Hello, we are in the process of developing a medical imaging application that performs server side rendering and presents the results to users via a web page. This currently runs well on a Windows Server 2012 R2 instance with an AMD FirePro V5900 running in an Intel i7 workstation.
We are looking to scale this up to a full rack mounted server, potentially serving several tens to low hundreds of concurrent users. We are currently undertaking some research to determine a good setup to suggest for our customers to buy as a baseline server to run at their sites. To this end, we have purchased a Dell PowerEdge R730 with an AMD FirePro S9100 GPU installed. The OS is Windows Server 2012 R2. The latest drivers from the server section of the AMD web site (14.301.1010.1001-firepro-windows_server-retail.exe) are installed.
We are having problems getting the back-end code to recognise the GPU. As far as we can see, the GPU drivers are correctly installed, and the card is listed in device manager as working properly, although it is listed as a W9100. Catalyst recognises it and lists it in the system information section. GPU-Z also recognises it fine. Our rendering back end makes heavy use of OpenGL and DirectX, and we do not see the level of performance we would expect (considerably less than the workstation described above). Closer inspection in GPU-Z seems to suggest that the GPU is not actually being used (0 MHz GPU clock, no change in temperature, high system CPU usage). DxDiag and GLview only list the built in graphics driver (GDI Generic). I have tried installing later drivers for the FirePro Wxxxx cards, but these either don't work or do not help.
Is there a setting we need to enable to allow Windows to use the headless graphics card's resources (we obviously cannot plug a monitor, or dummy plug into it)? We have tried disabling the basic Windows graphics driver in the BIOS and also in Windows with no success. I believe that what we need to achieve is to get Windows to recognise the FirePro as the primary adaptor so DirectX/OpenGL default to using it. The marketing for this card seems to suggest that it has 'headless display support', what does this mean and do we need to do something to enable it?
To be clear, we are not virtualising desktops, running Citrix, XenApp or anything similar, we are simply using the server resources to render an image to send to a client web page.
Any help would be very much appreciated.