- Joined
- Aug 26, 2013
- Messages
- 38 (0.01/day)
- Location
- Deutschland
System Name | Ghetto Cluster |
---|---|
Processor | 12x Intel Celeron @ 2,4GHz |
Motherboard | Foxconn |
Cooling | 1 Fan/CPU, nothing else |
Memory | 48x 256MB DDR2 Sticks |
Video Card(s) | None |
Storage | 24x 120GB HDD + 1x 64GB SSD (Controller) + 4x 500GB HDD (Controller) |
Display(s) | 2x Old TFT Panel (for debugging) |
Case | Very old Gigabyte Reference Cases |
Audio Device(s) | Hah |
Power Supply | 12x 235W NoName |
Software | minx7 (smaller than Tiny7, faster for Cluster Solutions) |
Hi!
Currently, we're devloping an experimental framework called CPUGFX. CPUGFX is a framework which enables developers to utilize a custom number of CPU cores exactly the same way as a GPU.
An example configuration:
Assume a PC using an Octacore CPU runs a game that puts nearly 100% load on the GPU, but not on the CPU. You can now use CPU cores which are in idle mode to calculate additional, non-crucial, effects to improve the players visual experience. CPUGFX' performance (what a surprise) scales with the number of cores available.
Shaders and Effects which should be calculated using the CPU are just passed to the CPUGFX framework which takes care of translating the shader source (e.g. GLSL) to CPU instructions. CPUGFX adds no overhead once the application is compiled, because the shader will be precompiled.
Small and fast Shaders that rely on CPU-based Physics can be offloaded to CPUGFX in order to eliminate GPU-CPU communication overhead (expect for APU / Integrated Graphics).
Here is a famous GLSL shader, compiled by CPUGFX, that creates a realistic iris (this is just a snapshot, the iris reacts to light exposure when used ingame):
Following is a sandbox test of the CPUGFX iris shader in a DirectX test enviroment. This test uses the CPU simultaniously for both smoke physics and the iris shader:
(we know that this is not a realistic scene )
The state of this project is: It works in the most cases. That means, we have a lot to improve, before we can create a serious tech demo, but we are not that far away.
Currently, we're devloping an experimental framework called CPUGFX. CPUGFX is a framework which enables developers to utilize a custom number of CPU cores exactly the same way as a GPU.
An example configuration:
Assume a PC using an Octacore CPU runs a game that puts nearly 100% load on the GPU, but not on the CPU. You can now use CPU cores which are in idle mode to calculate additional, non-crucial, effects to improve the players visual experience. CPUGFX' performance (what a surprise) scales with the number of cores available.
Shaders and Effects which should be calculated using the CPU are just passed to the CPUGFX framework which takes care of translating the shader source (e.g. GLSL) to CPU instructions. CPUGFX adds no overhead once the application is compiled, because the shader will be precompiled.
Small and fast Shaders that rely on CPU-based Physics can be offloaded to CPUGFX in order to eliminate GPU-CPU communication overhead (expect for APU / Integrated Graphics).
Here is a famous GLSL shader, compiled by CPUGFX, that creates a realistic iris (this is just a snapshot, the iris reacts to light exposure when used ingame):
Following is a sandbox test of the CPUGFX iris shader in a DirectX test enviroment. This test uses the CPU simultaniously for both smoke physics and the iris shader:
The state of this project is: It works in the most cases. That means, we have a lot to improve, before we can create a serious tech demo, but we are not that far away.