VMDirectPath and ATI Radeon

[taylorjonl]

I am using VMware ESXi and I am trying to setup a guest that is Windows 7 that will have an ATI Radeon video card passed through to it.  I actually had this working on a previous system but I had to reinstall.  Now when I do this the guest fails to start and I get the following:

Error message from localhost.XXXXXXXXXXX:
PCIPassthru 004:00.0: Guest tried to (null)map
32 device pages (with base address of 0xb5d20)
to a range occupied by main memory. This is
outside of the PCI Hole. Add pciHole.start =
"2909" to the configuration file and then power
on the VM.
error
12/23/2010 1:04:36 PM
media
User

When I do as it asks, the guest now starts but gets an immediate BSOD concerning memory management.  Any ideas on why this is occuring and why it worked at one point but now it fails?

[somedude1234]

Can I ask you what the BIOS settings are for your ICH10R SATA controller?

Have  a board with similar chipset (Intel S3420GPLC) and trying to  pass-through the controller to Nexenta as you are however disks do not  show up either on the napp-it page or when I issue "format -e".

How are your HD's mentioned under zpool? CxtxDx or as SDx?

Also my PCI id for this controller is 8086:3B22 when it is in AHCI mode which seems to be different from yours.

If I change it to either "IDE"or "Matrix RAID" the PCI id changes to 8086:3B21 or 8086:2822 but never to 3B34 as you mentioned.

Have no problem with passing through a LSI1068e controller but would like to have more than 8 disks in the VM.

Hopefully you can point me in the right direction.

Apologies for the late reply, I'm spending more time on the road these days than at home.

SATA is configured as AHCI.  I never tried IDE or Matrix RAID.  My drives show up as CxTxDx in the Nexenta web management console as well as in the "format -e" output:

AVAILABLE DISK SELECTIONS:
       0. c1t0d0 <DEFAULT cyl 2085 alt 2 hd 255 sec 63>
          /pci@0,0/pci15ad,1976@10/sd@0,0
       1. c2t1d0 <ATA-SAMSUNG HD204UI-0001-1.82TB>
          /pci@0,0/pci15ad,7a0@15/pci15d9,60b@0/disk@1,0
       2. c2t2d0 <ATA-SAMSUNG HD204UI-0001-1.82TB>
          /pci@0,0/pci15ad,7a0@15/pci15d9,60b@0/disk@2,0
       3. c2t3d0 <ATA-SAMSUNG HD204UI-0001-1.82TB>
          /pci@0,0/pci15ad,7a0@15/pci15d9,60b@0/disk@3,0
       4. c2t4d0 <ATA-SAMSUNG HD204UI-0001-1.82TB>
          /pci@0,0/pci15ad,7a0@15/pci15d9,60b@0/disk@4,0
       5. c2t5d0 <ATA-SAMSUNG HD204UI-0001-1.82TB>
          /pci@0,0/pci15ad,7a0@15/pci15d9,60b@0/disk@5,0

c1t0d0 is the VMDK file that Nexenta is installed on, it resides on the VMFS datastore which is using the left over space on the same 60GB SSD that ESXi is installed on.  "c1" is the VMware "LSI Logic Parallel" SCSI controller.

c2t[1..5]d0 are the Samsung 2TB SATA HDDs which are connected to the Intel Ibex Peak 6 port SATA AHCI controller which is presented to the Nexenta VM via VMDirectPath.

According to the Intel 3400 Chipset Datashet

Section 14.1.2: The value of Device ID (DID) is dynamically assigned depending on the value of the MAP register.

Section 14.1.30 shows how to decode the MAP register.

My AHCI DID of 3B34 decodes to:

0011 1011 0011 0100

- Bits 15..8 are reserved (0011 1011)

- Bits 7..6 show the current mode, which according to the spec (00) means I'm running IDE mode but the BIOS disagrees, I will double check this

- Bit 5 has to do with the number of SATA ports exposed, (1) means all 6x ports from Controller 1 and none from Controller 2

- Bits 4..2 are reserved, mine are (101), but the meaning is ?

