Trials and tribulations installing a 1070 in a M6800

You figured it out! Great work :-D
So in the end, an issue with the stock tables included in the Dell BIOS being "not quite right" (initializing stuff into an incorrect memory location)?
I suspect that this "fix" would allow any Pascal card to work in the M6800 (with Optimus disabled — requires eDP)... I know some people would like to use the 1070, or maybe 1060 or P5200, and it seems like you're on the way to a solution.

Doesn't seem like this will fix Pascal cards (other than P5000 with ES vBIOS) on the M6700. In this case, a BSOD isn't what you get, you simply can't get past the initial BIOS screens into booting an OS; the system totally hangs.

 

Yep.

This is very similar to the types of issues that the guys that put together "hackintoshes" go through. The CloverEFI bootloader that they use has some extensive support for both dynamically and statically patching DSDTs/SSDTs. OSX expects different names/paths and all sorts of stuff in a much different manner than Windows.

I basically hard coded some static patches (dumped the tables, altered them, recompiled, applied the patches by overwriting memory at boot time). If I can figure out a way to use Clover's dynamic patching, it would be much more robust and much more accessible. I'm pretty sure it can handle the DSDT/SSDT patching, but it might need an extension to outright create the FACS table.

 

Congratulations,and thanks for the hard work.

So how are you planning on making this work?.Are you going to mod the Vbios?

 

By the way....How much power is the 1070gtx draining?
Is your cpu going to bottleneck the M6800?

 

At the moment, I'm doing it as an extension to a bootloader on a USB stick. It's pretty low risk that way, and no chance of bricking the card or corrupting the OS and easy to test in a VM. I'm not sure modding the vBIOS is even an option.

When it's running at 100% GPU utilization, it's pulling almost 120W. Haven't run anymore benchmarks yet, but I wouldn't think the CPU would be the bottleneck, rather it will be total power draw.

 

Probably not because of NVIDIA's digital signature requirement with the Pascal line.
But, the vBIOS obviously can have something to do with it — we have the Quadro P5000 doing the exact same thing (BSOD at boot related to ACPI) but it works with the ES vBIOS. I wonder if it is possible to figure out why that is? If so then it may be possible to "vet" different 1070 vBIOS images for candidates without having to try flashing them all.

 

So should be ok with the regular 240w adapter.

 

It has already been determined that if you fully load the GPU (GeForce 1070 or Quadro P5000) then the CPU gets capped and cannot reach turbo speeds. This has been observed in Precision M6700 and M6800. I don't know what would cause this other than an internal power limitation (I doubt attaching an Alienware 330W PSU would help any, but haven't tried it). These systems were not designed for GPUs pulling more than 100W.

 

I spent some more time working on relocating the DSDT OpRegions rather than just deleting them and managed to get a working boot again. They are supposed to be located in the ACPI NVS and forcing them into the lower region seems to have worked.

I'm still recompiling the whole DSDT and there are a few unresolved references to a dynamic SSDT that is generated by the BIOS. That dynamic SSDT is defining power states for the CPU. (I assume the SSDT is dynamic because of all the BIOS options for Turbo, C states, etc ). A neat side effect of leaving that SSDT unresolved is that the CPU is getting locked at 3.6GHz (Max MHz when all cores are active), because there aren't any power states defined.

I ran some 3DMark benchmarks with my CPU locked at 3.6Ghz and compared results with a run from my VM. (Which was capping the CPU at 2.9Ghz)

TimeSpy Comparison: https://www.3dmark.com/compare/spy/6662346/spy/6342470
FireStrike Comparison: https://www.3dmark.com/compare/fs/18793062/fs/18255035

(My 1070 is reporting as Generic VGA for some reason.)

CPU performance was noticeably better (15% in Timespy!). However the GPU never exceeded 90% utilization. In the VM, it was reaching 99-100% for the graphics portions of the test. I think this confirms that it is the overall system power that is the limiting state. Allocating more power for the GPU just leaves less for the CPU and vice versa.

 

Actually, throwing in the 980M for comparison (capped at 100W power) and that had the best CPU score of all three. So the 1070 is definitely leeching power from the CPU:

https://www.3dmark.com/compare/spy/6662346/spy/6342470/spy/5987462