gpu technology develops rapidly as cpu does. with the virtualization technology evolves, more and more users start to deploy virtual desktop solution, expecting to achieve similar display performance as physical machines.
the maturity of gpu virtualization technology benefits more and more applications, and become an essential requirement.
convergence of gpu and virtualization technology becomes a hot topic in vdi industry. it greatly reduces the cost for graphic users. meanwhile, data security is also improved.
traditionally, virtual machine simulates standard vga driver by cpu. the vm only supports basic display capability and suitable for weak display requirement scenarios.
to introduce true cpu to vm, there are different ways such as api forwarding, direct pass-through and full gpu virtualization.
virtualization method | introduction | advantage | disadvantage |
api forwarding | hypervisor accepts vm’s display request, and forward to driver of graphic card | easy implementation on vm side easy sharing of graphic card among multiple vms | without true graphic card, vm can not utilize hardware acceleration some programs may not compatible |
direct pass-through | hypervisor allocates graphic card to designated vm by io mapping | similar performance as true machine, with better compatibility | graphic card can not be shared among multiple vms |
full gpu virtualization | graphic card driver of hypervisor provides dynamic resource dispatching to multiple vms | good performance excellent compatibility shared graphic card among multiple vms |
|
kvm is common virtualization technology for linux platform. intel implements gpu full virtualization via gvt-g technology. kvmgt is an open source program of gvt-g. it supports intel 4th generation (haswell) or upper (broadwell, skylake) gpu. it introduces better 3d acceleration capability for vm without external independent graphic card. in most cases, more than 80% of display performance in local mode can be achieved for vm.
embedway skyline gpu virtualization solution
embedway skyline high density media processing platform targets for high performance computing applications, such as video/picture encoding/decoding, transcoding, image processing and analyzing etc.
skyline media processing platform adopts latest intel skylake 6th generation core i7 or xeon e3 processor. integrated with iristm pro graphics p580, it supports gpu full virtualization, achieving great improvement of vm display performance. it also provides basic 3d graphic processing capability and hardware mjpeg/h.264/h.265 encoding for vm desktop. the platform greatly offloads cpu loading as to allocates more resources to vms. the cpu gpu computing architecture is quite suitable for vdi applications with media processing, like 3d, video, and image processing.
skyline media processing platform is 2u height unit with 4 processing nodes, integrating 16 skylake e3/i7 processors. it is characterized as high density and high performance and price ratio. take i7-6785r or e3-1585lv5 for example, each gpu supports sharing for 8 vms (windows), and 128 vms on a single unit.
embedway skyline high density media processing platform maximizes hardware capability of intel cpu and gpu. by adopting kvmgt technology, it provides excellent display performance for vm, improves vm applications compatibility, and reduces deployment and operation cost.