- Bits 1..0 are the Map Value (MV), mine is (00), this is related to bits 7:6 according to the spec

CORRECTION: I DOUBLE CHECKED AND MY SATA CONTROLLER IS ACTUALLY 8086:3B22, SAME AS YOURS BELOW

Your AHCI DID of 3B22 decodes to:

0011 1011 0010 0010

- Bits 15..8 are reserved (0011 1011), same as mine

- Bits 7..6 show the current mode, which according to the spec (00) means you are running IDE mode as well, very strange

- Bit 5 has to do with the number of SATA ports exposed, (1) means all 6x ports from Controller 1 and none from Controller 2

- Bits 4..2 are reserved, yours are (000), but the meaning is ?

- Bits 1..0 are the Map Value (MV), yours is (00), this is related to bits 7:6 according to the spec

Your IDE DID of 3B21 decodes to:

0011 1011 0010 0001

- Bits 15..8 are reserved (0011 1011), same as mine

- Bits 7..6 show the current mode, which according to the spec (00) means you are running IDE mode as well, very strange

- Bit 5 has to do with the number of SATA ports exposed, (1) means all 6x ports from Controller 1 and none from Controller 2

- Bits 4..2 are reserved, yours are (000), but the meaning is ?

- Bits 1..0 are the Map Value (MV), yours is (01), this is related to bits 7:6 according to the spec

Your RAID DID of 2822 decodes to:

0000 1011 0000 0110

- Bits 15..8 are reserved (0000 1011), this is different but the meaning is not in the Intel spec.

- Bits 7..6 show the current mode, which according to the spec (00) means you are running IDE mode

- Bit 5 has to do with the number of SATA ports exposed, (0) means all 4x ports from Controller 1 and 2x ports from Controller 2

- Bits 4..2 are reserved, yours are (001), but the meaning is ?

- Bits 1..0 are the Map Value (MV), yours is (10), this is related to bits 7:6 according to the spec

After actually going through this I'm more confused than I was when I started.  I'm probably missing something here but the spec isn't entirely clear.

Regardless of what the DID is, your OS should still see it.  Did you reboot the ESXi system after chaning the VMDirectPath settings?  AFAIK this is always required.  Also, does the controller show up in the output of "lspci"?  Here's what is shown on my system:

00:00.0 Host bridge: Intel Corporation 440BX/ZX/DX - 82443BX/ZX/DX Host bridge (rev 01)
00:01.0 PCI bridge: Intel Corporation 440BX/ZX/DX - 82443BX/ZX/DX AGP bridge (rev 01)
00:07.0 ISA bridge: Intel Corporation 82371AB/EB/MB PIIX4 ISA (rev 08)
00:07.1 IDE interface: Intel Corporation 82371AB/EB/MB PIIX4 IDE (rev 01)
00:07.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 08)
00:07.7 System peripheral: VMware Virtual Machine Communication Interface (rev 10)
00:0f.0 VGA compatible controller: VMware SVGA II Adapter
00:10.0 SCSI storage controller: LSI Logic / Symbios Logic 53c1030 PCI-X Fusion-MPT Dual Ultra320 SCSI (rev 01)
00:11.0 PCI bridge: VMware PCI bridge (rev 02)
00:15.0 PCI bridge: VMware PCI Express Root Port (rev 01)
00:15.1 PCI bridge: VMware PCI Express Root Port (rev 01)

.

.

.

00:18.6 PCI bridge: VMware PCI Express Root Port (rev 01)
00:18.7 PCI bridge: VMware PCI Express Root Port (rev 01)
02:00.0 Ethernet controller: Intel Corporation 82545EM Gigabit Ethernet Controller (Copper) (rev 01)
03:00.0 SATA controller: Intel Corporation 5 Series/3400 Series Chipset 6 port SATA AHCI Controller (rev 05)

Message was edited by: somedude1234 Corrected my SATA controller VID:DID, added format -e and lspci output

[somedude1234]

I can report another success story with the 2GB-ish memory limitation when using VMDirectpath.

I followed the advice of SomeMoose:

1. Shut down my Win7 x64 workstation VM

2. Added pciHole.start = 1200 and pciHole.end = 2200 to Options/Advanced/General/Configuration Parameters

3. Changed my VM's configured memory to 6GB (6144 MB), added a memory reservation of 6144 MB

4. Started up the VM, confirmed 6GB present and no BSODs!

HUGE thank you to SomeMoose for this discovery, I now have a completely virtualized workstation without compromises.

I strongly recommend that everyone who is having issues with their GPU passthrough give this method a shot, regardless of whether you are trying to exceed 2GB of RAM on your VM or not.

