NVIDIA
Grid 2.0

Should be linux supported for K1/K2 vGPU (shared) ?

Yes, we need it for current deployment of K1/K2.
No, we are preparing to use linux only with new M60/M6 and new NVidia CCU licenses.
Irrelevant, we do not use linux at all. What is linux ?
Ah, right.
Ah, right.

-=Tobias

#16
Posted 10/07/2015 02:19 PM   
[quote="mcerveny"][quote="Jason_Southern"][quote="mcerveny"] Will it be available to K1/K2 boards (eg. [b]linux support[/b]) ? [/quote] No - GRID 2.0 is Maxwell (M60/6) only and future architectures. It will not be added to Kepler. If you require Linux or 4K displays in vGPU, it's GRID 2.0 only. [/quote] Why NVidia does not support linux guest for K1/K2 vGPU sharing (there should not be any technical problem to support it for K1/K2) ? [/quote] [b]Yes, we can![/b] (unofficially) I tried to enable linux centos6.4 3D support (Xorg/NV-GLX) for K2 [b]vGPU K260Q[/b]. I was successful in 20 minutes. I tested glxinfo, glxgears, GLMark and Unigine Heaven. All tests worked ok (4 parallel linux). Ok, NVidia, your turn. [img]http://s33.postimg.org/6mjhqr4xb/linux_nvidia_grid_vgpu.jpg[/img] M.C>
mcerveny said:
Jason_Southern said:
mcerveny said:
Will it be available to K1/K2 boards (eg. linux support) ?

No - GRID 2.0 is Maxwell (M60/6) only and future architectures. It will not be added to Kepler. If you require Linux or 4K displays in vGPU, it's GRID 2.0 only.

Why NVidia does not support linux guest for K1/K2 vGPU sharing (there should not be any technical problem to support it for K1/K2) ?

Yes, we can! (unofficially)
I tried to enable linux centos6.4 3D support (Xorg/NV-GLX) for K2 vGPU K260Q. I was successful in 20 minutes. I tested glxinfo, glxgears, GLMark and Unigine Heaven. All tests worked ok (4 parallel linux). Ok, NVidia, your turn.

Image

M.C>

#17
Posted 10/07/2015 07:46 PM   
My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.
My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.

-=Tobias

#18
Posted 10/07/2015 10:40 PM   
[quote="tjkreidl"]My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.[/quote] Not everyone in the planet needs windows... Mac support would also be great if it was possible..
tjkreidl said:My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.


Not everyone in the planet needs windows... Mac support would also be great if it was possible..

#19
Posted 10/09/2015 02:18 PM   
My understanding now is that there is no current licensing model how GPU passthrough via an application like XenApp (as opposed to a user / vGPU combination) will be handled in GRID 2.0 and may take weeks to sort out. I have heard that licensing information, at least in general, may be released as early as sometime next week. The license server itself will apparently be similar to the one Citrix uses, namely a product from Flexera, but will have to be run totally separately as a Windows instance specifically to handle GRID 2.0 licensing. As to Macs, Apple refuses to allow MacOS to be virtualized and marketed. I guess they make too much money on their hardware sales to want to lose control over that.
My understanding now is that there is no current licensing model how GPU passthrough via an application like XenApp (as opposed to a user / vGPU combination) will be handled in GRID 2.0 and may take weeks to sort out. I have heard that licensing information, at least in general, may be released as early as sometime next week. The license server itself will apparently be similar to the one Citrix uses, namely a product from Flexera, but will have to be run totally separately as a Windows instance specifically to handle GRID 2.0 licensing.

As to Macs, Apple refuses to allow MacOS to be virtualized and marketed. I guess they make too much money on their hardware sales to want to lose control over that.

-=Tobias

#20
Posted 10/09/2015 10:41 PM   
Sorry, duplicate post, please ignore.
Sorry, duplicate post, please ignore.

-=Tobias

#21
Posted 10/09/2015 10:42 PM   
[quote="tjkreidl"]My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.[/quote] It's already supported in K1/2 when used in passthrough. Linux vGPU is a 2.0 feature.
tjkreidl said:My vote to add official support for K1/K2 guest Linux VMs, please!!! Not everyone on the planet needs/wants 4k displays.


It's already supported in K1/2 when used in passthrough.

Linux vGPU is a 2.0 feature.

Jason Southern, Regional Lead for ProVis Sales - EMEA: NVIDIA Ltd.

