CorePix Semiconductor effectively determines GPU task complexity

GPU chip startup CorePulse Semiconductor recently completed its Pre-A round of financing. The founder of CorePulse said, "Today, CorePulse has achieved independent innovation in core technology links in the field of GPU chip design, and we still insist on making good products down to earth."

At present, when executing tasks issued by the CPU, the graphics processor usually schedules each task to be executed. Since the complexity of each task varies, high-complexity tasks usually occupy more GPU computing resources. For high-complexity tasks, scheduling is used to reduce the GPU computing resource usage and save power. Priority is also something that needs to be considered when scheduling tasks. Although high-priority tasks are more complex, GPU resources will tend to be more inclined to high-priority tasks.

To this end, Xintong Semiconductor applied for an invention patent entitled "Method, device and medium for determining task complexity applied to GPU" on September 6, 2021 (application number: 202111036976.0), and the applicant was Xi'an Xintong Semiconductor Technology Co., Ltd.

Figure 1 Block diagram of computing device

1 is a block diagram of a computing device, which includes: a processor 6, a system memory 10, and a GPU 12. A display processor 14, a transceiver module 3, a user interface 4, and a display 8 may also be included. The transceiver module and the display processor may be part of the same integrated circuit (IC) as the processor and/or GPU. The processor may be a microprocessor, such as a central processing unit (CPU), which is configured to process instructions of a computer program for execution. The processor may include a general or special processor that controls the operation of the computing device. A user may provide input to the computing device to cause the processor to execute one or more software applications. In addition, the processor may also execute a GPU driver 22 that controls the operation of the GPU. The user may provide input to the computing device via one or more input devices.

The GPU performs graphics operations to render one or more graphics primitives to a display. Thus, when one of the software applications executing on the processor requires graphics processing, the processor may provide graphics commands and graphics data to the GPU for rendering to the display. The graphics data may include drawing commands, state information, primitive information, texture information, etc. In some cases, the GPU may have a highly parallel structure built in that provides more efficient processing of complex graphics-related operations than a processor.

Graphics memory 40 may be part of the GPU. Thus, the GPU may read or write data in the graphics memory without using a bus. In other words, the GPU may process data locally using local storage rather than off-chip memory. Such graphics memory may be referred to as on-chip memory. This allows the GPU to operate in a more efficient manner by eliminating the need for the GPU to read and write data via a bus, which may experience heavy bus traffic.

Figure 2 Schematic diagram of the method for determining task complexity applied to GPU

The figure is a flow chart of a method for determining task complexity applied to a GPU, the method comprising: first receiving a currently executed task issued by a GPU driver 22 (S401); then dividing the task into a plurality of subtasks, and polling and scheduling the subtasks to a processing core 300 as a programmable execution core 24 for execution (S402); corresponding to all processing cores 300 being assigned corresponding subtasks and the remaining subtasks not being assigned, predicting the time required to execute the remaining subtasks based on the warp execution time of the processing core 300 to which the subtasks have been assigned (S403); finally, when the time required to execute the remaining subtasks meets the set judgment condition, determining that the currently executed task is a high-complexity task (S404).

In short, CoreTone Semiconductor's patent for determining task complexity in the GPU can determine the task complexity in real time by polling and scheduling subtasks to the processing core as a programmable execution core to execute and predict the task duration, thereby providing a basis for real-time scheduling of GPU computing resources.

CoreTouch Semiconductor is a high-tech enterprise focusing on GPU chip design in the field of graphics and image rendering and high-performance computing. With the vision of "supporting a civilized world with free computing power", it is committed to building an industry-leading GPU chip design platform. In the future, CoreTouch will further promote the implementation of GPU-related application scenarios.