GPU Data Transfer via DVI Interface Bypassing PCIe Bottleneck
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for transferring network data from a graphics processing unit (GPU) to a remote external device are inefficient, leading to increased delay and wastage of computer system resources, particularly due to limited PCIe bus bandwidth and asynchronous memory operations, which cannot meet the demands of high-bandwidth applications.
Innovation Solution
A device and system that convert TMDS signals from a graphics card into network data using a data processing module and network transmitter, allowing for direct transfer via a network, with optional wired or wireless transmission, and including features like encryption, power management, and control signals to optimize data transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If data is transferred from GPU to external device via PCIE bus, then data transfer can be achieved, but bandwidth is limited and delay is increased
Solution Approach 1:
The patent introduces a display interface (DVI/HDMI) as an intermediary transmission path between the GPU and external device. Instead of directly using the PCIE bus for data transfer, the system converts GPU output data into display interface signals that can be transmitted at higher speeds, thereby reducing transfer delay and freeing up PCIE bus resources for other operations.
Solution Approach 2:
The patent changes the transmission parameters by converting data from PCIE protocol to display interface protocol (TMDS, HDMI). This parameter transformation enables the data to be transmitted through a different channel with higher bandwidth characteristics, thus reducing transfer time and improving overall system efficiency.
2Productivity
If CPU copies data from video memory to internal memory, then data can be transferred to network, but CPU resources and memory resources are wasted
Solution Approach 1:
The patent extracts the data transfer function from the CPU's responsibility. Instead of requiring the CPU to copy data from video memory to internal memory before network transmission, the system directly outputs processed data from the GPU through the display interface to the external device, eliminating the intermediate copying step and reducing CPU and memory resource consumption.
Solution Approach 2:
The GPU performs self-service by directly outputting processed data through its built-in display interface circuitry without requiring CPU intervention for data copying. The display interface circuitry within the GPU handles the conversion and transmission autonomously, improving productivity while conserving CPU and memory resources.
3Quantity of substance
If PCIE bus is used for data transfer, then connection is established, but bandwidth is limited and cannot satisfy high-bandwidth applications
Solution Approach 1:
The patent leverages the universality of display interfaces which were originally designed for video output but can also serve as high-speed data transmission channels. By reusing the existing display interface infrastructure (DVI/HDMI) for general data transfer purposes, the system achieves high bandwidth without adding complex dedicated transmission hardware, thus satisfying high-bandwidth application requirements while maintaining reasonable system complexity.
Data Source
AI summary
A device, system and method for transferring network data are presented. The device includes: a data processing module configured to convert a data signal in a TMDS signal output via a DVI of a graphics card to network data for being transferred via a network; and a network transmitter for receiving the network data and transferring the network data to an external device via the network. The data signal is generated by encoding texture data generated by a GPU of the graphics card by a digital video sender of the graphics card. The texture data is generated by binding a pointer to general computing data stored in a device memory of the graphics card to a texture stored in the device memory by the GPU. The general computing data is generated by general computation executed by the GPU.