HALP! System won't boot with GPU powered

[Josh | NFC]

Guys, this is really weird. SO weird that I have to ask for help, and I know alot of you are wizards at this kind of thing...

Right now I have five identical GPUs I am trying to test. They all work flawlessly in my standard bench.

I have four Z170 boards in the office, two from Asus, two from Gigabyte.

Three of them work great with the GPU I am trying to use, just as they should. One of them, the Asus Z170-S, does not.

The board in question will NOT power on with the GPU installed. If I plug the HDMI cable to the IGP, and UNPLUG the power connector to the GPU, it will boot to Windows.

ALL other GPUs in my office boot fine in the board.

Things I have tried:

-Making sure the card is in UEFI mode

-Disable SecureBoot

-Doublecheck CSM

-Tried multiple powersupplies and cable sets and dedicated a PSU to the GPU.

-Cried in a corner

Things I have not tried:

-Update the BIOS (I have the latest version time of this writing).

-Giving up.


I want some crazy ideas on how to fix this. Supply more voltage to the GPUs? Switch to PCIE v1? Eat a bucket of icecream? I don't like weird unsolvable problems.

 

[jØrd]

said board powers on w/ any of the other identical cards installed?
The GPU that wouldn't boot works fine in any of the other 3 boards?
Have you retested w/ a different GPU in said board, on the off chance something has changed w/ it since last time you used it w/ a different GPU?
Does said board boot w/ a not-GPU in the same slot?

 

[Phuncz]

If the board doesn't even power on (fans not spinning or twitching a moment), this isn't BIOS-related and more likely hardware-related.
Check the PCIe connector for debris or try a different port.

EDIT: for reference, picture of the board in question:

the board doesn't have another power connector to give additional power to the PCIe connectors so that's not it either.

Have you plugged in the PEG 8-pin on top ?

If you have any other PCIe card (network, sound, SATA, etc.), you could try plugging it in the gray PCIe slot, even if the PCIe connector is shorter. If that works, you atleast know some part of the slot is working correctly.

 

[Josh | NFC]

Thank you for the replies!!!

said board powers on w/ any of the other identical cards installed?

Nope! It will not not post with any combination of the identical GPUs in any PCIE slot if they are plugged into the psu.

The board will boot no issue whatsoever with my 1080, and even 1070s in SLI!

The GPU that wouldn't boot works fine in any of the other 3 boards?

To clarify, all five of these GPUs work individually in every other board I have tested in the office--they work great in the other Z170 boards.

Have you retested w/ a different GPU in said board, on the off chance something has changed w/ it since last time you used it w/ a different GPU?

Yup! I tried two slot-powered GPUs and a bunch of 10 series Nvidia GPUs. All work great!

Does said board boot w/ a not-GPU in the same slot?

Great suggestion! I didn't think to because it booted with a 1070 in there. I just came back from trying this and yes, it boots with a 8x SDI capture card in there and windows recognizes it.

If the board doesn't even power on (fans not spinning or twitching a moment), this isn't BIOS-related and more likely hardware-related.
Check the PCIe connector for debris or try a different port.

The CPU fan does indeed twitch slightly when I hit the power button--the first time. I need to drain power for it to twitch again. I tried several power supplies, and even a second set of PSU cables for my main PSU.

the board doesn't have another power connector to give additional power to the PCIe connectors so that's not it either.
Have you plugged in the PEG 8-pin on top ?

Yeah, I was hoping it was something stupid like that but unfortunately everything is all plugged in. Keep in mind the system benches absolutely fine with my retail cards--just NOT with these particular GPUs I'm supposed to be testing...

If you have any other PCIe card (network, sound, SATA, etc.), you could try plugging it in the gray PCIe slot, even if the PCIe connector is shorter. If that works, you atleast know some part of the slot is working correctly.

I gave that a scoot and it works great. Works with 1070, 1080, 750ti, and a 430GT too.

 

[Phuncz]

Keep in mind the system benches absolutely fine with my retail cards--just NOT with these particular GPUs I'm supposed to be testing...

