Folding Forum

Yesterday I upgrade the GPU in my machine from an old AMD HD 7870 to a new Radeon 5700 XT. When the client runs it tries 65.254.110.245 and 18.218.241.186 which both report Empty work server assignment. Are there no work units for this GPU type?

Please see this topic - viewtopic.php?f=83&t=31710. There is a known problem with the drivers provided by AMD for this new series of video cards and the support for OpenCL processing. This currently affects F@h and a number of other projects using OpenCL to process on GPU's.

Once AMD releases drivers with working OpenCL support, the card will be whitelisted. In the meantime you can select to follow the topic to see when reports come in as new driver versions come out.

I think direction is more focused to Vulkan, maybe FAH should start to look in Vulkan compute API also? I think Khronos is slowly fading away from OpenCL and OpenGL and focus is in Vulkan now.

https://pcper.com/2017/05/breaking-open ... to-vulkan/

Since the Nvidia driver seem locked on OpenCL 1.2, it really doesn't matter if at some point in the future OpenCL and Vulkan have the same API.
The API F@H has to write to is OpenCL 1.2, or it loses cross vendor compatibility.

Deciding to follow CUDA or OpenCL 2.2 means different projects for different vendors, doubling the programmer workload and halving his responsiveness.

I heared it is a bug in FAH OpenMM and will maybe fixed in upcoming FAH Core_22.

JimboPalmer wrote:Since the Nvidia driver seem locked on OpenCL 1.2, it really doesn't matter if at some point in the future OpenCL and Vulkan have the same API.
The API F@H has to write to is OpenCL 1.2, or it loses cross vendor compatibility.

Deciding to follow CUDA or OpenCL 2.2 means different projects for different vendors, doubling the programmer workload and halving his responsiveness.

I'm quite sure that nvidia does support Vulkan.

Waste of resources would be CUDA support. Soon there will be 3rd GPU vendor in game also so best route to go is stantards what every party support.

Vulkan may be the way to go in the future, but they have not yet merged in the GPU computing functionality. The current release was focused on replacing OpenGL for displaying 3D video. As I understand their announcements, the ability to take over for OpenCL is somewhere in the future, not currently an option.

So it is an emerging standard, but not ready yet to be a replacement for programming GPU processing cores.

JiiPee wrote:Waste of resources would be CUDA support. Soon there will be 3rd GPU vendor in game also so best route to go is stantards what every party support.

Yes, that is what i said, it slows F@H development to vary from OpenCL 1.2.