On March 4th, NVIDIA announced some big changes in both their licensing model as well as their pricing. At their annual GPU Technology Conference (GTC) in San Jose (CA), NVIDIA informed all the GRID enthusiasts and GRID curious people out there about these big changes. Now, you may think, “Why do you consider this a big change?” Let me clarify on this.
With GRID 2.0 NVIDIA decided to introduce a software licensing model, which wasn’t really appreciated by existing NVIDIA GRID 1.0 (K1 & K2) customers. But ever since NVIDIA announced GRID 1.0 End Of Sale (EOS) last month (March 2016) customers became stressed out. Everyone started calculating and came out with somehow ‘difficult’ business cases, so to speak. Today, things have changed, and they changed in a good way!
NVIDIA has simplified their GRID offering, simplified the business model as well as the License Management. It is tempting to compare the old model and pricing with the newly announced ones, but I won’t go there. The old model and pricing is ‘water under the bridge’, let’s focus on the current situation. I do have comparison charts if you are interested, just let me know. Before we dig into what’s new, let me refresh your memory about NVIDIA GRID.
NVIDIA GRID
In my early years in IT I was introduced with Server Based Computing (SBC), a big leap coming from the concept of ‘super computers’ and Character User Interface (CUI) based terminals. SBC, and later on VDI, brought us flexible and easy to manage environments which brought almost all user functionality to their endpoints. People were able to perform their job regardless of their location, device and connection. Work became less of ‘a place’ and more ‘a thing you do’.
Work isn’t a place, its something you do
With virtual desktops many benefits were introduced but there was always a small part of the application set which didn’t fit in. Applications requiring rich graphics were not the best candidate for SBC/VDI environments. We used reverse seamless techniques to cope with these kind of challenges. This resulted in a centralized environment with virtual machines providing secure access to applications and data where some endpoint devices provided extra functionality with graphic intense applications. These endpoints are equipped with a GPU. This is just one recognizable example.
Almost three years ago NVIDIA introduced GRID 1.0 with the K1 and K2 GPU’s. These Keplar cards brought us lots opportunities with GPU virtualization. Virtualization techniques we all know from server virtualization became available for GPU’s. Within a virtualized environment we already used pass-through configurations for specific use cases. These are 1:1 solution. One user could leverage one GPU. Citrix HDX 3D Pro and Teradici Remote Workstation cards were used as well.
With the NVIDIA vGPU techniques this changed. With a virtualized GPU, all of a sudden, there is a 1:N solution. Based on the vGPU profile you are able to maximize up to 16 concurrent users per GPU. With boards containing multiple GPU’s this results into 32 concurrent users.
Time Slicing
Another huge benefit with GRID is the fact you are able to actually use a GPU to its full potential. I know designers use their expensive physical workstations to their full potential, but in the end of the day, they are humans just like me. Every now and then they need a cup of coffee, take a bathroom break or end up in a discussion with a fellow designer on which design is better. My point is, the GPU isn’t used 100% of the time. This is where GRID benefits with something called time slicing. Time slicing is a well-known technology that hypervisors (e.g. vSphere, XenServer, Hyper-V) use to share physical resources between virtual machines. NVIDIA GRID uses this same technology to share the GPU between multiple virtual machines. So, while the designer is taking his lunch break or in a meeting someone else is able to use the GPU, which won’t be possible in a physical world. This time slicing allows the distribution of pooled resources based on actual need. NVIDIA GRID uses time slicing to share the 3D engine between virtual machines. But it even gets better. Even when the GPU (and CPU) is under load a designer isn’t zooming or rotating all of the time. The CPU and GPU load results in spikes, with free time in between. This free time is usable to other workloads on different VM’s making the GRID model extremely efficient.
Erik Bohnhorst (Performance Engineering Lead Architect at NVIDIA) presented a great session on Time Slicing at the NVIDIA GRID Days. (I short blog describing these first GRID Days can be found here.)
NVIDIA GRID Benefits
NVIDIA GRID changed the game of 2D/3D Graphics in combination with virtual desktop environments and allows people to perform their jobs in a more clever way with a rich User eXperience (UX)! Speaking of UX, the NVIDIA GRID model isn’t just for the designers and power users. Everybody can leverage from a GPU, receiving a great UX. Applications like Microsoft Office or web applications hugely benefit from a GPU. So whether you are a designer or a knowledge worker using Office applications most of the day, they all deserve a workstation class UX. User eXperience is King!
So, why do we need a GPU in each and every virtual desktop (SBC and VDI)?
