Sticking x16 Video Card in x8 slot

Zarathustra[H]

Extremely [H]
Joined
Oct 29, 2000
Messages
38,858
So we've known for some time that you can either cut/melt out the back of a smaller PCIe slot or cut off the leads on the video card so it will fit, resulting in the cards working in an x8 slot.

Usually this is done with video cards that use an PCIe power connector to pull their power, resulting in no power being pulled from the motherboard.

I was planning on doing this with a couple of Quadro 2000 video cards in a server motherboard.

The thing with the Quadro 2000 is that it doesn't have a PCIe power connector, and instead pulls all of its power from the motherboard, all 62w of it under load.

I just realized that PCIe x8 slots officially per the PCIe specification only provide 25W of power. (PCIe x1 slots only provide 10w). The full x16 slots provide 75 watt.

I'm guessing there is a risk of a problem here if the video cards pull too much power over the 8x slot, depending on how the motherboard is designed, if they overprovisioned the power to the 8x slots at all.

Does anyone have any thoughts about this?

I only plan on using them for video playback at 1080p, so I may not even ever hit the 25w limitation of x8.

Does anyone have any thoughts regarding this?

Much Obliged,
Matt
 
It won't work. You can't reasonably expect the card to stay under 25w of power use even if you try going easy on it. And that is presuming the card will even boot without the power connections an x16 size slot would provide.

They make riser cables to attach a 16x card to an 8x slot AND provide power from a molex connector. Look for one of these:

http://www.moddiy.com/products/PCI%...der-Cable-w{47}Molex-%2b-Solid-Capacitor.html
 
Last edited:
Should work fine, all of the power comes from the small finger before the key which is all pinned no matter what size the slot is. The rest is I/O lanes.
 
Should work fine, all of the power comes from the small finger before the key which is all pinned no matter what size the slot is. The rest is I/O lanes.

I'm aware of this, but if they followed the PCIe spec to the T, the fact that they only installed x8 slots would have allowed them to only provide enough power to support 25W per slot.

I know that only the first few contacts provide the power, but the power delivery circuitry (caps, etc) supporting the slots may not be able to max out 75W as expected for an x16 card per slot, if they are only designed for x8...

Now they could have done this in many ways. Its a Supermicro X8DTE so it has 6 8x slots.

They could have pooled it, providing 25*6 watt to be shared among all the slots. They could have been sticklers to the spec, in which case going over 25W for any given slot would be a problem, or they could have added a nice safety margin, maybe as high as the 75w per slot x16 slots would use just to be safe, considering how.many different passive risers there are out there allowing 16x cards to be inserted in x8 (or smaller) slots.

The devil here is in the details, and unfortunately the manual is not this detailed.

I have sent an email to Supermicro support. Lets see if they are willing to find me.an answer for an older product like this...
 
The most likely scenario is that you will be able to use one of the quadro 2000's. I see that mb has a pair of 8-pin 12v connectors. So it should have ample power capacity. The only way it won't work is if the pci-e slots are fused to protect from drawing more than spec power. I doubt they are, but it is possible. Lately I have seen mb's with fused USB headers, but I don't believe I have encountered one with fused pci-e slots.

BTW, even those cards without an auxiliary power connector should only draw 25w at boot, I believe. I think that's what the spec dictates. After initial boot / post, then they can draw more. I suspect this was added specifically for this type of situation, where a card will be drawing above spec power. I have owned mb's that only came with one 16x slot but still supported SLI. I think the issue is total power draw through the mb not exceeding its power envelope. How that draw breaks down is not as critical.

As was mentioned, if you want to be absolutely sure there is no problem you can use a riser with a molex connector. Those are inexpensive.
 
Your fine. The card will just down grade your ego status to lowbie lanes and you will get less performance.

Just ram it in the slot, turn on the system, and play some Fallout 4.
 
Your fine. The card will just down grade your ego status to lowbie lanes and you will get less performance.

Just ram it in the slot, turn on the system, and play some Fallout 4.

Well, that IS the way it works with any videogames that uses a 12v PCIe power connector. They generally.pull next to no.power from the motherboard. With lower end boards that don't have the connector, t hey pull all their power from.the motherboard, and the. You have to pay attention to these things.

These are not for gaming though. They are going in an ESXi server and will be direct forwarded to guests for emote video display using HDMI over Ethernet adapters.

If I didn't have to do direct forwarding I'd just get a set of gt 720's and not worry about the power at all. Unfortunately consumer geforce cards don't direct forward (Nicosia intentionally sabotaged this). AMD cards do, but I don't like their video acceleration performance under linux. Very buggy.
 
The most likely scenario is that you will be able to use one of the quadro 2000's. I see that mb has a pair of 8-pin 12v connectors. So it should have ample power capacity. The only way it won't work is if the pci-e slots are fused to protect from drawing more than spec power. I doubt they are, but it is possible. Lately I have seen mb's with fused USB headers, but I don't believe I have encountered one with fused pci-e slots.

Interesting. When you say fused, are these old fashioned fuses that burn out?

In other words, if I try it and trip the fuse, am I looking at a paper weight until I find out how to replace a fuse, or is it something that just trips a reset to protect components leading to an improper shutdown and reboot?
 
Well, I looked up the specs of all the other expansion cards I have installed.

If I total all of them up, along with the two planned Quadro's I wind up with a total of between 158 and 166w.

If the power is pooled over the six 8x slots, that means I have a total of 150w to work with, plus whatever safety margin they built in.

It's close enough that it might just work, especially since the Quadro's will never see full load...
 
Zarathustra[H];1042066171 said:
Interesting. When you say fused, are these old fashioned fuses that burn out?

In other words, if I try it and trip the fuse, am I looking at a paper weight until I find out how to replace a fuse, or is it something that just trips a reset to protect components leading to an improper shutdown and reboot?

For USB protection I have seen polyfuses used - https://en.wikipedia.org/wiki/Resettable_fuse

Here is an image of a raspberry pi with a pair of polyfuses (it was used on earlier revisions, newer ones have done away with them).
1.jpg


If the pci-e slots are fused with polyfuses, it will reset itself once current has been removed from the circuit. I don't think a mobo manufacturer would use an old school style fuse that would need to be replaced once blown.
 
For USB protection I have seen polyfuses used - https://en.wikipedia.org/wiki/Resettable_fuse

Here is an image of a raspberry pi with a pair of polyfuses (it was used on earlier revisions, newer ones have done away with them).
1.jpg


If the pci-e slots are fused with polyfuses, it will reset itself once current has been removed from the circuit. I don't think a mobo manufacturer would use an old school style fuse that would need to be replaced once blown.

Thanks, thjis is very helpful.

If anyone is interested, I have continued a broader discussion on my entire project over here.
 
Back
Top