157 results about "Display memory" patented technology

The invention is suitable for the field of GPU management, and provides an intelligent dynamic management method for GPU computing power and a cloud platform, and the method comprises the following steps: collecting task data and GPU state data, and generating a task queue and a resource allocation strategy in combination with a scheduling plug-in; based on the task queue and the GPU real-time load, dynamically adjusting a resource allocation proportion through reinforcement learning to analyze a task dependency relationship and generate a migration plan; optimizing a communication path and adjusting asynchronous transmission delay according to the task dependency relationship and the GPU communication topology, and outputting a synchronous state mark; and monitoring abnormity in combination with the synchronization state and the GPU hardware state, executing thermal migration according to the migration plan, performing video memory recovery, and updating the resource idle list. According to the invention, through algorithm innovation and hardware collaborative optimization, intelligent, dynamic and efficient resource scheduling in the multi-GPU system is realized.

The invention discloses a heterogeneous GPU resource management scheduling method, a computer device, a medium and a product. The method comprises the following steps: acquiring computing power resources of each node, including a GPU model, a GPU video memory, a GPU number, a computing power segmentation scheme and a GPU use condition; automatically segmenting the heterogeneous GPU of each node according to the computing power segmentation scheme of each node to obtain resource segmentation information; obtaining an expected computing power and an expected video memory of a to-be-executed task; screening out target nodes meeting the expected computing power and the expected video memory according to a node analysis strategy; and according to the sub-resource analysis strategy, screening out sub-resources meeting the expected computing power and the expected video memory, recording the sub-resources as target sub-resources, and allocating the to-be-executed task to the target sub-resources. According to the method, GPU resource fragmentation is effectively reduced through global and node resource conjoint analysis scheduling, meanwhile, pooling management, dynamic configuration and intelligent scheduling of heterogeneous computing equipment can be achieved, and the resource utilization rate is effectively increased.

The invention discloses a CFD (computational fluid dynamics) boundary processing acceleration method based on GPU (graphics processing unit) parallelism, and belongs to the technical field of fluid mechanics related calculation. The problem that in the prior art, a traditional CFD boundary processing acceleration method is low in calculation efficiency is solved. The method comprises the following steps: generating a computational grid required for simulating a flow field, and reading grid information and boundary condition information of computational simulation; initializing a flow field according to the calculation parameters, copying flow field data and flow field boundary information from a main memory to a video memory, establishing a boundary pointer array in the video memory, and instantiating a boundary pointed by each pointer; according to the boundary pointer array, the flow field data and the flow field boundary information, time step cycle calculation is carried out by utilizing a GPU, and physical quantity updating of a boundary virtual unit corresponding to the thread is realized; and after time step cycle calculation convergence, video memory flow field data are copied to a main memory, and flow field simulation calculation is completed. The method effectively improves the calculation efficiency, and can be applied to CFD boundary processing.

The invention relates to a multi-node resource scheduling method for a cloud three-dimensional modeling task, and relates to the technical field of cloud computing resource scheduling, and the method comprises the specific steps: analyzing a three-dimensional modeling task log stream in real time, determining a current modeling stage through a regular matching key operation instruction, and according to a label of the current stage, carrying out the multi-node resource scheduling of the cloud three-dimensional modeling task. The method comprises the following steps: extracting a GPU / CPU matching rule and a video memory capacity requirement of a corresponding stage from a preset resource configuration template library, dynamically calculating a block parameter N of a to-be-reconstructed grid based on the video memory capacity requirement, and dividing the to-be-reconstructed grid into N * N * N sub-blocks according to spatial positions; by analyzing the three-dimensional modeling task log stream in real time and utilizing the regular matching key operation instruction, the current modeling stage can be accurately recognized, and then the GPU / CPU matching rule and the video memory capacity requirement of the corresponding stage are extracted from the preset resource configuration template library.

