GPU Memory Controller Data Path Width Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing memory controller and display controller data path in GPUs is inefficient due to the need to read and transmit multiple rows of data, leading to wasted die area and increased costs, as the memory controller reads extra data that is not used by the display controller.
Innovation Solution
The implementation of a graphics processing unit with a data path that transmits only one row of data at a time from the memory controller to the display controller, using an enhanced command format that includes a row field and sector mask to selectively discard and transmit only the required data, optimizing the data path width and reducing unnecessary data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the memory controller reads multiple rows of data to fulfill a display controller request, then the data access requirement is met, but the data path width must be increased to accommodate the extra data bandwidth
Solution Approach 1:
The patent extracts and transmits only the specific row of data requested by the display controller from the multiple rows read by the memory controller. The memory controller reads N rows (where N≥2) from memory, but only the required row is extracted and transmitted through the data path to the display controller, while the other rows are discarded. This extraction principle allows the data path to be sized for only one row width rather than N rows, resolving the contradiction between meeting data access requirements and minimizing data path width.
2Productivity
If a data buffer is added to the display controller to store discarded data for subsequent use, then overall memory use is improved, but the data path must still be large enough to carry multiple rows of data
Solution Approach 1:
Instead of buffering all N rows read from memory and transmitting them through a wide data path, the invention extracts only the specific row needed by the display controller and transmits only that row through a narrow data path. The extraction happens at the memory controller level, preventing unnecessary data from entering the data path in the first place, thus avoiding both the wide data path requirement and the need for large buffers.
3Adaptability or versatility
If the data path is sized to accommodate multiple rows of data, then all data can be transmitted, but die area is wasted due to unused data capacity
Solution Approach 1:
The patent applies extraction by identifying and transmitting only the specific row of data that the display controller needs from the N rows read by the memory controller. This ensures the data path is sized precisely for one row width, maximizing adaptability for the actual data transmission requirement while minimizing die area by avoiding provision for N-row capacity that would remain unused.
Solution Approach 2:
The invention changes the data path width parameter from N-row width to 1-row width, where N≥2 is the number of rows read by the memory controller. This parameter change is enabled by the extraction mechanism that separates the needed data from the unnecessary data, allowing the data path to be optimized for actual usage rather than maximum potential usage, thus reducing die area while maintaining transmission capability for the required data.
Data Source
AI summary
Embodiments of the present invention set forth a technique for optimizing the on-chip data path between a memory controller and a display controller within a graphics processing unit (GPU). A row selection field and a sector mask are included within a memory access command transmitted from the display controller to the memory controller indicating which row of data is being requested from memory. The memory controller responds to the memory access command by returning only the row of data corresponding to the requested row to the display controller over the on-chip data path. Any extraneous data received by the memory controller in the process of accessing the specifically requested row of data is stripped out and not transmitted back to the display controller. One advantage of the present invention is that the width of the on-chip data path can be reduced by a factor of two or more as a result of the greater operational efficiency gained by stripping out extraneous data before transmitting the data to the display controller.