- Flexibility
Work isn’t a place, it’s something you do. - Access
A virtual desktop delivers applications and data on any device, any location and any connection. - Security and Control
Information (IP) is kept within the secured datacenter. - Application Integration
Applications and their data is kept close to each other, within the datacenter. There is no need to copy large data sets back and forth towards different locations. - BYO
A virtual, centralized desktop enables secure access to applications and data in a BYO concept. - Disaster Recovery
Multi-side and/or multi-datacenter on different geo locations is easier to accomplish.
So we all benefit from a GPU in our virtual desktops and we all should have one, but what about the money? First, the licensing model.
Licensing Model
The previous licensing model contained three different license models. There was a Virtual PC, a Virtual Workstation and a Virtual Workstation Extended license available. These licenses where only available as Perpetual License. As mentioned, I won’t compare the previous models with the ones announced. The old model is ‘water under the bridge’.
So, what’s new you wonder?
First of all, NVIDIA introduced an Annual Subscription license model next to the existing Perpetual license model. Second, the Virtual Workstation Extended is gone, the vGPU profiles and support (NVIDIA CUDA, OpenCL and GPU Pass-through) available in this Extended license are now available with the Virtual Workstation license. And third, a new license is introduced with the Virtual Applications license.
As of today the following licensing models are available.
NVIDIA GRID Virtual Applications
The Virtual Applications model is introduced for those organizations who use RDSH or Citrix XenApp environments. These shared hosted desktop environment can really benefit a GPU. Both for the ever important User eXeperience (UX) as well as User Density. A lot of Windows applications e.g. Adobe Photoshop and Microsoft Office application can leverage a GPU the UX improves drastically. And since we don’t bother the CPU that much with instructions which are ‘better off’ being processed at a GPU the User Density increases.
NVIDIA GRID Virtual PC
The Virtual PC is intended for users who want a virtual desktop and need great UX. Delivering applications, browsers and HD video with the same experience as people are used to when they are using a physical PC.
NVIDIA GRID Virtual Workstation
The Virtual Workstation is available for those who require a professional graphics virtual workstation, the Power Users, Designers and Engineers. The ones designing the next Boeing or doing all the cool stuff at BMW. These man and woman use Adobe CS6, AutoDesk AutoCad, Esri and Catia on a day to day basis. These are the ones creating a new world.
Let’s compare these models to each other.
Licensing Prices
As mentioned NVIDIA announced new GRID prices. Today NVIDIA offers two business models with Perpetual Licenses and Annual Subscription, both with their own pricing, off course. “So, what about these prices already”, I hear you thinking.
With these new business models and prices organizations gain ‘freedom of choice’. Obtain an Annual Subscription, with SUMS included, just for one year or invest some extra and get a Perpetual license in which you’ll be GRID Enabled for life. With Perpetual licenses SUMS is required only the first year, after the first year it’s optional.
Also be aware there is an Education Pricing program. Use it to your advantage, if applicable.
Annual Subscription provides software entitlement, Support, Updates, Maintenance (SUM) for 1 year. If you prefer more, you’ll get a 3 year entitlement and pay upfront.
SUMS
Last but not least, the SUMS offering is simplified. With the introduction of GRID 2.0 SUSM was offered in 2 flavors, Basic and Production. The only difference between those 2 offerings was phone support. Well, they opened up some extra landlines. Everybody who buys SUMS or an Annual Subscription is allowed to give the green team a call.
Take a look at the NVIDIA GIRD Packaging and Licensing Guide for more details or contact me if you have any questions on GRID.
Final Thoughts
NVIDIA has brought a great second generation of GRID cards with the M60 and M6 boards. Nevertheless the introduction of licenses didn’t brought them the traction which everyone was hoping for. I truly believe in GPU for the masses. Each and every user deserves a workstation style user experience with their virtual desktops. With the introduction of the new licensing model and pricing this could be more in reach, but will it be enough, I’m not sure to be honest. The first few business cases will prove.
NVIDIA has raised the bar for GPU virtualization. Competitors like AMD, with the FirePro S7150/S7150 x2, are closing in. Maybe their offerings are ‘good enough’. Only people (users) can tell. I know quadro certified can be a huge deal, but for lots of environments this isn’t a requirement if you are just pursuing a ‘good enough’ User eXperience.
Nevertheless, the NVIDIA GRID team achieved something new, something great and explored new grounds. They’ve got a track record and proved they do a great job. Competition will drive all competitors to aim higher, which is beneficial to all of us.