According to the method, PCIe extended storage equipment is identified by responding to a system starting signal, and initialization and address mapping control logic is executed to construct a first address space addressable by a GPU (Graphics Processing Unit); then, loading model data according to an AI task instruction, and dynamically dividing hot, warm and cold multi-level data popularity levels by monitoring memory access frequency; a migration data block is generated based on the popularity level, a migration data flow is organized and verified through priority scheduling logic, and intelligent data scheduling between extended storage and a GPU video memory is achieved; finally, the resources are released after the task is completed. Therefore, dynamic expansion of the physical video memory capacity is effectively achieved, large-scale AI model localization training is met, dependence on a high-end GPU chip is remarkably reduced, and data safety and system economy are improved.

The invention provides a spatio-temporal data volume rendering method oriented to an unreal engine, and belongs to the technical field of spatio-temporal data volume rendering. Ocean spatio-temporal data in a NetCDF format is acquired, a two-dimensional volume texture is generated, and a video memory partition management system is established to divide a GPU video memory into an upper layer video memory grid and a lower layer video memory grid; a double-layer game algorithm is adopted to realize differentiated resource allocation, a volume rendering material object is created in an unreal engine, a volume coloring framework based on light stepping is established, and the sampling step length is dynamically adjusted through an adaptive light stepping algorithm, so that efficient volume texture sampling based on a video memory grid is realized. And executing ray casting calculation to generate a three-dimensional visualization effect, establishing dynamic interaction control of space-time dimension to realize synchronous display of multi-source ocean data, and solving the technical problem of low rendering efficiency caused by unreasonable video memory resource allocation in the ocean space-time data three-dimensional visualization rendering process.

The embodiment of the invention provides a video memory control method, device, equipment and system and a computer storage medium. The method comprises the steps that in response to an obtained drive upgrading request of an image processing unit (GPU), a before-upgrading page table corresponding to the GPU is obtained, and the before-upgrading page table comprises a mapping relation between a physical address and a virtual address of a GPU video memory; determining a data cleaning queue corresponding to the GPU, wherein the data cleaning queue is used for performing global cleaning on data in a GPU video memory; under the condition that the data cleaning queue is locked, driving upgrading operation is carried out on the GPU based on the driving upgrading request, an upgraded page table corresponding to the GPU is obtained, and the upgraded page table comprises a mapping relation between a physical address and a virtual address of a GPU video memory; and regulating and controlling data in the GPU video memory based on the pre-upgrading page table and the post-upgrading page table. In the embodiment of the invention, the driver upgrading operation of the GPU can be realized without operations such as video memory backup, shutdown, restart and the like.

Content-addressable memory-in-display (CAMiD) electronic displays and circuitry are provided. An electronic display may include a pixel active array and content-addressable control cells to control display pixels of the pixel active array using control signals propagated to the pixel active array on column lines by column driver circuits of the content-addressable control cells.

The invention discloses a big language model dialogue system for Token-level resident memory and a use method. The big language model dialogue system comprises an inference engine module, a big language model module, an independent memory model module and an external storage module, relates to the technical field of big language model memory enhancement, after a current dialogue Token sequence is received, whether historical dialogue information is needed or not is judged through a big language model module, and if yes, the historical dialogue information is called from an external storage module through an independent memory model module, so that information displayed by finally responding to the Token sequence is more complete, and user experience is improved. The problems that a model cannot completely retain and utilize long-distance historical information, so that questions cannot be answered and information is forgotten are solved; an external storage module is arranged, and model parameters of an independent memory model module and Token-level data of all historical conversations are resided through a memory mapping technology; and zero occupation of the GPU video memory is realized.

The invention relates to the technical field of computer graphics, and particularly discloses a three-dimensional intelligent visualization method and device for multi-source data loading, a medium and equipment.The method comprises the steps that an oblique photography model is loaded and dynamically adjusted, and an oblique photography live-action base is constructed; generating semantic 3D tiles data on the basis of the base; identifying a key entity object, carrying out lightweight processing on a high-precision GLB / GLTF model of the key entity object, and dynamically implanting the high-precision GLB / GLTF model into a scene; performing dynamic analysis and spatial superposition on the KML data associated with the key object, and constructing a multi-level three-dimensional scene information system from entity annotation to macroscopic annotation; dynamically coupling the KML elements with terrain elevation data to form a high-precision digital elevation base supporting visualization; based on a video memory dynamic partitioning mechanism, optimized resource scheduling is performed on a substrate, and adaptive balance of terrain complexity and system load is realized. According to the invention, the precision and efficiency of multi-source heterogeneous data fusion and the system stability can be improved.

