illeatmyhat

On ‎5‎/‎25‎/‎2016 at 3:21 AM, davide445 said:

Simply not sure why is working, since it's not the standard set up where the PSU can receive info from mobo about needed power.

That's not how power supplies work.

They're very much passive devices. You stick a load on, and it sources as much power as it needs (or fail spectacularly, if your power supply is insufficient. I would say this results in "undefined behavior".)

Electricity is about as automatic as it gets. I suggest looking up Ohm's Law.

To get into detail, semiconductor devices (a graphics card) have a variable load. For example, a transistor's resistance depends on whether it's on or off. An oversimplification would be that the collective resistance of these transistors is directly related to the load of the device.

So long as the power supply can source as much power as the maximum load, you won't hit that undefined behavior condition. Manufacturers provide some protections against this, but they're not perfect. No matter what, it ends in the computer crashing.

On ‎5‎/‎24‎/‎2016 at 9:54 AM, bartimeaus said:

I've read accounts that EXP GDC does not work with windows 10 at all or that it only works when using an external monitor.

If it only works when using an external monitor as the primary display device, then it essentially doesn't work at all, according to the thread's tagline.

Optimus is described as having PCI-e 1.2x compressed links, along with being able to use the internal display. If you don't have one, you almost certainly don't have the other.

I use a PE4C 3.0 and have the experience that you describe. While the eGPU does work, it only works on an external monitor.

Furthermore, it automatically attempts to negotiate PCI-e Gen 2, which crashes the computer upon doing anything significant, because my ExpressCard and eGPU combination only support Gen 1.

So yes, there is value in using Nando's Chainloader, because it allows you to forcibly negotiate the connection to Gen 1 if necessary.

Are you saying that you're using an Optimus setup that only has the internal screen, and that it's limited to 48FPS,

or are you saying that you have an external screen, but are rendering to the internal screen while capped to 48FPS?

Then it's only one rail. 12V, at 18A. You could have figured that out by looking at the label on the Dell adapter.

 Incidentally you're using two 6-pin connectors, which means your card tops out at 150W (theoretical max, as opposed to actual usage) from the second power supply.

But really, does this even matter? We already know that the Dell adapter can deliver enough power to the graphics card (because it doesn't cause your PC to crash)
We already know that the internal PSU can deliver enough power to the entire system excluding the GPU (because it was designed to do so given an unmodified system.)

What's the point of this again?

After testing my configuration on an external monitor, everything works. Installing either desktop or notebook drivers works fine. A peculiarity with my ThinkPad and PE4C 3.0 setup is that I need to run the eGPU in PCIe Gen 1.

So it turns out that the problem is indeed Windows 10, if that could be said to be a problem.

Currently, Nvidia does not ``support" Optimus on Windows 10 for anything other than Thunderbolt, in the sense that we cannot render to the internal display, and that we cannot use PCIe 1.2x compression.

So the solution to my problem would be to install Windows 8.1, which would be egregious in my opinion.

So, the ``real" solution to my problem is to instead run Windows 8.1 on a Hyper-V VM, and then add the .VHDX to the boot menu using bcdedit. This allows me to run the Windows 8.1 VM on bare metal (with a 20% I/O penalty on the VHDX, irrelevant for an SSD). With Steam or what have you pointing to the appropriate directory, I'll have avoided cluttering my partitions while maintaining compatibility with Optimus and Nando's chainloader.

It's just math. You add up the power consumption of every component used by the first power supply.

The PCI-e slot provides 75W of power by itself. Each 6-pin connector provides 75W of power. Each 8-pin connector provides 150W of power.

Check this number against the rating of the first power supply. You can get this from the manufacturer's manual. Typically prebuilt power supplies usually have a very low maximum watt rating (~200-300W), but are built to supply power at a rate very close to the limit. Having said that, power supplies generally perform their best at 50% load.

You also want to be careful of providing too much power from a single rail. Check your second power supply to see how much power the 12V rail can provide. If that number is lower than the card requires (add it up based on the connectors used), then you will need to use multiple rails. This means either using a second cable if your power supply has multiple rails, or even a third power supply if the lone 12V rail can't deliver enough power by itself.

Prebuilts are nasty. Last time I got one I took a hammer to it and chopped up the motherboard so that I could fit in an ATX PSU and a GTX 980.

I'm assuming what's happening here is that you've got the power pins from the PE4C connected to the GPU, and that's it.

This is obviously not going to work. The PE4C requires a signal from the ExpressCard/mPCIe adapter before it will activate the power supply.

Let's just solve the problem instead of trying to make this contraption work. Essentially all you really need is to figure out how to power on the PSU.

You don't even need the PE4C.

Lazy people will jumper pin 14 and 15 on the ATX connector and then the power supply will be on permanently. While that isn't a terrible solution, it isn't a clean one.

Smart people will slave the second power supply to the first, so that both power on at the same time.

There exists a professional product that will do this for you, called the Add2PSU, available for $12.95, which I would consider ``inexpensive", if it were not a glorified jumper cable.

Personally I would stick to my guns and just jumper the damn thing, but you do you.

Even after doing all that, it didn't work in either setting.

I have the Nvidia Control Panel set up so that everything runs on the "High End Nvidia Graphics".

It all just gets pushed onto the HD4000 when the K2000M isn't available.

Thinking about it, it is probably related to the internal monitor and Windows 10.

Nvidia released beta eGPU support on Windows 10, but only for Thunderbolt. Since I'm using the PE4C 3.0, this doesn't apply to me.

I will test it on an external monitor and report back.

2 hours ago, Arbystrider said:

Ensure that you have the notebook drivers installed, not the desktop ones.

While it's true I had the desktop drivers installed, I've done a clean install with the notebook drivers, and no luck.

Still can't run anything using either Nando's chainloader or by running them alone.

2 hours ago, gharimanto said:

SW 1 = 2

SW 2 = 1

BPlus needs to document what the hell these do so people aren't blindly flipping switches until something works. I stuck with SW1 = 1 and SW2 = 2.

My current setup is:

ThinkPad W530

i7 3720QM

Nvidia Quadro K2000M

32GB RAM

Windows 10 64-bit on MBR

The eGPU is a GTX 770 connected with a PE4C 3.0.

I could try it with a GTX 980, but I don't have immediate access to it as it's in my workplace at the moment

I have tried using both Nando4's DIY eGPU setup 1.3, as well as simply running vanilla Windows with recent Nvidia graphics drivers (365.19, 365.10, 364.79) as reported in this thread

The same version of graphics drivers are being run on the eGPU and the dGPU (365.19 at the moment).
 

Both have the same result:

There are no errors in the Device Manager.

The eGPU is being detected properly by all devices (3D Mark, GeForce Experience, Speccy, Nvidia Control Panel, etc.)

However, I cannot run any applications on the GPU.

I'm somewhat at a loss as for what to do now because I'm pretty sure I've tried everything, and everything on the forums seems to indicate that everything should run perfectly at this point.

The settings for the PE4C are:
SW1:1

SW2:2

The settings for Nando's DIY eGPU Setup v1.3:

endpoint = 56.25GB

ignore dGPU

PCI compact iGPU + eGPU

32-bit on eGPU

dGPU off

startup.bat:

call speedup lbacache
call vidwait 60 
call vidinit -d %eGPU% 
call pci 
:end 
call chainload bootmgr noremap 

The reason I am chainloading with bootmgr noremap is because I am booting from an mSATA SSD, which is considered hdd1 (as opposed to hdd0).