[NTShad0w]

Jsnow,

my HW (I didn't make it after 2 hard long days... but only one graphics card tested and some other limits)

ESXi v5.0 b469512

Server Class Mobo: Supermicro X8DAH-F, 2x L5520 CPU (8 cores + HT)

http://www.supermicro.com/products/motherboard/QPI/5500/X8DAH_-F.cfm

Chipset is a dual 5520 (Tylersburg)

96GB of ram at 1066MHz in triple channell

onboard Matrox G200eW Graphics  -  I didn't test it as a VMDP but maybe its a good idea to test it, I'll do it asap

ADATA 128GB SSD as ESX system and VM SWAP

2x HP P800 Controllers with 10x SATA HDD

2x Intel 82576 NIC - works perfectly as a VMDP

Asus NVidia GT240 512MB RAM Graphics card seems to be VMDP adapter - it's on chips codename GT215

Tested on:

- MS Win 7 Pro PL x64 Image with snapshots, has 2x vCPU and 2GB of ram (fully reserved)

- VMTools was the newest from ESXi v5.0

- VM HW Version was 8

- only one or two VMDP devices added to VM

- I have on the other vm VMDP NIC Card, but vm is not active and it rather should not mather at all

- All tests with almost same result - VMDP Device (GT240) is listed, NVidia drivers are installed (and saw the card) but card doesn't start (see attachment, error code 43) - GT240 devace in Device Manager start only when I Disable (or uninstall) "VMware SVGA 3D (Microsoft Corporation - WDDM)" but as expected drivers dont want to put it as a primary or secondary display in the system

- I test only in MS Win7 Pro x64, /Update - I tested also MS Win7 Pro x32 EN - same results

- I test only on this ESXi v5.0 build (469512) - it's first 5.0 build

- I tested only latest NVidia driver v280.26 64bit

- I know and aknowledge that vmware/ESXi 4.x/5.x can/will renumerate any PCI/PCIe card, not only VMDP cards and it may result to recognize card as new one and if it is a RAID Controller may eventually "reassign" new SID's to a DataStores which will result to...a lot of prolems (You will not see your DataStores anymore...:( and can't connect them normally, only from CLI)

I'm a VM Infrastructure Architect and personally VM passionate, genrally if it will be easy to enable this card in my enviroment, I will enable it, but unfortunately I failed to do so...:(, maybe it's just doesn't work or it is needed very specific advanced configuration to work or... a just a little specific change but nobody know where - coz it's unsupported (right now) to VMDP Graphic/Display adapters, but I know that a lot of configs works very well.

If I buy another card... probably ATI or just use my desktop ATI 5870 I will give you an update.

kind regards

NTShad0w

[NTShad0w]

SomeMoose,

I can confirm that vm values:

pciHole.start = "1200" and pciHole.end = "2200"

(Options/Advanced/General/Configuration Parameters) - in properties of VM.

seems to working for using a VM with VMDP with more than 2/3GB of RAM !!!!, I just start it with 4GB and... still testing this f...g GT240 but machine start and see whole memory and... dont looks like want to crash :smileysilly:

grats SomeMoose :smileysilly:

cheers

NTShad0w

[NTShad0w]

jsnow and mates,

it looks like I have a success with integrated server based old Matrox MGA G200eW 16MB adpater :smileysilly:.

HW/SW:

ESXi v5.0 b469512

Server Class Mobo: Supermicro X8DAH-F, 2x L5520 CPU (8 cores + HT)

http://www.supermicro.com/products/motherboard/QPI/5500/X8DAH_-F.cfm

Chipset is a dual 5520 (Tylersburg)

96GB of ram at 1066MHz in triple channell

onboard Matrox G200eW 16MB Graphics  -  I didn't test it as a VMDP but maybe its a good idea to test it, I'll do it asap

ADATA 128GB SSD as ESX system and VM SWAP

2x HP P800 Controllers with 10x SATA HDD

2x Intel 82576 NIC - works perfectly as a VMDP

Asus NVidia GT240 512MB RAM Graphics card - faulty tested with VMDP by me on this config, its still in the server and is enabled VMDP dev in ESXi

as you can saw my 2 posts above I failed to do VMDP on Asus NVidia Geforce GT250 (GT215 codename chip) but I try this with my integrated server graphics... without touch any hw (I even don't remove this NVidia) just enable in ESXi VMPD for Matrox and reboot the ESXi server, add this card to test Win7 x64, install drivers, reboot vm and after that I saw same effect as before that my new adapter is visible, drivers installed but dont working together with VMware SVGA 3D (Microsoft Corporation WDDM)..adapter...:( but... when I disable VMware adapter and reenable Matrox in Device Manager, win7 asks me to reboot... and after reboot I observe my "Windows start" "hang" on starting but it's resume on my monitor connected to a server :smileysilly: and working as espected

