Hi guys,
So I have two GTX 950M videocards taken from ASUS M32CD4 desktops. Card itself is a GM107 chip (GTX 750/ti) paired with 2Gb DDR3 memory, hdmi and VGA ports. It is actual thing and it got reviewed here: watch?v=acATaZCh8g0
The problem is - they don’t work in 3D on anything, except two or three ASUS models. The guy who made the video, actually have original ASUS board, and had no issues. Others, who go it aftermarket, can’t make it run. Symptoms are: generic error 43 when drivers are installed, CPU-Z shows no memory / clocks. I can dump and edit card BIOS, it is valid. Bios file is twice the size of regular 950m. I can flash bios from another mobile 950m with GDDR5 and get device working, but i lost video outputs and, because no 2gb ddr3 memory card exist, I run optimus but got artifacts in 3d ( because memory type wrong). I also tried to change my Z170 board ID to ASUS, but no luck.
I believe something exist in bios to let card works correctly, or even it is just a check for some parameter in BIOS. I have downloaded M32CD4 bios and comparing it to the similar generic asus H110 one for now.
May be you know what can be the issue or what modules relate to card support? I don’t want to lose display output.
media2.24aul.ru/imgs/5d09cedc50b12400012420f2/asus-geforce-gtx-950m-2-gb-2-13935386.jpg
I got similar card with similar behaviour.
I have several findings about the idea of “Modding that gpu VBIOS instead of motherboard bios”:
- The original VBIOS VGA Bios Collection: Asus GTX 950 0 MB | TechPowerUp that gives code 43 on windows works perfectly fine with official nvidia driver on linux (tried with v510). Opengl, displaying, vulkan - all works fine. So it looks to be some “software error injection made by windows driver”.
- MSI 82.07.99.00.11 bios gives 3D device withput display that accepts driver fine both on windows and linux but gives wrong compuatation resuls due to bad memory timnings or something like this.
- I tried combining the parts of Asus VBIOS and other VBIOSes. My preliminary result is that the section that causes the above error injection is the 0x600-0xFE00 Section. Swpping parts 0x0-0x600 and 0xFE00-end from the MSI 82.07.99.00.11 bios that is 3D device which does work fine in linux but gives error 43 on windows. So the same “software error injection made by windows driver” is still in place for such mod.
Any updates on this? I ended up with this GPU as well and I would like to use it in Windows 
So this is super weird.
I have actually found this asus motherboard and tried to run 950m that i keep all this time. Tried with kaby lake i3 and it did not work. Tried with Xeon E3 1225v5 and it did not work. Then I specially bought and tried with i5 6400T and it did indeed worked right away.
And it is not all. I got another case of “locked” GPU system that I also bought. I can confirm that stock Zotac vBIOS gives error if paired with anything except skylake i5 (even kabylake for some reason) BUT it works when flashed to MSI bios (which is not even fully compatible).
Zotac EN1080K Error Code 43 / Zotac Vr Go 2.0 GTX 1070 vBIOS Request - #17 by whargus
At this point I am not sure if this is BIOS ACPI bug/limitation or vendor NVIDIA driver lock (but why lock CPU???). I have both systems currently and can provide any data from it or even give access (windows/linux) if you want.