So I guess you're testing pre-production cards. This could mean incompatibility. What CPU are you using ? It might be caused by that. Since PCIe lanes for GPU are communicating straight with the CPU, it might be an incompatibility and either the CPU may need a microcode update (unlikely, included in BIOS updates) or the issue lies in the GPU's BIOS (likely).

Have you tried the card in the other PCIe slots ?

 

[EdZ]

So just to make sure I have things correct:

- Multiple cards of GPU X DO NOT allow the Asus Z170-S to POST IF the PSU is connected to the GPU
- GPU X DOES post in the Asus Z170-S IF the PSU is disconnected form the GPU
- GPU X DOES POST on other boards with the PSU connected to it
- This behaviour occurs in all PCIe 16x slots of the Asus Z170-S
- Other PCIe cards (including 10xx series GPUs, a bus-powered 750ti and a bus-powered capture card) DO NOT have and issue regardless of whether the PSU is connected to the card or not

If all these are correct, one possible issue could be ground linking (or voltage mismatch) problem on GPU X (or on the motherboard, if other cards just happen to be working around the issue, which is not unheard of) that the Asus Z170-S is having an issue with but other boards are not. To test this; try powering GPU X with a completely separate PSU from the one powering the board.
A more comprehensive but much riskier option would be use Kapton tape to mask off parts of GPU X's PCIe header to isolate it from 12V, 3.3V, GND, or a combination of the three. This risks blowing either the GPU, motherboard, or both, so is probably not worth it unless you have plenty of spares and GPU X's vendor is willing to pay you for the replacements.

Phuncz's suggestion could also be correct, though I'd expect the same CPU in a different motherboard to have the same issue in that case. If you happen to have a PCIe riser to sacrifice by removing the tab at the end of the slot (or willing to modify the slots on the board), you could try installing GPU X into one of the PCIe 1x slots that are connected to the PCH and see if the board POSTs. That could differentiate between a problem with the PCIe bus between the CPU and GPU, and a problem with the wiring of the motherboard itself.

 

[Josh | NFC]

So I guess you're testing pre-production cards. This could mean incompatibility. What CPU are you using ? It might be caused by that. Since PCIe lanes for GPU are communicating straight with the CPU, it might be an incompatibility and either the CPU may need a microcode update (unlikely, included in BIOS updates) or the issue lies in the GPU's BIOS (likely).

Have you tried the card in the other PCIe slots ?

I thought from the beginning that it was a BIOS/UEFI issue (btw thank you for the detailed explanation) but before I "gave up" I wanted to be sure I wasn't doing something stupid. I was assured the system would be compatible and people are expecting it.

I did try the GPUs in all the compatible slots, for funsies, but no-go. No options for it in the UEFI. I am wondering if ASUS has some power tuning options for the PCIE slot that I could play with, but everything in that regard looks above me.

So just to make sure I have things correct:

- Multiple cards of GPU X DO NOT allow the Asus Z170-S to POST IF the PSU is connected to the GPU
- GPU X DOES post in the Asus Z170-S IF the PSU is disconnected form the GPU
- GPU X DOES POST on other boards with the PSU connected to it
- This behaviour occurs in all PCIe 16x slots of the Asus Z170-S
- Other PCIe cards (including 10xx series GPUs, a bus-powered 750ti and a bus-powered capture card) DO NOT have and issue regardless of whether the PSU is connected to the card or not

Yes, you flawlessly stated the situation! I did try using a HDPLEX to power the GPU separate from the motherboard, but now I am going to try your two advanced ideas...

wish me luck.

Thanks guys, I really appreciate you.

 

[Josh | NFC]

Well, my office smells like burnt copper, para-cord, and crystalized rubber now. Just like old times...

 

[Phuncz]

I did try the GPUs in all the compatible slots, for funsies, but no-go. No options for it in the UEFI. I am wondering if ASUS has some power tuning options for the PCIE slot that I could play with, but everything in that regard looks above me.

