol, there is not way to get a "round trip" on a clock edge, so the clock is likely sent both directions with the data (or the clock is recovered from the data, like what is done in FibreChannel or Ethernet). But I have no read the specs on PCI Express.

 

[BrownTown]

it would best be done by using an asnchronous link in which the CPU sent rendering information to a large input buffer on the video card, and requests by the video card were handled in a imilar manner. I'm not usre if the added delay would be noticile to the human eye though. Also, in order to work best you would have to have a large on board cache, and a large amount of video memmory so that you had to wait for textures as little as possible, and could reduce the bus traffic to a reasonable amount.

This would of course require two cards, the actual video card, and then a smaller PCIe card which could talk to the vido card over the asynchronous link. Basically its just way to expensive and slower to boot, so people would jsut use water cooling to solve the heat problem for alot less money, and alot better performance.

 

[Peter]

There is a standard for PCIE-over-cable, but that requires the mainboard to bring a plug, or a PCIE bridge card in the slot.

Homebrew? Nah. You are aware you are talking about 2.5 GHz signalling?

 

[imgod2u]

If I remember the Intel presentations for PCI-E, this kind of design was one of the intentions of PCI-E. I'm not sure exactly what happened to that idea, perhaps it's still being worked on. That being said, it's possible, although I'm not sure if you have the correct equipment at home to do it. Some of the advantages of PCI-E was that it is a high-speed serial link and so you eliminate the issues of clock and data skew. You still have a bit of crosstalk across your wires (assuming you're using a 16x PCI-E with a link in both directions) but that can be solved by using decoupling capacitors and by impedence matching. Truth be told, I have no idea how to get a 2.5 GHz signal across a few feet. The most I've ever done is get that kind of signal across a few inches on a PCB. But then again, I'm not Intel. I'm sure it can be done. LVDS probes are made that can sense signals up to the 10GHz and they transmit that sampling data over to the oscilliscope over pretty long wires so it must be possible.

 

[NeoPTLD]

It has a lot to do with the speed of electron flow. The electrical signals do not travel anywhere near the speed of light.
http://www.jimloy.com/physics/electric.htm

 

[MrDudeMan]

Originally posted by: NeoPTLD
It has a lot to do with the speed of electron flow. The electrical signals do not travel anywhere near the speed of light.
http://www.jimloy.com/physics/electric.htm

Why does it have a lot do to with the speed of electron flow? The potential is the result of the electric field, which does infact propogate at c. The electrons don't need to move anywhere close to c for this to be true.

 

[CrispyFried]

Originally posted by: Bigsm00th

Originally posted by: NeoPTLD
It has a lot to do with the speed of electron flow. The electrical signals do not travel anywhere near the speed of light.
http://www.jimloy.com/physics/electric.htm

Why does it have a lot do to with the speed of electron flow? The potential is the result of the electric field, which does infact propogate at c. The electrons don't need to move anywhere close to c for this to be true.

IIRC, the speed of the actual electrons in a conductor is only a few feet a second, you can run faster.

anyway as to the original question, many computers used (and maybe still do) riser cards for pci cards (a card that plugs into a standard pci slot and has several pci slots on it, it basically turns the cards 90 degrees and was used on slim case computers). the riser card wasnt anything fancy, just a few caps and resisters, not active stuff at all. and some company (cant remember) makes pci expansion cages that gives you an external case with a bunch of pci slots, but I cant remember exactly how it was connected to the computer its used with. anyway that was pci, so I dont know if that can be done to pci-e. if it can though I wouldnt trust a cable, as someone already mentioned, but a pc board might do it. might need some active electronics to keep everything happy though.

 

[Peter]

You can't make many PCI slots from one PCI slot without using a PCI-to-PCI bridge chip. The reason is that a PCI slot is a device position, not the entire bus (unlike ISA).

Those old boards you mention had a special riser slot that was NOT suitable for anything but its mating riser card.

VIA's EPIA boards make their PCI slot a bit "special" so their simple riser supports two cards - at the expense of the board's own slot not being quite as PCI compliant as it looks.