My first post here, and I'd like to collect my thoughts on this somewhat troubling assertion by SciresM:
Regarding the embedded driver, it's been pointed out by some that Nvidia employs a Unified Driver Architecture (UDA) that "allows one set of drivers to be used across the entire range of NVIDIA products—including consumer and workstation products". [Page 30]
When asked, however, SciresM was adamant that it wouldn't work because the GPU driver embedded is "low level".
This brings me to the other issue raised by SciresM about "pre-compiled native shader assembly". Here Nvdia also possesses a seemingly relevant patent: Unified assembly instruction set for graphics processing
If I understand this correctly (not an expert), it suggests that by using a unified assembly instruction set, the shader programs can be run on any GPUs that support the instruction set. However, I'm not sure if this patent has been used in any released products though—can't find any evidence online.
Regarding the embedded driver, it's been pointed out by some that Nvidia employs a Unified Driver Architecture (UDA) that "allows one set of drivers to be used across the entire range of NVIDIA products—including consumer and workstation products". [Page 30]
- According to Nvidia, "UDA guarantees forward and backward compatibility with software drivers, simplifying upgrading".
- "[On] every chip NVIDIA releases they have a small portion of silicon that is indicated by the green Unified Driver Hardware [Abstraction] Layer (HAL) on each GPU [...]. At the same time, each driver release contains code that interfaces directly with that Unified Driver HAL, so even before [a new GPU] was out, NVIDIA had drivers that could run the chip."
- "UDA provides extendibility for new features. This is very important for end users; without it, supporting multiple GPUs would be limited to providing the lowest common denominator in features across all GPUs." [Page 30]
- This Nvidia patent seems relevant: Unified driver architecture device identifier strategy to improve forward and backward compatibility of a unified driver architecture (my read is that either Dane can use the Maxwell driver—if compatible—or the OS can load a new driver for Dane—if available—but this is not my area of expertise)
When asked, however, SciresM was adamant that it wouldn't work because the GPU driver embedded is "low level".
This brings me to the other issue raised by SciresM about "pre-compiled native shader assembly". Here Nvdia also possesses a seemingly relevant patent: Unified assembly instruction set for graphics processing
- "[A] unified instruction set that allows shader programs of different types to use a common instruction set. The unified instruction set provides easy access for new graphics hardware features and faster compile times for shader programs written using the unified instruction set".
- "Functions that use the unified instruction set may be included in shader, vertex, or geometry programs without modification. Existing shader programs written in high-level languages may be compiled to produce shader programs that use the unified instruction set."
- "Shader programs written using the unified instruction set are assembled to produce shader microcode for execution by any graphics processors designed to support the unified instruction set." (emphasis mine)
If I understand this correctly (not an expert), it suggests that by using a unified assembly instruction set, the shader programs can be run on any GPUs that support the instruction set. However, I'm not sure if this patent has been used in any released products though—can't find any evidence online.