#22
Posted 10/10/2015 08:55 PM   
[quote="tjkreidl"]My understanding now is that there is no current licensing model how GPU passthrough via an application like XenApp (as opposed to a user / vGPU combination) will be handled in GRID 2.0 and may take weeks to sort out. I have heard that icensing information, at least in general, may be released as early as sometime next week. The license server itself will apparently be similar to the one Citrix uses, namely a product from Flexera, but will have to be run totally separately as a Windows instance specifically to handle GRID 2.0 licensing. [/quote] Licensing system is available and GA today. Anyone that trials 2.0 will see how it can be deployed. It's based on FlexNet from Flexera, is either a windows or linux based server and we're looking into developing a virtual appliance too. Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count. [quote="tjkreidl"] As to Macs, Apple refuses to allow MacOS to be virtualized and marketed. I guess they make too much money on their hardware sales to want to lose control over that.[/quote] Correct, It's all about controlling the user experience. Apple only allows OS X to be deployed on Apple hardware, which is why VMware Fusion can virtualise OS X and maintain the EULA. Since Apple no longer produce server hardware, it's limited. Whlist we're all well aware of the Hackintosh movement, Enterpise businesses such as Citrix & VMware aren't going to circumvent another vendors license restrictions.
tjkreidl said:My understanding now is that there is no current licensing model how GPU passthrough via an application like XenApp (as opposed to a user / vGPU combination) will be handled in GRID 2.0 and may take weeks to sort out. I have heard that icensing information, at least in general, may be released as early as sometime next week. The license server itself will apparently be similar to the one Citrix uses, namely a product from Flexera, but will have to be run totally separately as a Windows instance specifically to handle GRID 2.0 licensing.


Licensing system is available and GA today. Anyone that trials 2.0 will see how it can be deployed.

It's based on FlexNet from Flexera, is either a windows or linux based server and we're looking into developing a virtual appliance too.

Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count.


tjkreidl said:
As to Macs, Apple refuses to allow MacOS to be virtualized and marketed. I guess they make too much money on their hardware sales to want to lose control over that.


Correct, It's all about controlling the user experience.

Apple only allows OS X to be deployed on Apple hardware, which is why VMware Fusion can virtualise OS X and maintain the EULA. Since Apple no longer produce server hardware, it's limited. Whlist we're all well aware of the Hackintosh movement, Enterpise businesses such as Citrix & VMware aren't going to circumvent another vendors license restrictions.

Jason Southern, Regional Lead for ProVis Sales - EMEA: NVIDIA Ltd.

#23
Posted 10/10/2015 09:04 PM   
Jason, First off, thanks for the updated information on the licensing. Nice to see the Flexera option for Linux is also there (though Citrix requires the Windows version for XenDesktop/XenApp for some strange reason). Go figure. Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended" fits a GPU passthrough description according to [url]http://www.nvidia.com/object/grid-technology.html[/url]) and presumably it would require the full, 8 GB configuration to be able to map to one of the GPU engines, right? Or is there flexibility to use [i]less[/i] than the full engine (which seems kind of silly on the surface, but there may be real use cases that would justify this)? In any case, I've heard about some pretty scary pricing for the Virtual Workstation Extended license -- not that incomparable with the price of the device itself -- let alone if you wanted to map to both engines for two GPU passthrough instances. The statement "Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count" makes me feel [i][u]very[/u][/i] uneasy. It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. Buying into an unknown like this is IMHO a real put-off, as it is tied into unknown costs once you have committed to the initial purchase. Surely, NVIDIA should be able to come up with something more definitive (which ought to have really been thought out and worked out prior to the release). GPU passthrough is as important in many environments as vGPU. Customers should have reasonable information on what they are buying into up-front.
Jason,
First off, thanks for the updated information on the licensing. Nice to see the Flexera option for Linux is also there (though Citrix requires the Windows version for XenDesktop/XenApp for some strange reason). Go figure.

Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended" fits a GPU passthrough description according to http://www.nvidia.com/object/grid-technology.html) and presumably it would require the full, 8 GB configuration to be able to map to one of the GPU engines, right? Or is there flexibility to use less than the full engine (which seems kind of silly on the surface, but there may be real use cases that would justify this)? In any case, I've heard about some pretty scary pricing for the Virtual Workstation Extended license -- not that incomparable with the price of the device itself -- let alone if you wanted to map to both engines for two GPU passthrough instances.

The statement "Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count" makes me feel very uneasy. It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. Buying into an unknown like this is IMHO a real put-off, as it is tied into unknown costs once you have committed to the initial purchase. Surely, NVIDIA should be able to come up with something more definitive (which ought to have really been thought out and worked out prior to the release). GPU passthrough is as important in many environments as vGPU. Customers should have reasonable information on what they are buying into up-front.

