DIY eGPU experiences

[epiphany47]

Confirm there is a "PCI BUS" section in your \config\devcon.txt file which will be ignored if it's in English, or otherwise prohibit all 32-bit allocation if it's another language. The latter not good.

Then try 36-bit compaction, moving the 4500MHD to 36-bit space and keeping the 9800GT in 32-bit space.

If using hallucinate's script then check that the I/O ports allocated are free in Device Manager. Conflicting I/O area will also throw an error 12.

If no go, run the Setup 1.x diagnostics, archive the contents of the \diag directory, upload somewhere and link here.

- devcon.txt is in English, and has been attached

- tried 4500mhd->36bit, 9800gt->32bit, and my 9800gt was recognized, but my 4500mhd got an error 12 and booted in vga mode (see attached screenshot intel36bit-nvidia32bit.png)

- hallucinate's script is set to use port 4 and that port is open according the the Everest output in my previous post - that ought to be okay, right?

- attached compressed diag directory

if there's anything else i'm missing, let me know... i'm starting to think i'll have to do the remapping manually?

[jeremyshaw]

It seems that it'll be a little bit tricky to get required info. If environment variables are set by the process, then they're specific for this process and cannot be seen globally (from SET command). Please try to use Process Explorer as described here.

Ah, I'm sorry for being late. I'll get you the result tomorrow. I saw what you meant in PE earlier, but it's a late worknight today/early workday tommorrow.

[User Retired 2]

- tried 4500mhd->36bit, 9800gt->32bit, and my 9800gt was recognized, but my 4500mhd got an error 12 and booted in vga mode (see attached screenshot intel36bit-nvidia32bit.png)
..
if there's anything else i'm missing, let me know... i'm starting to think i'll have to do the remapping manually?

hallucinate's script won't work because your 9800GT is requesting a 512MB+16MB+32MB BAR. No way it's going to get that in 32-bit space when TOLUD=3GB.

 5: 0: 0, BAR=1 size=0x20000000 prefetch=1 64b=1
 5: 0: 0, BAR=0 size=0x01000000 prefetch=0 64b=0
 5: 0: 0, BAR=3 size=0x02000000 prefetch=0 64b=0

So then either the 4500MHD or 9800GT or both need to go to 36-bit space. Try updating the NVidia and 4500MHD drivers to the latest and test. Or maybe you can flash a vbios on the 9800GT so that only requires 256MB BAR=1? Or alternatively, try a different video card altogether. A GTS450/GTX460 would give you x1.Opt performance.

This is the first time I've seen a non-SLI desktop video card require > 256MB PCI config space.

[TsumeAwase]

Series-6 mobile chipset supports pci-e 2.0 (5GT/s) Snippet taken from the just-released Series-6 Intel chipset datasheet shown on the right confirms this. So we finally have expresscard 2.0 (x1 2.0). This chipset is used with Sandy Bridge CPUs.

Unfortunately, the support for true PCIe 2.0 5GT/s doesn't mean automatic support for Expresscard 2.0. Expresscard 2.0 interface should be implemented by notebook vendor separately. So far I haven't seen any such announcements.

[acer47ok]

Is it possible convert a memory stick duo or sdhc card slot into a pci-e type slot?

[ntsan]

Is it possible convert a memory stick duo or sdhc card slot into a pci-e type slot?

even if u can, the speed is far to slow..

[User Retired 2]

Unfortunately, the support for true PCIe 2.0 5GT/s doesn't mean automatic support for Expresscard 2.0. Expresscard 2.0 interface should be implemented by notebook vendor separately. So far I haven't seen any such announcements.

The only thing vendors need to do is conform to the Expresscard 2.0 electromagnetic spec AND ensure they don't switch the port to 2.5GT/s. The datasheet shows it's pretty easy to manually switch the port to 5GT/s if the vendor down-clocked it to not claim Expresscard 2.0 compliance.

I'd say we'll see a lot of Expresscard 2.0 capable systems announced. Why? Primarily because of USB 3.0, something that Intel's Series-6 chipset doesn't included native support for. So to get it at it's 5Gbps spec will require a Expresscard 2.0 interface to host a USB 3.0 EC card.

Harmonic Inversion(HIT) has more DIY ViDock accessories

HIT have the US$6 SWEX and US$20 PM3N in stock. US/Canada folks can buy from HIT instead of hwtools and save on shipping costs.

[Skiddy]

Following Nando's further advice, this is a new update on the implementation of DiY Vidock on a HP 2540P (i7 640LM, 4GB ram, intel HD+Nvidia GTX460, W7 64)

- Setup 1.x did solve the sound problem. I did not test further to see the effects on resuming from standby (previously impossible). Am I supposed to run Setup 1.x everytime I intend to use the Vidock?