The invention relates to an ultra-high-definition image restoration method, device and equipment based on clustering center feature scanning. The method comprises the steps that a clustering center scanning unit and a space-channel feature modulator are constructed, the clustering center scanning unit constructs a plurality of center features through a feature aggregation stage sparse sampling input feature map, a cosine similarity matrix of pixels and the center features is calculated, similar pixels are screened through an activation function and aggregated to generate clustering core features, and the clustering core features are clustered through the space-channel feature modulator; the similarity matrix is utilized to establish association mapping of core and non-core features through a fractional diffusion stage, global modeling weight is transmitted, a space-channel feature modulator extracts global features and texture detail features through parallel attention branches, the two modules are fused based on an asymmetric encoder-decoder architecture, and the texture detail features are obtained. According to the method, an ultra-high-definition image restoration network is constructed, low-quality ultra-high-definition images are restored after training, high-quality images are output, and the problems of video memory bottleneck and detail blurring in the prior art can be effectively solved.

The invention provides a reasoning instance management method and device. The method is applied to inference service and comprises the steps that when a first message from an inference engine is received, a GPU resource pool is inquired, the inference engine is used for monitoring the GPU resource use condition of an inference instance, and the first message is sent out when the inference engine monitors that the video memory of the inference instance is insufficient; if the GPU card with the same type as the current GPU card exists in the GPU resource pool, the identification information of the GPU card is sent to the reasoning instance, so that the GPU card is newly added to the reasoning instance on the basis of the current GPU card; otherwise, the identification information of the GPU card with the larger video memory is sent to the reasoning instance, so that the reasoning instance replaces the current GPU card with the GPU card with the larger video memory. According to the method, intelligent monitoring, dynamic scheduling and fault recovery of GPU resources are realized through close cooperation of an inference engine and an inference service, so that a stable and reliable execution environment is provided for an inference task.

The embodiment of the invention provides a GPU computing power scheduling method and device, equipment, a medium and a program product, and relates to the technical field of computing power scheduling, and the method comprises the steps: dividing a computing unit and video memory resources of a GPU into a plurality of dynamically adjustable partitions according to a real-time task demand; calculating the priority weight of each task; preemptive scheduling and cooperative scheduling are combined, resource preemption is carried out on the first priority task, and time division multiplexing is carried out on the non-preemptive task; executing a dynamic voltage frequency adjustment (DVFS) and task merging strategy; through dynamic resource partitioning, dynamic priority evaluation, a hybrid scheduling strategy and an energy efficiency optimization algorithm, the GPU resource utilization rate and the energy efficiency ratio in a multi-task scene are remarkably improved, and meanwhile, energy consumption is reduced by combining a DVFS and a task merging strategy.

The present application relates to the technical field of project management, in particular to a large model batch inference and data flow optimization system for MOE architecture. The present application sets a collaborative architecture of a request access module, an environment perception module, an expert routing engine, a resource scheduling module and a dynamic optimization control module, uses the request access module to extract the text length and the subject type to provide the basis for accurate routing, collects the GPU display memory, the I / O bandwidth and the request queue depth in real time through the environment perception module, comprehensively monitors the system load, activates the expert sub-network according to the request characteristics through the expert routing engine to avoid invalid calculation, manages the weight loading and resource allocation through the resource scheduling module to reduce the I / O bottleneck, and finally intelligently triggers the optimization strategy based on the routing conflict factor through the dynamic optimization control module to solve the problems of large inference delay fluctuation and unbalanced resource utilization mentioned in the background technology, and realize stable low-delay response and resource collaborative optimization in a high-concurrency scenario.

