Nvidia patents method to fix uneven time sharing for virtual graphics cards

1. A method comprising: identifying, by at least one processing device, a plurality of virtual processing units associated with a physical processing unit, wherein each of the plurality of virtual processing units is associated with a virtual machine of a plurality of virtual machines, each of the plurality of virtual machines being allocated predefined discrete time slices by a hypervisor to execute computing tasks using respective virtual processing units in round-robin order during respective allocated predefined discrete time slices; obtaining, by the at least one processing device and for a first virtual machine of the plurality of virtual machines, a first overhead time value associated with the first virtual machine during a first predefined discrete time slice using a first virtual processing unit of the plurality of virtual processing units; obtaining, by the at least one processing device and for a second virtual machine of the plurality of virtual machines, a second overhead time value associated with the second virtual machine during a second predefined discrete time slice using a second virtual processing unit of the plurality of virtual processing units; determining, by the at least one processing device and based on the first overhead time value associated with the first virtual machine, whether the second overhead time value associated with the second virtual machine satisfies a compensation threshold criterion, wherein the compensation threshold criterion is satisfied if a difference between the second overhead time value associated with the second virtual machine and the first overhead time value associated with the first virtual machine is greater than a respective assigned execution time period of the second virtual machine; and responsive to determining that the second overhead time value associated with the second virtual machine satisfies the compensation threshold criterion, causing, by the at least one processing device, the hypervisor to allocate a third predefined discrete time slice to the second virtual machine after the second predefined discrete time slice based on the difference between the second overhead time value associated with the second virtual machine and the first overhead time value associated with the first virtual machine, wherein causing the hypervisor to allocate the third predefined discrete time slice to the second virtual machine comprises executing at least one additional computing task of the second virtual machine immediately following the second predefined discrete time slice and prior to allocating a predefined discrete time slice to execute any other computing task corresponding to any other virtual machine of the plurality of virtual machines.