Resource-utilization-based workload re-allocation system
A resource-allocation-based workload re-allocation system includes a computing device with a GPU processing system including GPU processing subsystems, and a GPU memory system including a respective GPU memory subsystem utilized by each GPU processing subsystem. A resource utilization monitoring and workload re-allocation system that is coupled to the computing device identifies graphics workloads that are each allocated a respective memory capacity provided by one of the GPU memory subsystems and, for each of those graphics workloads, determines a respective GPU memory subsystem utilization of the respective memory capacity allocated to that graphics workload. Based on the respective GPU memory allocation utilizations, at least some of the plurality of graphics workloads are re-allocated by the resource utilization monitoring and workload re-allocation system from a first GPU memory subsystem to at least one different GPU memory subsystem, and memory capacity provided by the first GPU memory subsystem is allocated to non-graphics workload(s).
1. A resource-allocation-based workload re-allocation system, comprising: a computing device including:
a Graphics Processing Unit (GPU) processing system including a plurality of GPU processing subsystems; and
a GPU memory system including a respective GPU memory subsystem utilized by each of the plurality of GPU processing subsystems; and
a resource utilization monitoring and workload re-allocation system that is coupled to the computing device and that is configured to:
identify a plurality of graphics workloads that are each allocated a respective memory capacity provided by one of the GPU memory subsystems included in the GPU memory system;
determine, for each of the plurality of graphics workloads, a respective GPU memory subsystem utilization of the respective memory capacity allocated to that graphics workload;
cause, based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a re-allocation of at least some of the plurality of graphics workloads from a first GPU memory subsystem included in the GPU memory system to at least one different GPU memory subsystem included in the GPU memory system; and
cause memory capacity provided by the first GPU memory subsystem to be allocated to at least one non-graphics workload.
2. The system of claim 1 , wherein the causing the re-allocation of the at least some of the plurality of graphics workloads from the first GPU memory subsystem to the at least one different GPU memory subsystem includes:
causing the reallocation of all of the graphics workloads that are allocated a respective memory capacity provided by the first GPU memory subsystem such that all of those graphics workloads are re-allocated to the at least one different GPU memory subsystem.
3. The system of claim 1 , wherein the computing device is configured to:
determine, for each of the plurality of graphics workloads, a respective GPU processing subsystem utilization of the GPU processing subsystem that utilizes the GPU memory subsystem that includes the respective memory capacity allocated to that graphics workload, wherein the at least some of the plurality of graphics workloads that are re-allocated from the first GPU memory subsystem to the at least one different GPU memory subsystem include a first subset of the plurality of graphics workloads that have each have a lower respective GPU processing subsystem utilization relative to a second subset of the plurality workloads.
4. The system of claim 1 , wherein the computing device is configured to:
cause, based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a reduction of the amount of the respective memory capacity allocated to the at least some of the plurality of graphics workloads.
5. The system of claim 1 , wherein the at least one non-graphics workload includes at least one of an Artificial Intelligence (AI) workload, a Machine Learning (ML) workload, and a Deep Learning (DL) workload.
6. The system of claim 1 , wherein the respective memory capacity allocated to each of the plurality of graphics workloads includes a frame buffer capacity, and wherein the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads includes a frame buffer utilization.
7. The system of claim 1 , wherein the GPU processing system in the computing device is configured to perform the plurality of graphics workloads and the at least one non-graphics workload at the same time.
8. The IHS of claim 7 , wherein the resource utilization monitoring and workload re-allocation engine is configured to:
determine, for each of the plurality of graphics workloads, a respective GPU processing subsystem utilization of a GPU processing subsystem that utilizes the GPU memory subsystem that includes the respective memory capacity allocated to that graphics workload, wherein the at least some of the plurality of graphics workloads that are re-allocated from the first GPU memory subsystem to the at least one different GPU memory subsystem include a first subset of the plurality of graphics workloads that have each have a lower respective GPU processing subsystem utilization relative to a second subset of the plurality workloads.
9. The IHS of claim 7 , wherein the resource utilization monitoring and workload re-allocation engine is configured to:
cause, based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a reduction of the amount of the respective memory capacity allocated to the at least some of the plurality of graphics workloads.