This Matrox MGA G200eW 16MB is a Supermicro Mobo integrated adapter, is set in BIOS as primary, ESXi start on this adapter only and during start screen freeze (as expected). The fun thing is that I not get USB redirect so I dont have a mouse or eyb but I observe that when I click on the vmclient v, window (which is black or just saw last screen of starting windows) I can control a machine via my laptop mouse and keyb but its happen on Matrox card insteed of my vmware vga card :smileysilly: and its working ok:). Of course by this card is very old and server based it doesn't have any acceleration and working rather very slow, but it's normal for such adapters, they are only for start system in basic resolution.

I test it with 2 vm, MS Win7 Pro x64 bit (4x vCPU and 4GB vRAM, pciHole.start=1200, pciHole.end=2200) and with Win2k8 R2 x64 DataCenter (4x vCPU and 16GB of RAM - I dont use options like pciHole.start or pciHole.end here!!, this vm also have second VMDP - intel NIC, its working perfectly). So the parameters "pciHole.start or pciHole.end"

are required for devices that use/map some memory (something like windows swap), its related to almost all modern graphics cards and a lot of other cards like e-fax cards (like Eicon Diva) or a lot of athers, I think that soon when this technology (VMDP) will be more often, vendors will give exact info what part of memory need to be reserved for such card, generally if you have card with 1GB of RAM you (In my opinion) need to reserve exactly such memory in ram (via parameters pciHole.start or pciHole.end) to use more than 2GB of ram and to have vm system stable, probably you have to try some values to get best one.

cons...

its evidently in my situation that VMware SVGA 3D (Microsoft Corporation WDDM) from my vmtools block other graphics cards to work... and some cards (or rather drivers) may work when you disable VMware SVGA 3D in Device manager and some like my NVidia and a lot of that other mates tests didn't

So my recomendation for next tests is to:

- check and described how to use WDDM v1.1 or newer drivers (I even don't know how to check version of WDDM...) but I know that there are a newest models like 2.0, 2.1:

http://en.wikipedia.org/wiki/Windows_Display_Driver_Model

So I don't know what model of WDDM use my NVidia GT240 (with 280.26 drivers) and VMware SVGA 3D,  how to check this?

- check how to enable our VMDP graphics adapter together with VMware SVGA or alone without VMware SVGA...(what we know not all drivers will work, maybe someone know the reason of that?)

- check that it is related to some VM Tools or VM HW version?

- check what pciHole.start or pciHole.end parameters will be best for such card

- asks VMware Corp to give a little support for that, or even help to achive a good results or give some tweaks how to disable or remove VMware SVGA adapter entirely - I think its a main problem, when windows will se only 1 adapter it should work in most cases with almost every graphics card I expect !!!

- so maybe someone know how to totally remove VMware SVGA adapter from VM on ESXi...?

kind regards and Good Luck for all of You mates

NTShad0w

[Nathanw20111014]

Just wanted to add some successes and failures with GPU passthrough.   Using ESXi 5 I could never get a nVidia Quadro 4000 to work; however I was able to get a ATI 4550 and a Fire Pro V7900 to work with full hardware acceleration.  One thing I noticed with the ATI cards the vmware SVGA 3D display adapter always had to be enabled in the device manager before it would allow me to move my primary display to the ATI adapters in the display properties.  If I either disabled or uninstalled the vmware SVGA display adapter in the device manager it would not allow me to change the primary display to the ATI adapter in the display adapters.  It appears that the vmware SVGA adapter in the device manager is required before you can switch over to the ATI adapter in the display properties.  Also if the vmware SVGA adapter was the primary display and tha ATI card was the extended secondary display the catalyst control center would not come up nor any hardware acceleration would work.  It would only work once the ATI adapter was the primary display.  This may have been the cause of some problems listed here earlier about catalyst cc not coming up because it didn't detect a ATI adapter or green YouTube screens. 