A method for sharing a distributed video memory across process groups and a computing device, the method comprising: in response to receiving a recovery instruction from an upper application, a first process in a plurality of processes traverses metadata of each virtual address in a first metadata table thereof, mapping a first physical address to each of the plurality of first virtual addresses of the local mapping type, wherein the recovery indication is sent by the upper application after determining that the second process group releases the video memories of the plurality of GPUs; under the condition that metadata of a target virtual address in the plurality of first virtual addresses comprises a reference relationship between a second virtual address in a second process and the target virtual address, the first process provides a first export value corresponding to the second virtual address and a target physical address to the second process, the target physical address is used for exporting the video memory from the second process, and the target virtual address and the target physical address are mapped.

This application discloses a method and apparatus for managing video memory blocks, belonging to the field of data processing technology. The method includes: acquiring the occupancy status and access activity curves of video memory blocks within multiple consecutive address segments; for each consecutive address segment, performing frequency domain transformation and encoding based on the occupancy status and access activity curves of the video memory blocks within the segment to generate a video memory convolution spectrum characterizing the fragmentation degree and temporal access activity of that segment of video memory; determining candidate migration regions based on the video memory convolution spectra of each consecutive address segment; determining the differential tensors of candidate video memory blocks within the candidate migration regions; performing a comprehensive evaluation based on the differential tensors of all candidate video memory blocks under preset bandwidth and quality of service constraints to generate a video memory migration plan; and calling the underlying interface to execute the batch migration of candidate video memory blocks according to the video memory migration plan.

The application provides a Gaussian ball clustering method for three-dimensional reconstruction of a highway traffic accident site, comprising the following steps: S1, collecting multi-view images of the accident site, solving camera parameters and initializing a three-dimensional Gaussian ball scene; S2, constructing a rendering training loop based on the initialized three-dimensional Gaussian ball scene, and adaptively adjusting the Gaussian ball density through cloning or splitting operations; S3, extracting the spatial and color features of the three-dimensional Gaussian ball, and constructing a multi-modal physical feature set; S4, calculating an adaptive discrete step length based on the multi-modal physical feature set, and mapping the continuous features into discrete coordinates; S5, generating a multi-modal hash index based on the discrete coordinates, and realizing continuous rearrangement of the same type of Gaussian ball in the display memory through GPU parallel sorting; S6, performing physical conservation formula parameter aggregation on the same type of Gaussian ball continuously stored in the display memory, releasing redundant display memory and simplifying the three-dimensional Gaussian ball set; and S7, completing all training iterations, and exporting the final simplified three-dimensional Gaussian ball set into a general PLY format file.

The application provides a perspective image reconstruction method and a readable storage medium. The perspective image reconstruction method comprises the following steps: setting a plurality of calculation nodes in series based on a perspective image reconstruction process, and the plurality of calculation nodes work in parallel; inputting a view data to a first calculation node of the plurality of calculation nodes in a time period; and outputting a perspective image by a last calculation node of the plurality of calculation nodes. In this way, each calculation node operates independently and does not interfere with each other, and each calculation node processes one view data at a time, the data amount involved in each execution is small, the scheduling is convenient, the view data can be discarded immediately after being used, and the consumption of memory and display memory is reduced. Thus, the problem that the reconstruction method in the prior art has a long delay, a low refresh frequency and cannot meet the real-time requirement of the perspective image in special occasions is solved.

The application discloses a liquid crystal display module, a liquid crystal display control method and device, and applies to the field of display control, and comprises the following steps: obtaining to-be-displayed content of a liquid crystal display screen; filling a dot matrix image corresponding to the to-be-displayed content in a pre-allocated virtual display memory; and performing a full-screen refreshing operation on the liquid crystal display screen according to the dot matrix image, so as to present the to-be-displayed content on a display area of the liquid crystal display screen. Through the application, the problem of complex coordinate calculation work and display area flickering caused by cross-page display of the liquid crystal display screen is at least solved to a certain extent.