Asus does have/had some wonky active voltage protection in the BIOS (something like Anti Power Surge), it gave some trouble to people with PSUs during heavy load.

Well, my office smells like burnt copper, para-cord, and crystalized rubber now. Just like old times...

Ow boy, that sounds bad.

 

[Josh | NFC]

Hooked everything up as suggested, powered on system then...

PSSSSHHHHHHHHHHHHHHHHHHHHH POP pop pop POP

Flame shot out of my power cables and erupted in smoke. Smoke alarm didn't go off, which actually scares me. Thanks Kiddle.

I dived for the breaker switch, but before I made it I heard an unexpected sound...

*BEEP*

Wait, was that...a post peep?

I unplug everything and discard my charred and now brittle cables.

I sit down for a moment, stunned, and wondering just how much damage I did.

PSUs look good, they don't smell bad...well, they smell better than everything else in the room now. Connectors on the GPUs look good too despite the cable side pulling a T1000.

Am I crazy enough to set everything back up like normal? Check.

I go for broke because, screw it, that's why. I put two GPU-"X"s in connected to the same PSU as the motherboard. I flip the breaker.

... nothing

...
...

*BEEP*


You've got to be kidding me...

Dr. EdZ...you just defibrillated my video cards.

 

[Phuncz]

LOL this is one epic thread !
Kids don't try this at home !

But what was damaged in this process ? Because it seems like it was the cause for the issue apparently.

 

[EdZ]

Bwahahahah! IT'S ALIVE!

My guess from the other side of the planet with a text-only description: somewhere, probably on the GPU, is a polarised smoothing capacitor installed the wrong way. Well, WAS installed...

 

[jeshikat]

Are they after-market or custom-made cables? It seems strange to me that they failed so spectacularly without damaging the connectors on either the GPU or PSU yet the current draw wasn't enough to trip the OCP in the PSU.

 

[Josh | NFC]

But what was damaged in this process ? Because it seems like it was the cause for the issue apparently.

Apparently the motherboard? Because it works now with all the GPUs?

I really don't know...if I wasn't so pressed to meet my deadlines I would be in a straight jacket as this is the second weirdest thing I have ever seen building PCs.

My guess from the other side of the planet with a text-only description: somewhere, probably on the GPU, is a polarised smoothing capacitor installed the wrong way. Well, WAS installed...

Hahahaha! Yeah something like that...

seriously though if you think of why the board and cards love each other now let me know...

Are they after-market or custom-made cables? It seems strange to me that they failed so spectacularly without damaging the connectors on either the GPU or PSU yet the current draw wasn't enough to trip the OCP in the PSU.

Aftermarket. I had a set a primochill cables that were long enough for my new workbench. Check out this one--I find it fascinating how the cable made these little loops that burst the sleeving. Electricity is fun! Check out the PSU side connector...barely anything left.

 

[EdZ]

Holy overcurrent, Batman! I'm amazed the PSU failed to trip OCP before it could pump enough to burn out the actual cables. In fact, that might be what the PSU you were using previously was doing: tripping that 12V rail before it could fry anything.
I'm really thinking a reversed smoothing cap is to blame. It would essentially be acting as a short-circuit across a power bus somewhere. When the new PSU dumped UNLIMITED POWEEEEER through that 'short', it overloaded and popped the cap and turned it into an open circuit (and protected the components behind it in doing so), leaving the overall circuit working but just with less filtering.
It looks like the lead that blew up was pin 4, which is the voltage sense pin. Which is doubly-weird: Some of AMDs recent cards did silly things with the Sense pin, but that would result in things blowing up on the GPU, rather than the motherboard.

 

[jeshikat]

What model power supplies did you use?

 

[Josh | NFC]

Silverstone platinums. I've culled everything else because I love the soft connectors they use--easy to unplug their cables.

Well, those and HDPLEX/picos of course.

The cables blew up while using a Silverstone 700 sfx-L plat.

For anyone reading this on the outside--read the whole thread--it wasn't the PSU fault I do t think.