One of the issues I am having is trying to keep the vmware SVGA adapter from becoming the primary display after reboots or shutdown of the VM.  I always have to go back in and change it around to use the ATI as primary and disable the vmware display in the display properties.  I thought if I just disabled or uninstalled the vmware SVGA adapter in the device manager that would do the trick, but no such luck.  It would just not allow me to even move over to the ATI adapter since it would then default to some horrible 8 bit VGA adapter at 800x600 res. 

All this was done on a win 7 vm with 2 GB of ram and the 11.9 catalyst driver or newest fire pro driver off amd.com.  My next task is to look and see if ATI has any apis similar to nVidias nvapi to write scripts to manipulate displays and resolutions after each boot.  If anyone has any suggestions I'd greatly appreciate the info.  If anyone wants anymore details about my setup let me know.

[MattGagliardi]

I can report success with both the nVidia Quadro 2000 and 4000. I've also made some headway with getting the GPU assigned as the primary adapter/monitor and keeping it that way (though I had to disable the VMware adapter to do it)...the trick seems to be (at least for me) delaying the startup of the VMware Tools service. My working theory is that in doing that I allow the OS to "grab" the GPU first and designate it the primary monitor after reboots.

Until you get the GPU selected as the "primary" adapter is seems like a lot of software won't do hardware acceleration. Again, working theory...the software is looking to see what adapter is "primary" and if the VMware SVGA adapter is designated as such the software sees that it can't do full hardware acceleration and won't try it.

The difference between a successful person and others is not a lack of strength, not a lack of knowledge, but rather in a lack of will.

[Nathanw20111014]

Matt,

Great to hear about the Quadro 4000 (by they way what driver version did you use); I'm going to go back and try it some more.  I would definitely much rather have the Quadro since I'm needing to do some CUDA.

Awesome job with getting the ATI recognized as the primary adapter after reboot; that is exactly where I'm trying to get.  When you say you disabled the vmware svga adapter did you do it in the device manager?  On the vmware tools service are you talking about selecting the "Automatic (delayed start)" in the startup type drop down menu when you are in the Windows services window?  Any more details on that procedure would be greatly appreciated as that is exactly what I'm trying to do.

[MattGagliardi]

Nathan,

My solution on the delayed startup was to put in a dependency for the VMware Tools service...one of my app's services (the one that recognizes which video card is "primary") has to start before VMware Tools does. That seems to keep my guest settled with the GPU as primary. I suspect a delayed service start would also work, I've just not tried it.

Matt

The difference between a successful person and others is not a lack of strength, not a lack of knowledge, but rather in a lack of will.

[Nathanw20111014]

I tried the delayed start but didn't have any luck.  Just out of curiosity what "app" do you use that recognizes which video adapter is primary? 

[MattGagliardi]

It's an application that uses the video card for 3D rendering. Unless it picks up the 4000 as "primary" it won't work. The application is set to start/run as a service, so I just delay the start of VMware Tools until after the other service starts. My thinking was something along the lines of VMware Tools controlling the SVGA driver (which it may or may not), so by delaying the start of the tools I could delay the SVGA driver's activation. I'm not sure if that's what's really happening or not.

Perhaps there are some settings in the guest's BIOS that would allow you to delay/disable the SVGA driver (similar to a BIOS setting that would allow you to turn of "embedded" video in favor of a PCI card). Have you actually made the GPU the "primary" and disabled the VMware SVGA card?

The difference between a successful person and others is not a lack of strength, not a lack of knowledge, but rather in a lack of will.

[Nathanw20111014]

Matt,

Thank you for entertaining my curiosity. My situation is a little different than yours, but I may be able to do something similar.

I thought the same thing about the guest's BIOS, but no such luck. There are unfortunately no settings in there related to video. Its very very basic.

My situation is I have a high end 3D application (basically just like a game) that uses DirectX so it needs access to the video card that is passed through. Unless I go in and switch the video adapters around it comes up with the default vmware svga as the primary and the application will not run. I'm also using a "screen scrapper" remoting application to remote this session to a thin client. It works great once I move around the monitors to use the ATI adpater. What keeps happening is when the VM reboots and I remote in to this VM (I have a dual monitor setup) my primary display is the vmware svga and the secondary (extended) display is the ATI card. I then have to go in the display properties and move the adapters around so that I have dual monitors on the ATI card and nothing on the vmware svga.