-=Tobias

#24
Posted 10/10/2015 11:46 PM   
[quote="tjkreidl"]Jason, First off, thanks for the updated information on the licensing. Nice to see the Flexera option for Linux is also there (though Citrix requires the Windows version for XenDesktop/XenApp for some strange reason). Go figure. Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended" fits a GPU passthrough description according to [url]http://www.nvidia.com/object/grid-technology.html[/url]) and presumably it would require the full, 8 GB configuration to be able to map to one of the GPU engines, right? Or is there flexibility to use [i]less[/i] than the full engine (which seems kind of silly on the surface, but there may be real use cases that would justify this)? In any case, I've heard about some pretty scary pricing for the Virtual Workstation Extended license -- not that incomparable with the price of the device itself -- let alone if you wanted to map to both engines for two GPU passthrough instances. The statement "Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count" makes me feel [i][u]very[/u][/i] uneasy. It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. Buying into an unknown like this is IMHO a real put-off, as it is tied into unknown costs once you have committed to the initial purchase. Surely, NVIDIA should be able to come up with something more definitive (which ought to have really been thought out and worked out prior to the release). GPU passthrough is as important in many environments as vGPU. Customers should have reasonable information on what they are buying into up-front.[/quote] I agree with Tobias on this one. This could easily be a make or break for vGPU deployments depending on what the costs for this become. This could especially become painful for those who have already adopted the K1/K2 route that will need to upgrade infrastructure some day. More information on this would be greatly appreciated.
tjkreidl said:Jason,
First off, thanks for the updated information on the licensing. Nice to see the Flexera option for Linux is also there (though Citrix requires the Windows version for XenDesktop/XenApp for some strange reason). Go figure.

Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended" fits a GPU passthrough description according to http://www.nvidia.com/object/grid-technology.html) and presumably it would require the full, 8 GB configuration to be able to map to one of the GPU engines, right? Or is there flexibility to use less than the full engine (which seems kind of silly on the surface, but there may be real use cases that would justify this)? In any case, I've heard about some pretty scary pricing for the Virtual Workstation Extended license -- not that incomparable with the price of the device itself -- let alone if you wanted to map to both engines for two GPU passthrough instances.

The statement "Since GRID 2.0 requires a CCU, XenApp is licensed based on the users count. In the initial release this is not enforced but could be audited, however future releases will enforce the user count" makes me feel very uneasy. It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. Buying into an unknown like this is IMHO a real put-off, as it is tied into unknown costs once you have committed to the initial purchase. Surely, NVIDIA should be able to come up with something more definitive (which ought to have really been thought out and worked out prior to the release). GPU passthrough is as important in many environments as vGPU. Customers should have reasonable information on what they are buying into up-front.


I agree with Tobias on this one. This could easily be a make or break for vGPU deployments depending on what the costs for this become. This could especially become painful for those who have already adopted the K1/K2 route that will need to upgrade infrastructure some day. More information on this would be greatly appreciated.

#25
Posted 10/12/2015 01:16 PM   
[quote="tjkreidl"] Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended" [/quote] Correct, that's the version required for passthrough. [quote="tjkreidl"] It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. .[/quote] It's not undefined, every concurrent user requires a license. To use the bus analogy, you need a ticket to ride, but right now we're not putting inspectors in every coach. So, if you have a XenApp environment with 500 concurrent users connecting to GPU enabled servers, you need 500 licenses at the relevant level of features. We're looking to simplify the maangement / enforcement of this but just like an RDS CAL, you'll need a GRID 2.0 license if you're on Maxwell.
tjkreidl said:

Can you elucidate which model of licensing will be required for GPU passthrough (it seem only the "Virtual Workstation Extended"


Correct, that's the version required for passthrough.

tjkreidl said:
It's like buying a bus and being told you can fill it with as many passengers as you want, but eventually, this will be regulated and you will probably have to pay some additional and undetermined amount if you want to fill up more seats. .


It's not undefined, every concurrent user requires a license.

To use the bus analogy, you need a ticket to ride, but right now we're not putting inspectors in every coach.

So, if you have a XenApp environment with 500 concurrent users connecting to GPU enabled servers, you need 500 licenses at the relevant level of features.

We're looking to simplify the maangement / enforcement of this but just like an RDS CAL, you'll need a GRID 2.0 license if you're on Maxwell.

Jason Southern, Regional Lead for ProVis Sales - EMEA: NVIDIA Ltd.

#26
Posted 10/12/2015 05:40 PM   
[i]It's not undefined, every concurrent user requires a license.[/i] Yes, it will be defined by numbers, but not in the pricing. I would hope that the cost would not be as high as that of a vGPU user, but this remains as of this moment something that will still end up being an additional and undetermined cost.
It's not undefined, every concurrent user requires a license.