A multi-screen interaction method, device, system and storage medium solve the problem that private customization cannot be realized and display efficiency is low when multi-screen display in the multi-screen interaction process in the prior art. The multi-screen interaction method is used for a multi-screen interaction system including a display chip, a display memory, a main display screen and a secondary display screen. The display memory includes a plurality of layer display memory areas, a main display memory area and a virtual display memory area. The plurality of layer display memory areas respectively store graphical data of a plurality of application programs. The main display memory area stores synthesis data of the plurality of application programs. The plurality of application programs include a target application program. The multi-screen interaction method includes: determining a secondary display screen selected by a user and screen projection content based on a screen projection operation of the user; when the screen projection content is the target application program, storing graphical data of the target application program in the virtual display memory area; and sending the graphical data in the virtual display memory area and the synthesis data in the main display memory area to the secondary display screen and the main display screen respectively for display.

According to the large rotary equipment assembling unbalance optimization method based on GPU parallel acceleration and the video memory control technology, inherent attributes of a rotor are stored in a CUDA constant memory, a dynamic combination batch scheduling mechanism is adopted, and GPU computing resources can be fully utilized in a high-dimensional optimization space of million-level phase combination; and while the calculation accuracy is ensured, the use pressure of a video memory is remarkably reduced, kernel operation interruption and system crash are effectively avoided, and stable execution of a large-scale parallel calculation task is ensured.

The application discloses an LCD-TFT controller, and relates to the technical field of display control, which mainly comprises a direct memory access module (DMA), two-layer image processing units, a blending module (Blend) and a dithering module (Dither). The image processing unit of each layer comprises a 2x1 selector, a color look-up table (CLUT), a pixel format converter (PFC) and a matte color changing module. The PFC further comprises a YUV2RGB module, which is used for converting YUV format image source data into RGB format. The matte color changing module is used for selectively realizing the functions of matte and color changing under the control of an enabling signal and outputting. Compared with the traditional LCD-TFT controller, the application supports YUV color format, realizes the functions of automatically identifying color ranges and changing colors, and reduces the demand of an MCU on the size of a display memory.

The present disclosure provides an extended video memory-based model training method, a model inference method, an apparatus, a device, a medium and a product, relating to the technical field of storage devices. The extended video memory-based model training method comprises: sending a first DSM instruction to an SSD controller, and performing multiple rounds of iterative training by a GPU. Each round of iterative training comprises multiple processing stages. The processing stages comprise: the SSD controller loading data units before calculation corresponding to hierarchical identifiers to the GPU, and the GPU calculating calculation units based on the data units. The first DSM instruction marks the priority of the calculation units. The GPU unloads the calculation units to the SSD. The SSD controller unloads the calculation units of high priority to SLC and unloads the calculation units of low priority to QLC. Through the technical solution of the present disclosure, the access delay of high-sensitive data is reduced without increasing the hardware cost, and the invalid erasing and writing of SLC is reduced to prolong the service life of the SSD.

The application discloses a kind of based on bird's-eye view's video memory efficient three-dimensional target detection method and device, and the application is based on reversible main stem network and forward-backward view conversion module composition image BEV feature extraction network.Reversible main stem network uses reversible architecture, by recalculation intermediate parameter to minimize memory occupation.Forward-backward view conversion module can carry out fine processing to the BEV feature extracted in front view projection, generate denser, more accurate BEV representation.Ring view RGB image is input to image BEV feature extraction network, and the BEV feature of image is obtained;Laser point cloud is input to point cloud BEV feature extraction network, and point cloud feature is extracted and projected to BEV plane, and the detection result of three-dimensional object is obtained after image and point cloud BEV feature are spliced by BEV encoder and carry out feature extraction by BEV decoder.

The invention discloses a method for changing the capacity of a video memory based on a printed circuit board, which can be used in the field of electronic circuits, and comprises the following steps: firstly, acquiring a first pin definition file of an original video memory chip and a second pin definition file of a target video memory chip on the printed circuit board; then, comparing the first pin definition file with the second pin definition file, and determining incompatible pins; then, based on a preset path planning rule, determining a fly line path corresponding to each incompatible pin; the path planning rule at least comprises path length control, interference source distance control and non-intersection of a plurality of fly lines; and finally, based on the fly line path, welding the target video memory chip to the printed circuit board of which the original video memory chip is removed. Therefore, redesign, proofing and PCB manufacturing are not needed, capacity expansion can be realized by using existing hardware resources, the time cost and the economic cost of video memory capacity change are greatly reduced, and the problems of long period and large investment of a traditional scheme are solved.

