-
公开(公告)号:US12259822B2
公开(公告)日:2025-03-25
申请号:US17242028
申请日:2021-04-27
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US20190294556A1
公开(公告)日:2019-09-26
申请号:US16442138
申请日:2019-06-14
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
IPC: G06F12/1009 , G06T1/60 , G06T1/20 , G06F9/455 , G06F13/16
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US11853796B2
公开(公告)日:2023-12-26
申请号:US16063618
申请日:2016-12-09
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Jackson Dsouza
CPC classification number: G06F9/4887 , G06F9/4818 , G06F9/4881 , G06T1/20
Abstract: The operation of a GPU is controlled based on one or more deadlines by which one or more GPU tasks must be completed and estimates of the time required to complete the execution of a first GPU task (which is currently being executed) and the time required to execute one or more other GPU tasks (which are not currently being executed). Based on a comparison between the deadline(s) and the estimates, the operating parameters of the GPU may be changed.
-
公开(公告)号:US20210263857A1
公开(公告)日:2021-08-26
申请号:US17242028
申请日:2021-04-27
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
IPC: G06F12/1009 , G06F13/16 , G06F9/455 , G06T1/20 , G06T1/60
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US10366012B2
公开(公告)日:2019-07-30
申请号:US15367370
申请日:2016-12-02
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and while each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US11847492B2
公开(公告)日:2023-12-19
申请号:US16063518
申请日:2016-12-09
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Jackson Dsouza
CPC classification number: G06F9/4887 , G06F9/4881 , G06T1/20 , Y02D10/00
Abstract: The operation of a GPU is controlled based on one or more deadlines by which one or more GPU tasks must be completed and estimates of the time required to complete the execution of a first GPU task (which is currently being executed) and the time required to execute one or more other GPU tasks (which are not currently being executed). Based on a comparison between the deadline(s) and the estimates, context switching may or may not be triggered.
-
公开(公告)号:US11016906B2
公开(公告)日:2021-05-25
申请号:US17013720
申请日:2020-09-07
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
IPC: G06F12/1009 , G06F13/16 , G06F9/455 , G06T1/20 , G06T1/60
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US10802985B2
公开(公告)日:2020-10-13
申请号:US16442138
申请日:2019-06-14
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
IPC: G06F12/1009 , G06F13/16 , G06F9/455 , G06T1/20 , G06T1/60
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
公开(公告)号:US20170161204A1
公开(公告)日:2017-06-08
申请号:US15367370
申请日:2016-12-02
Applicant: Imagination Technologies Limited
Inventor: Dave Roberts , Mario Sopena Novales , John W. Howson
IPC: G06F12/1009 , G06T1/60 , G06T1/20 , G06F9/455
CPC classification number: G06F12/1009 , G06F9/45558 , G06F13/16 , G06F2009/45583 , G06F2212/1024 , G06F2212/1052 , G06F2212/151 , G06F2212/152 , G06F2212/651 , G06F2212/657 , G06T1/20 , G06T1/60
Abstract: A method of GPU virtualization comprises allocating each virtual machine (or operating system running on a VM) an identifier by the hypervisor and then this identifier is used to tag every transaction deriving from a GPU workload operating within a given VM context (i.e. every GPU transaction on the system bus which interconnects the CPU, GPU and other peripherals). Additionally, dedicated portions of a memory resource (which may be GPU registers or RAM) are provided for each VM and whilst each VM can only see their allocated portion of the memory, a microprocessor within the GPU can see all of the memory. Access control is achieved using root memory management units which are configured by the hypervisor and which map guest physical addresses to actual memory addresses based on the identifier associated with the transaction.
-
-
-
-
-
-
-
-