So far what I have tried once everything is the way I like it (dual ATI displays) I go in the device manager and disable the vmware svga display adapter and it works fine until I reboot the VM and then it goes into an even lower resolution standard vga 800x600 (since the vmware svga driver is disabled and/or uninstalled) which does not even give me the option in display properties to move the displays around. That is were I came to the thought that maybe the vmware svga driver needs to be enabled (WDDM driver; also actually the wiki for WDDM mentions this feature in ver 1.1) to allow multiple video adapters in one OS. The vmware svga (when the "enable 3D" box is checked in the vm properties using the vmware client) will do some 3D, but my application frame rate is not good enough to work.

So after all that I may try and start up my application as a service and see if that works. Does this 3D rendering application of yours just use the 4000 as a computational task or does it actually display graphics to the user? When I do disable or uninstall the vmware svga adapter and reboot the device manager does show the passed through card as working properly. I just cannot switch to it for displaying any nice graphics or high resolutions. It may however be available for an application to use for rendering and other tasks that don't require any user viewable graphics. Thanks again for describing your situtaion.

[MattGagliardi]

Using it for both computational and display. When you do your modification of the monitors/adapters how are you accessing the machine (desktop)?

[Nathanw20111014]

Matt,

I am using a remote login application (rgs) not windows remote desktop or the console window from the vshere client.

[Nathanw20111014]

By the way how are you accessing your desktop?

[djsecrist]

Has anyone had luck on ESXi 5 with an ATI 6750 with 2GB of RAM on the graphics card?  I have tried many different values for the pcihole.start and .end but I keep getting the BSOD on atikmpag.sys during Windows 7 boot.  I can get into Windows if I disable the vmware graphics card from within Safe Mode, but then the Catalyst drivers complain that no ATI card is installed.  I am not sure if this matters, but this is my only video card in the physical machine and I am trying to pass it through to a guest.  Any ideas?  Thanks.

[Kamerat]

Install VMware Tools, reboot and then shut down.

Configure your 6750 for passthrough.

Boot Windows into safe mode, then open Device Manager and disable your 6750. Reboot.

Install Catalyst, a reboot may be needed.

When you get picture on the screen connected to the 6750 you can open "Screen Resolution" (as it's called in Windows 7). Make the screen connected to the 6750 your primary screen, then you can disable your VMware screen.

Configure the pcihole options if you want to use more than 2816MB RAM (2304MB two GPUs inn passthrough to the same guest and 1792MB with three) in your guest os running with the 6750.

[Linjo]

VMware recently announced this togheter with Nvidia, could be an interesting read for the contibuters to this thread:

http://hothardware.com/News/Nvidia%2DVMWare%2DTeam%2DUp%2DTo%2DOffer%2DHighEnd%2DRendering%2DOn%2DMo...

// Linjo

[djsecrist]

Thanks for the help, Kamerat.  Unfortunately, I still can't get it to work.  Every time I install the ATI drivers package, I get a BSOD on atikmpag.sys.  I have tried it with 2816MB of guest system RAM, and it still doesn't work.  Should I be configuring the VM from the Vmware console?  Should I use the pcihole.start and end values of 1200 and 2200 for a 6750 with 2GB of RAM?  Does it matter that the 6750 is my only physical video card in the system?  Any other thoughts?  Thanks again.

[BAM279]

I normally use VNC to remotely setup/install my VMs that use a physical video card (because it doesnt replace/interfere with the graphics driver, whereas the likes of RDP does - not sure about the vmware console).

Even with VNC, getting at least one BSOD in the atixxxx.sys file is part and parcel of the process, in my experience.  I recently set up yet another VM with which to pass the ATI 3450 HD card through, and this is how the process went:

1.  First boot up of the VM with the physical card passed through has Windows 7 detect the card and trying to install the built-in microsoft ATI driver, which causes BSOD.

2. Reboot VM into safe mode, disable the ATI card and reboot system normally

3. Once booted up, run the ATI Catalyst driver installation - despite the ATI card being disabled in Device Manager, the install process should still detect the card and begin installing the drivers (may get another BSOD during this).

4. Reboot VM into safe mode and go into device manager and now you should see the ATI card listed with the official ATI driver.  Disable it and reboot normally.

5.  Once booted normally into the VM, go into device manager and enable the ATI card - it should enable fine and you can set up your display to output video via monitor 2 which is the ATI adapter.

Good luck and if it fails, just keep trying until it does... it should work eventually

BAM.