The invention discloses a model loading reasoning method and device, a medium and a product, which are applied to a financial science and technology scene, and the method comprises the following steps: registering a model related file in a model memory through a controller, and initializing the model related file; segmenting a model related file into a specified size through a model memory, establishing storage mapping of a video memory and a rapid memory, and feeding back model segmentation information to a controller; state monitoring information is obtained in real time through the state monitoring assembly and fed back to the controller; and when the gateway receives a model reasoning request, performing model loading reasoning through the controller according to the state monitoring information. Model related files are segmented through the model memory, and storage mapping is established, so that flexible segmentation and rapid calling of the model can be realized, and support is provided for scheduling; by monitoring state information, the controller comprehensively masters the system state to assist in decision making; and after the gateway receives the request, the controller performs model loading reasoning according to the request, so that resources can be flexibly scheduled, and the response speed, the accuracy and the resource utilization rate are improved.

The application relates to an OpenGL multi-class object uniform drawing method based on VBO internal coding, which comprises the following steps: constructing a mixed data structure containing vertex coordinates, color components, texture coordinates and object type marks, and storing the mixed data structure in the memory of a VBO in an interlaced arrangement; defining a uniform vertex shader and a fragment shader; judging whether texture mapping is needed according to the object type marks during a rendering process; and deleting the VBO object and the shader program after the rendering is completed, and releasing the display memory and the memory resources. The application links map elements with different rendering logics, realizes the uniform drawing of heterogeneous objects under a single pipeline, fully utilizes the parallel processing capacity of a GPU, avoids frequent switching of the shader, effectively saves the communication bandwidth between a CPU and the GPU, simplifies the development difficulty of multi-map element mixed rendering in a complex scene, and facilitates the horizontal expansion of a system in large-scale geographic information visualization.

Embodiments of the present application provide a display memory control method, device, equipment, system and computer storage medium. The method comprises: in response to an obtained driving upgrade request of a graphics processing unit (GPU), obtaining a pre-upgrade page table corresponding to the GPU, the pre-upgrade page table comprising a mapping relationship between a physical address and a virtual address of GPU display memory; determining a data cleaning queue corresponding to the GPU, the data cleaning queue being used for globally cleaning data in the GPU display memory; under the condition of locking the data cleaning queue, performing a driving upgrade operation on the GPU based on the driving upgrade request, obtaining a post-upgrade page table corresponding to the GPU, the post-upgrade page table comprising a mapping relationship between the physical address and the virtual address of the GPU display memory; and based on the pre-upgrade page table and the post-upgrade page table, regulating the data in the GPU display memory. In the embodiments, the driving upgrade operation of the GPU can be implemented without the need for operations such as display memory backup, shutdown and restart.

The application provides a GPU display memory and system memory cooperative management method and system of a cloud mobile platform, the method comprising: acquiring GPU display memory parameters of the GPU display memory and system memory parameters of the system memory of the cloud mobile platform in real time, and calculating a resource imbalance index and a unified stress value according to the GPU display memory parameters and the system memory parameters; constructing a unified resource view according to the GPU display memory parameters, the system memory parameters, the unified stress value and the resource imbalance index, and making corresponding cooperative scheduling decisions for the GPU display memory and / or the system memory based on the unified resource view; detecting a current session state of a user, adopting a corresponding resource allocation strategy according to the current session state, and adjusting resource allocation of the GPU display memory and the system memory through a resource smooth switching mechanism when the session state of the user is switched; and predicting an application to be started by the user based on historical behaviors of the user, and performing resource pre-allocation processing on the application to be started. The application can realize cooperative scheduling of the GPU display memory and the system memory.