10. An Information Handling System (IHS), comprising:
a processing system; and
a memory system that is coupled to the processing system and that includes instructions that, when executed by the processing system, cause the processing system to provide a resource utilization monitoring and workload re-allocation engine that is configured to:
identify a plurality of graphics workloads that are each allocated a respective memory capacity provided by one of a plurality of Graphics Processing Unit (GPU) memory subsystems included in a GPU memory system;
determine, for each of the plurality of graphics workloads, a respective GPU memory subsystem utilization of the respective memory capacity allocated to that graphics workload;
cause, based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a re-allocation of at least some of the plurality of graphics workloads from a first GPU memory subsystem included in the GPU memory system to at least one different GPU memory subsystem included in the GPU memory system; and
cause memory capacity provided by the first GPU memory subsystem to be allocated to at least one non-graphics workload.
11. The IHS of claim 10 , wherein the causing the re-allocation of the at least some of the plurality of graphics workloads from the first GPU memory subsystem to the at least one different GPU memory subsystem includes:
causing the reallocation of all of the graphics workloads that are allocated a respective memory capacity provided by the first GPU memory subsystem such that all of those graphics workloads are re-allocated to the at least one different GPU memory subsystem.
12. The IHS of claim 10 , wherein the at least one non-graphics workload includes at least one of an Artificial Intelligence (AI) workload, a Machine Learning (ML) workload, and a Deep Learning (DL) workload.
13. The IHS of claim 10 , wherein the respective memory capacity allocated to each of the plurality of graphics workloads includes a frame buffer capacity, and wherein the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads includes a frame buffer utilization.
14. A method for re-allocating workloads based on resource utilization, comprising:
identifying, by a resource utilization monitoring and workload re-allocation system, a plurality of graphics workloads that are each allocated a respective memory capacity provided by one of a plurality of Graphics Processing Unit (GPU) memory subsystems included in a GPU memory system;
determining, by the resource utilization monitoring and workload re-allocation system for each of the plurality of graphics workloads, a respective GPU memory subsystem utilization of the respective memory capacity allocated to that graphics workload;
causing, by the resource utilization monitoring and workload re-allocation system based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a re-allocation of at least some of the plurality of graphics workloads from a first GPU memory subsystem included in the GPU memory system to at least one different GPU memory subsystem included in the GPU memory system; and
causing, by the resource utilization monitoring and workload re-allocation system, memory capacity provided by the first GPU memory subsystem to be allocated to at least one non-graphics workload.
15. The method of claim 14 , wherein the causing the re-allocation of the at least some of the plurality of graphics workloads from the first GPU memory subsystem to the at least one different GPU memory subsystem includes:
causing the reallocation of all of the graphics workloads that are allocated a respective memory capacity provided by the first GPU memory subsystem such that all of those graphics workloads are re-allocated to the at least one different GPU memory subsystem.
16. The method of claim 14 , further comprising:
determining, by the resource utilization monitoring and workload re-allocation system for each of the plurality of graphics workloads, a respective GPU processing subsystem utilization of a GPU processing subsystem that utilizes the GPU memory subsystem that includes the respective memory capacity allocated to that graphics workload, wherein the at least some of the plurality of graphics workloads that are re-allocated from the first GPU memory subsystem to the at least one different GPU memory subsystem include a first subset of the plurality of graphics workloads that have each have a lower respective GPU processing subsystem utilization relative to a second subset of the plurality workloads.
17. The method of claim 14 , further comprising:
causing, by the resource utilization monitoring and workload re-allocation system based on the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads, a reduction of the amount of the respective memory capacity allocated to the at least some of the plurality of graphics workloads.
18. The method of claim 14 , wherein the at least one non-graphics workload includes at least one of an Artificial Intelligence (AI) workload, a Machine Learning (ML) workload, and a Deep Learning (DL) workload.
19. The method of claim 14 , wherein the respective memory capacity allocated to each of the plurality of graphics workloads includes a frame buffer capacity, and wherein the respective GPU memory subsystem utilization determined for each of the plurality of graphics workloads includes a frame buffer utilization.
20. The method of claim 14 , further comprising:
performing, by a GPU processing system that utilizes the GPU memory system, the plurality of graphics workloads and the at least one non-graphics workload at the same time.