virtualbox.org

> emulators are really not worth it, besides their theoretical security advantages as compared to Full virtualization like vbox?
emulators are worth it for some usecases, where you needs to emulate another architecture - x86 on ARM mobile phone, or Sony PlayStation on x86 PC or 64-bit x64 CPU on 32-bit x86 CPU.

>So in Virtualbox, the real cpu is used but other aspects like hdd, soundcard and graphics are emulated right?
Yes.
Although some components are para-virtual now (VirtualBox Graphics Adapter with it's 3D Acceleration, VirtIO para-virtual network, Guest Control). BIOSes are also emulated.

>Tech like VT-d is supposed to allow VGA passthrough so the graphics card of the host can be safely used
Yes, theoretically. In VBox 4.1 it is currently limited to pass-though of NICs, not VGA cards.

-Technologov

Technologov wrote: emulators are worth it for some usecases, where you needs to emulate another architecture - x86 on ARM mobile phone, or Sony PlayStation on x86 PC or 64-bit x64 CPU on 32-bit x86 CPU.

I've read that something like a PS3 emulator would need obscene amounts of processing power to work: since it's 3.2Ghz it would need a rig with a 32Ghz processor to be functional. Which seems a little far off in the future as of now. So emulation is great for bringing back old games but can never be used to create a viable alternative to consoles of any current era since there will always be a need for much larger amounts of power at least 10 fold of the platform at the time.

I wonder if iPhone game devs would have similar demands to create their work


When qemu is working with a x86 on an x86 platform it functions as a virtualizer with dynamic binary translation. Does it do the same when the guest is x64 on an x86 host? If it is the case then it would be fast enough and thenVBox could have an emulation mode for this OS arch -x64 on x86 - combination without the need for VT-x. (I don't get why Intel doesn't make this a basic feature until now )

In the essay emulation performance is listed as 0.1%-5% of normal speed. To put this in perspective, let's say an operation that takes 10 seconds to complete on the real host cpu, would it take 200 seconds on an emulator? - in other words emulation is 20x slower (?)

Emulation, if fast, go to maybe 10% of host speed. 3 GHz processor can at best emulate a 300 MHz processor. Slow emulators (such as Bochs) have a penalty of 400x compared to host performance, not 10x.

So emulation is great for bringing back old games but can never be used to create a viable alternative to consoles of any current era since there will always be a need for much larger amounts of power at least 10 fold of the platform at the time.

Yes, this is true.

>When qemu is working with a x86 on an x86 platform it functions as a virtualizer with dynamic binary translation.

I think the differentiating point is kernel driver. Doing 'dynamic binary translation' of x86-on-x86 without kernel driver is still en emulation, and performance is slow. (10% at best of host speed). Using even primitive kernel driver (kqemu, VirtualPC) jumps performance to above 60% of host's speed on many workloads.

Again, many thanks. My current version landed at 934 MB, I hope to stay under the 1000MB mark due to limitations of our web site at KTH.

Best regards
Johnny