[toadwart]

I had an idea, but don't know if it's technically feasible, or a pie in the sky idea. Unfortunately the unknown's are due strictly to my lack of knowledge. I'm not an Electrical Engineer, so I'm just throwing this out as an idea to mull over.

What I propose is to move the video card(s) outside of our existing CPU box? Via an extension cable of sorts?

We would create a PCI-x (PCI/AGP whatever) cable, with a slot at one end, and a plug at the other, that you could plug into the interface on the motherboard, and a run a cable outside of the box to a plug on the other end. So, your video card would now sit outside of your computer box, either open air, or in a secondary enclosure. Now this cable, plug, and slot would have to be manufactured, and perhaps the signal degradation over the contacts and cable length would make this a moot point. Is the data path length critical enough with the transfer loads between the GPU and the bus to make this too difficult? Essentially though, if possible to do, we'd be able to segregate the GPU from the rest of the heat sources in the box, allowing us to cool the GPU seperately from the rest of the system, and not adding CPU and GPU heat to the same environment. I understand that the rest of the system has to send data to the video card through the bus system in order for the video card to process the data and then display it on our screens, but are we at, or over the load capacity of the transfer bus? If not we could take some small transfer capacity loss in order to decrease our heat issues. This same idea could theoretically be applied to other portions of the system as well, and I assume that the system manufacturers have worked this all out before, but has it ever been re-addressed with newer technology, transfer pipelines, etc?

I don't know, most likely not possible, but if we could figure out how to do it, we could segregate the computer components quite a bit better, and as a result have better, more targeted cooling options for various parts, allowing for greater capabilities. Right now we're trying to cool different parts all jammed into the same area, and losing efficiency in cooling, and subsequent clocking abilities, because the cooling system on the CPU is fighting some of the heat from the HDD's, and the GPU's as well as the CPU heat, thus decreasing it's overall effectiveness at cooling the CPU.

[Adragontattoo]

seen the idea tossed around a few times but I have yet to see it as more then an idea.

[Nephewkp]

by using cables like that you would run into serious issues with signal delay and signal quality issues. Also considering the number of contacts for the various busses, the cables would not be very flexible and probably extremely expensive.

[BossBorot]

It would prob be more practical to solder longer leads from the mobo to the socket and replace the wires with copper, silver, or gold.

[Evilsizer]

as irrc there is lots of talk about dx10 cards being external with their own psu as well.

edit
here we go
http://www.tomshardware.com/2006/07/...ion/page4.html

[emboss]

Feasable, but too expensive at current tech levels. Basic problem is getting the data around. PICe x16 does 4 GB/s each way, which is about 40 Gb/s (each way). Take eSATA, which is rated at 3 Gb/s (one way) with a 2 meter cable. So you'd require 14 eSATA lanes each way - 28 total - to get sufficient bandwidth to high-end cards. While not impossible, a cable of sufficient quality (well shielded strands to reduce cross-talk especially) would not be cheap, and nor would the interface ICs at each end.

So yes, it is possible, but would probably only make financial sense for high-end cards where the cost of the interconnect would be a smaller part of the total price.

On the other hand, there's not a whole lot of point to it. We already have a nice, fast interconnect (PCIe) and centralising things like the power supply improves efficiency. The only problem at the moment is cooling since card manufacturers cannot rely on the case to provide cooling.

For example, a very efficient way of foxing GPU cooling would be to compartmentalise the PCI(e) slots away from the CPU and have a wind-tunnel like thing at the bottom of the case. A nice big 120mm fan could then force air through this tunnel and out the back of the case through heatsinks not dissimilar to current designs (fins running lengthwise down the card). Your GPU wouldn't require a fan at all in this case (and indeed, adding a fan would probably make it cool worse.

One problem would be the requirement for thermal feedback from the cards to the front fan to make it run quiet when not under load, and also some sort of thermal protection to protect the cards in case someone ran them in a non-compliant case.

Hmm, I might actually try something like this in my case build later this year ... might be hard getting the heatsinks though.

[Adragontattoo]

Emboss you may have just been hijacked! If I get either cubes up and running I may have to steal that idea and use it.