ncrmnt

Try putting NVreg_EnablePCIeGen3=1 to /etc/modprobe.d/nvidia.conf Might well do the trick. I didn't have the moment to test that one yet.

Hey folks, I think I found a way to play on these cards without the need for integrated graphics and/or driver patching. It looks like the linux driver is not crippled in any way, so you can easily start X11, connect to it using x11vnc, install steam and use remote play/in-home streaming for you convenience. I've thrown up a howto at my blog: https://ncrmnt.org/2019/08/04/linux-gaming-with-p106-100/ The problems - SLI & NVENC do not work. Did anybody manage to get NvENC working on these cards? (at least in windows). NvENC is the only thing missing to make it an awesome headless gaming rig. As for SLI, it seems that the virtualized Q35 chipset is not certified by nvidia. While windows drivers may be patched for SLI, I haven't found any info on how to fix 'em that in linux.

Link, please Either they are gurus or scammers. Something tells me it's more likely the latter

Thanks a lot for clarifying this point. If that's the case, there are no resistor straps I hoped for, and the only way we can make this work - hack'n'patch nvidia drivers OR use nouveau driver ;(

Hm... And you have the GPU-Z screenshot of the P106-100 with the GTX1060 BIOS?

Don't lose faith! Thanks to your experiments we've learned quite a lot on how this thingy works. First of all, if it booted, the pinout is compatible and it's a complement on your great soldering skills on its own! Second, can you post at least the resources tab from devices manager (or gpu-z) screenshots with different vbioses? (E.g. with original 1060 and p106-100 vbios) I wonder if flashing different vbioses will affect these in any way and/if they change. Or, at least if they look different than on a usual P106. The error 43/10 you've got are probably due to your modification to vbios file with the hex editor. Don't bother modding bioses, since, as I've said they are all digitally signed and tamper-proof. Driver detects that an refuses to start. The only way is to flash either the original GTX1060 or the original P106-100 vbios and give them a shot. On a side note: There's also a way to spoof pci IDs, when the card is being passed through to the virtual machine. This way we can try to install regular 1060 driver. But this will require you setting up a linux machine to host a VM and will require hardware with working IOMMU. (Process not for the faint-hearted) One of the things I was going to try myself once my P106 arrives.

AFAIK it's useless to mod it. It has a digital signature, like a checksum. If you replace a string, driver will know the rom has been tampered and refuse to work. All we can do is swap for another vbios from the collection. According to this: https://www.techpowerup.com/gpu-specs/nvidia-gp106.g797 GP106-100-A1 and GP106-400-A1 look roughly the same thing and work with the same DDR. Can you boot some live linux distro (e.g. ubuntu or debian) on a test rig with internal graphics? I'll tell you what commands to put in the terminal. The interesting stuff is, whether the PCIe card is visible on the bus after your surgical operations and what VID/PID it exposes (just type lspci in the terminal app). On previous generations VID/PID were set by resistor straps. If this is still the case and it works - we'll see a different VID/PID, hence we'll need a different VBIOS from some 1060. As a side effect, if it is visible on the PCI bus you can flash those VBIOSes via linux nvflash. That will save you some time on desoldering the VBIOS chip every failed attempt. In other words - don't trust three beeps from bios! They lie! Linux has some more awesome low-level tools to play with and gives some meaningful errors when things go wrong. Not like "error 43, go figure out yourself or ask our support drone"

DoctorVGA, it looks like you are way ahead of me, I'm still waiting for my P106-100 to arrive. Had to register here just after I saw your comment First of all, can you boot an actual system and see if the modded card actually is present on the pci bus? An output of lspci (linux) would be really awesome, as well as giving it a spin with regular nvidia linux drivers. Just Xorg.0.log would prove very useful. My old GTS450 soft-modded into Quadro 2000 to enable GPU passthrough never got detected by BIOS, I was getting three beeps, but worked flawlessly when I booted the VM and installed drivers. So if it's not detected by BIOS it may not be a complete failure yet! Following the the above, can you backup/flash VBIOS? If so, you can try some different VBIOSes from the net, see if any of them actually works. Finally, you flash (or transplant the ic) P106-100 BIOS, and see if it detects and works as if it were P106-100. If I got things right, we have a limited set of options here: Either we have some resistor straps, like the previous generations, and the only problem you have is different GDDR/timings. Therefore a VBIOS from a working 1060 with the same GDDR is required for things to finally work. AFAIK VBIOS'es are digitally signed, so we can't alter it in any way. Or nvidia did get themselves some one-time programmable rom inside the chip. If that's the case, we're pretty much screwed and we can only use it as is and hope for nouveau guys to finally perfect their driver. (Or patch the binary drivers in even more weird ways)