Yes, it will be defined by numbers, but not in the pricing. I would hope that the cost would not be as high as that of a vGPU user, but this remains as of this moment something that will still end up being an additional and undetermined cost.

-=Tobias

#27
Posted 10/12/2015 06:58 PM   
[quote="Jason_Southern"] It's not undefined, every concurrent user requires a license. To use the bus analogy, you need a ticket to ride, but right now we're not putting inspectors in every coach. So, if you have a XenApp environment with 500 concurrent users connecting to GPU enabled servers, you need 500 licenses at the relevant level of features. We're looking to simplify the maangement / enforcement of this but just like an RDS CAL, you'll need a GRID 2.0 license if you're on Maxwell. [/quote] Sorry to ask a question about this. How will this affect in mix-mode environments, i.e. I am thinking of having some servers without GPUs, and others with? If I only allocate 20 users for the high end grid 2.0, but have 300 users on "normal" VMs, I won't have to purchase 320 2.0 licenses correct? Sorry I am still very new to Citrix, and am trying to design and build out a XenDesktop + Grid 2.0 environment. Thank you.
Jason_Southern said:

It's not undefined, every concurrent user requires a license.

To use the bus analogy, you need a ticket to ride, but right now we're not putting inspectors in every coach.

So, if you have a XenApp environment with 500 concurrent users connecting to GPU enabled servers, you need 500 licenses at the relevant level of features.

We're looking to simplify the maangement / enforcement of this but just like an RDS CAL, you'll need a GRID 2.0 license if you're on Maxwell.


Sorry to ask a question about this. How will this affect in mix-mode environments, i.e. I am thinking of having some servers without GPUs, and others with? If I only allocate 20 users for the high end grid 2.0, but have 300 users on "normal" VMs, I won't have to purchase 320 2.0 licenses correct?

Sorry I am still very new to Citrix, and am trying to design and build out a XenDesktop + Grid 2.0 environment.

Thank you.

#28
Posted 10/13/2015 04:21 AM   
Ken31, It'd be better for an NVIDIA employee to respond, but my understanding is that this goes by [u]concurrent[/u] connections. A vGPU connection and a license token (the kind would depend on which of the three options is configured for the VM) would be consumed as soon as the VM becomes active, that is, when the VM is booted up. So assuming all 20 are "Workstation Extended" users, you'd need 20 of that variety of license. The other 300 VMs would not be tapping into any GPU resources, and hence would not require any GPU licenses at all. If you only boot up a maximum of 15 of those VMs at any given time that [u]do[/u] access the GPU, then you'd be consuming a maximum of 15 licenses. The GPU passthrough licensing sounds like it's eventually also going to count how many active passthrough [i]sessions[/i] are seen on the VM supporting it (such as on a XenApp service) but how that cost will be determined remains to be worked out. On the surface, it would appear you need a "Workstation Extended" license to begin with, as that is evidently the only one that supports GPU passthrough (again, see [url]http://www.nvidia.com/object/grid-technology.html[/url] for details).
Ken31,

It'd be better for an NVIDIA employee to respond, but my understanding is that this goes by concurrent connections. A vGPU connection and a license token (the kind would depend on which of the three options is configured for the VM) would be consumed as soon as the VM becomes active, that is, when the VM is booted up. So assuming all 20 are "Workstation Extended" users, you'd need 20 of that variety of license. The other 300 VMs would not be tapping into any GPU resources, and hence would not require any GPU licenses at all. If you only boot up a maximum of 15 of those VMs at any given time that do access the GPU, then you'd be consuming a maximum of 15 licenses.

The GPU passthrough licensing sounds like it's eventually also going to count how many active passthrough sessions are seen on the VM supporting it (such as on a XenApp service) but how that cost will be determined remains to be worked out. On the surface, it would appear you need a "Workstation Extended" license to begin with, as that is evidently the only one that supports GPU passthrough (again, see http://www.nvidia.com/object/grid-technology.html for details).

-=Tobias

#29
Posted 10/13/2015 09:25 AM   
Tobias, Thanks for the reply. I am trying to get classification through my vendor, but thought I'd post up my question anyhow. Your answer does seem logical, but sometimes licensing isn't always logical! If I do get clarification, I will post up here. Ken
Tobias,

Thanks for the reply. I am trying to get classification through my vendor, but thought I'd post up my question anyhow. Your answer does seem logical, but sometimes licensing isn't always logical!

If I do get clarification, I will post up here.

Ken

#30
Posted 10/15/2015 12:24 AM   
Scroll To Top

Add Reply