hardware-assisted virtualization
After years of CPU vendors making virtualization painful, Intel and AMD did the honest thing and built virtualization support straight into the processor. Intel calls its version VT-x (with the instruction set named VMX); AMD calls its version AMD-V (instruction set SVM). The idea is simple to state: add a brand-new operating mode to the CPU whose entire job is to host guests cleanly, so the hypervisor no longer needs binary translation or a modified guest. It just runs unmodified guest operating systems and lets the hardware catch what needs catching.
Concretely, these extensions add two execution contexts. Root mode is where the hypervisor runs - the most privileged place, in full control. Non-root (guest) mode is a faithful copy of the normal privilege structure, BUT under the hypervisor's thumb: the guest's kernel can run in its own ring 0 believing it is fully privileged, while the hardware quietly diverts the dangerous moments to the host. The hypervisor configures, per virtual CPU, a data structure the manual calls the VMCS (Virtual Machine Control Structure on Intel; VMCB on AMD): a control block that says which guest operations should kick back to the host, and that stores the guest's entire register state so a guest can be paused and resumed exactly. To start a guest you execute a VM-entry; when the guest hits a configured condition, the CPU performs a VM-exit back into the hypervisor.
Why it changed everything: hardware assistance fixed the Popek-Goldberg gap directly - now the dangerous instructions DO cause a controlled exit, so plain trap-and-emulate works on x86 again, with no need to rewrite or paravirtualize the guest. It made running unmodified Windows or Linux as a guest fast and routine, and it is the bedrock under KVM, modern Hyper-V, ESXi, and the cloud. The honest caveat is that each VM-exit still costs hundreds to thousands of cycles, so the engineering goal shifted to making exits RARE - which is exactly why companion features like nested paging exist.
On Intel, a hypervisor sets up a VMCS for a virtual CPU, then runs the instruction vmlaunch to do a VM-entry. The guest runs at native speed until, say, it accesses an emulated device register; the hardware then performs a VM-exit, and the hypervisor reads the VMCS to learn exactly why and what to emulate.
VT-x/AMD-V add a guest mode plus a VMCS so unmodified guests run fast and exit only when needed.
Hardware assistance does not make virtualization free - every VM-exit still costs real cycles. Early VT-x was sometimes SLOWER than a good binary-translation hypervisor; the win came once exits were made rare and memory virtualization moved into hardware too.