- The directx issue is still here. I cannot run anything directx 9 or even 10 on the external monitor. Only directx 11 apps and the windows desktop will run. Even the directx10/11 option of Civilization 5 won't launch. I still get the message "D3DERR_NOTAVAILABLE" from GetDeviceCap, while dxdiag lists NO DDI version and "not available" for all 3 directx features under the "display" tab (with no option to enable them). However, everything works fine on the internal LCD..
There is little google information available on this issue. D3D devs forums basically say this:

It could be that because you've already created a REF type of device, then GetDeviceCaps is returning NOTAVAILABLE as you can't create another device on the same adapter (my D3D is a little rusty, but it looks like this could be the problem). What you might have to do is find another way of testing if the screen becomes unlocked (perhaps if your window re-gains focus), then destroy the device and re-create it.

They also say it could be linked to a multi-monitors set-up.

* My W7 64 is fairly clean. I never had to install directx on top of what's already in there (until this week)
* Reinstalling directx9.c or the latest full sdk package does not help
* UniOptimus 258.96, standard 258.96, modded 260.99: all nvidia drivers give the same result
* Cleaning the registry did not help
* Could it have to do with the modded drivers not being WHQL certified?
* Could it be a stupid thing as in "sequence in which you plug on vidock" (I switch it on and then plug the whole thing including EC2 in the laptop), an option in nvidia CP (either intel or nvidia as primary graphics) or somewhere else? An option in W7 screen resolution etc?
* Which version of the nvidia drivers are we supposed to use: desktop or mobile? (before modding them)
* Could additional Setup 1.x help solving that issue? (maybe I did not go far enough last time)

Any suggestion?


- Still, I ran a couple more benchmarks on what I could:

* 3DMark11 (the only 3Dmark that works on my external display): from 2881 to 3104 points (depending on the GTX460's o/c: from factory 765mhz to 866mhz), ie. almost the exact same score Docusync got with his own factory-o/ced GTX460 (2877)

* 3DMark06 (on internal LCD only) went from 9600 (factory o/c 765mhz) to 10418 (o/c 866) under drivers 260.99. I reckon that would be about 11300 points under the 259.96 drives.

* Surprisingly, the results were not exactly linear with progressive o/c:
i7-640LM + GTX460@765mhz: 2881pts
@802mhz (+37mhz): 2912 (+31pts)
@850mhz (+48mhz): 3071 (+159pts)
@866mhz (+16mhz): 3104 (+33pts)

[User Retired 2]

Am I supposed to run Setup 1.x everytime I intend to use the Vidock?

- The directx issue is still here. I cannot run anything directx 9 or even 10 on the external monitor. Only directx 11 apps and the windows desktop will run. Even the directx10/11 option of Civilization 5 won't launch. I still get the message "D3DERR_NOTAVAILABLE" from GetDeviceCap, while dxdiag lists NO DDI version and "not available" for all 3 directx features under the "display" tab (with no option to enable them). However, everything works fine on the internal LCD..

Thank you for posting your results. Somewhat baffled by why your external monitor doesn't work as I have not seen such an error during my testing.

First suggestion: run Driver Sweeper and remove ALL previous instances of the NVidia drivers, then do a fresh re-install. You want to use/mod the Verde (mobile) drivers as they are the ones that come with the x1.Opt tweak and internal LCD mode setup.

Second suggeston: plug in another LCD (hdmi/DVI) if have at hand to ensure it's not something to do with the monitor or any VGA/DVI adapters you may be using.

Yes, you would need to boot via Setup 1.x to overcome your sound issue. This is a minor inconvenience if you set it up with as a disk image install AND configure your startup.bat with all your necessary details in it, eg:

REM prompt/wait for GTX460 to be powered on 
call waitvid 60 10de:0e22

REM 'configure' the GTX460 so it doesn't hang win7 load
call pcicap load -d 10de:0e22

REM *Perform the PCI compaction and chainload to Win7*
call pci
call grub4dos w7

* This requires that you have previously saved a PCI Config dump of your GTX460 and done the necessary PCI compaction which will automatically save the commands required to do it to pci.bat.

Then the boot sequence to use your DIY ViDock would be:

1. Boot Win7 bootmenu, select "DIY ViDock Setup 1.x"

2. At Setup 1.x bootmenu, select default "Automated startup via startup.bat" 
option or let it count down. You will be prompted to power on your GTX460. Your 
startup.bat will run and do it's magic..

3. Now you will be back at the Win7 bootmenu, select "Windows 7 ..." upon which 
Win7 will bootup with the GTX460 active and any external LCDs attached to it 
showing your desktop/extended desktop.

Easy, per design. Use of Setup 1.x in this way can actually save you the inconvenience of having to do a standby, poweron GTX460, resume to get the GTX460 working in Win7. So it is faster at getting you to a working desktop on a monitor attached to your external video card.

[Skiddy]

Thank you once again, I will Driver_Sweeper and report.

I confirm that I get the same issue on a VGA monitor, a VGA projector and a HDMI monitor