2 gamers using seperate GPUs in one overkill PC possible?

[WhiteSkyMage]

edit: Wait. No. I'm dumb. The problem isn't the graphics drivers, it's the OS itself. Windows doesn't allow multiple windows to be in focus at the same time. You'd need to use the VM solution people are talking about. This will substantially reduce the performance of any game run within the VM (don't expect modern AAA games to play acceptably in a VM, if at all). But yea. That's the problem: window focus.

 

Do you mean there is will be performance loss for the host AND for the VM user? I thought that performance from CPU will spread in 1/2 for VM and host, and since the 2 GPUs (NOT in SLI config) are connected directly to the separate monitors, the performance loss might only occur from CPU's performance output, not the GPUs. For the RAM, I am not sure how it splits up (I wanna split it 8/8 or 16/16 so she has and I have and we can both play AAA games on max with good performance (having the CPU overclocked and fans , pump at the maximum ). 

 

I just don't get, which component will get performance output loss?

[ThatAngryGnome]

Dang a guy and his sister both having interest in gaming...the perfect relationship

[WhiteSkyMage]

Dang a guy and his sister both having interest in gaming...the perfect relationship

 

Man, it's normal? Are you jealous?

[Noctus]

Why would you even want to do this ?

[ThatAngryGnome]

Man, it's normal? Are you jealous?

Lmao I'm like the freaky internet gamer of the house =P

[ionbasa]

You can use a Linux KVM virtual machine manager. It supports PCIe passthrough, assuming that your CPU supports it through VT-d and VT-x (only non K intel CPUs  :mellow:).

Create 2 VMs each running windows, assign 1 GPU to each VM, along with the peripherals.

 

If you need help, look here: http://www.howtoforge.com/howtos/virtualization/kvm

[Guest]

You can actually do that, you can run an app that deploys a OS to each one of your monitors and use your monitor as a source of usb ports, and using shadowplay would be possible that way,

[WhiteSkyMage]

You can use a Linux KVM virtual machine manager. It supports PCIe passthrough, assuming that your CPU supports it through VT-d and VT-x (only non K intel CPUs  :mellow:).

Create 2 VMs each running windows, assign 1 GPU to each VM, along with the peripherals.

 

If you need help, look here: http://www.howtoforge.com/howtos/virtualization/kvm

 

hmm, sorry man but my CPU will be K intel (since Haswell-E are 2 K and 1 X CPU)

 

 

You can actually do that, you can run an app that deploys a OS to each one of your monitors and use your monitor as a source of usb ports, and using shadowplay would be possible that way,

 

What app? Is that VM as well? 

 

 

Guys I just wanna know one thing - which component will have a disadvantage and potentially a performance loss from a virtual machine? Why would there be a problem playing AAA games on virtual machine? 

[xjohnny3589]

Dang a guy and his sister both having interest in gaming...the perfect relationship

incest

[ThatAngryGnome]

incest

Well...that escalated quickly

[WhiteSkyMage]

Hey guys, I think I found some good guide for VMs. Ty for the help. Here is it: http://www.overclock.net/t/1205216/guide-create-a-gaming-virtual-machine

[The_Evenger]

maybe if you had some sort of virtual desktop thing going on... but i don't know if you would be able to get a second mouse on the screen without it interfering with the other...

 

but i might do some testing on this now, very interesting...

[WhiteSkyMage]

Well alright,

 

Looks like the LGA 2011 CPUs all support VT-d and VT-x. Let's hope that Haswell-E will also support it. 

 

So, which MBs support VT-d right now? What's the best Virtual Machine out there I can make for gaming?

[ionbasa]

Well alright,

 

Looks like the LGA 2011 CPUs all support VT-d and VT-x. Let's hope that Haswell-E will also support it. 

 

So, which MBs support VT-d right now? What's the best Virtual Machine out there I can make for gaming?

Just about any Asus Workstation board. 

As for what host OS, it would probably be a Linux, pick just about any flavor. I would suggest Ubuntu as it has a fairly easy to use GUI.

As for what virtualization software: Xen, KVM, or Virtualbox. All are free. 

As for the paid route: Vmware ESXi or Vspere is the only option that I know of.

[asquirrel]

Do you mean there is will be performance loss for the host AND for the VM user? I thought that performance from CPU will spread in 1/2 for VM and host, and since the 2 GPUs (NOT in SLI config) are connected directly to the separate monitors, the performance loss might only occur from CPU's performance output, not the GPUs. For the RAM, I am not sure how it splits up (I wanna split it 8/8 or 16/16 so she has and I have and we can both play AAA games on max with good performance (having the CPU overclocked and fans , pump at the maximum ). 

 

I just don't get, which component will get performance output loss?

The perf loss would be for the VM user only. The non-VM user will be unaffected. The reason why is that GPUs don't virtualize well. Some tasks it will be transparent, others...massive. It depends on the application/task itself as to whether the 'doesn't virtualize well' is a showstopper or a minor annoyance.

 

The loss of performance isn't from hardware, it's from software. Well..technically it's from hardware that hasn't been created yet. In order to virtualize access to a GPU, you have to:

1. Virtualize the PCI bus.
2. Virtualize the CPU interrupt handler, and tie that into the PCI bus
3. Create an interface between the virtual interrupt handler and the physical interrupt handler, check that you aren't clobbering another signal, etc.
4. For each GPU event:
5. Take the physical signal to/from the hardware, translate it to a virtual instruction, send that through the virtual interrupt handler, virtual CPU, virtual PCI bus, to the virtual graphics card.
6. decode *back* down to the physical layer
7. execute the instruction, wrap the result into the virtual layer
8. send the virtual answer back through the virtual PCI bus, virtual CPU, virtual interrupt handler --> physical interrupt handler
9. Display physical answer on screen

It adds steps. Those steps stack up if you are doing lots of tiny operations and slow you down A LOT. If your workload gives you a smaller number of larger tasks, the impact isn't so large.

[WhiteSkyMage]

The perf loss would be for the VM user only. The non-VM user will be unaffected. The reason why is that GPUs don't virtualize well. Some tasks it will be transparent, others...massive. It depends on the application/task itself as to whether the 'doesn't virtualize well' is a showstopper or a minor annoyance.

 

The loss of performance isn't from hardware, it's from software. Well..technically it's from hardware that hasn't been created yet. In order to virtualize access to a GPU, you have to:

• Virtualize the PCI bus.
• Virtualize the CPU interrupt handler, and tie that into the PCI bus
• Create an interface between the virtual interrupt handler and the physical interrupt handler, check that you aren't clobbering another signal, etc.
• For each GPU event:
• Take the physical signal to/from the hardware, translate it to a virtual instruction, send that through the virtual interrupt handler, virtual CPU, virtual PCI bus, to the virtual graphics card.
• decode *back* down to the physical layer
• execute the instruction, wrap the result into the virtual layer
• send the virtual answer back through the virtual PCI bus, virtual CPU, virtual interrupt handler --> physical interrupt handler
• Display physical answer on screen

It adds steps. Those steps stack up if you are doing lots of tiny operations and slow you down A LOT. If your workload gives you a smaller number of larger tasks, the impact isn't so large.

If i have a CPU and MB which support VT-d and VT-x, then will the performance loss be smaller?

[asquirrel]

If i have a CPU and MB which support VT-d and VT-x, then will the performance loss be smaller?

No. Those instructions help the CPU tasks, they do nothing for the GPU tasks AFAIK.