IBM, NVIDIA and Citrix have worked together to publish several articles, check them out here, detailing a new reference architecture including specific product and part information for purchasing the NeXtScale System components and NVIDIA GPUs, see the NeXtScale nx360 M4 Product Guide and the reference architecture. Over the next few weeks IBM will be expanding on this topic in a series of cross-industry blogs that I would watch out for on their site, here.

IBM has been a key partner in building up reference architectures for CAD virtualisation with the likes of PTC and AutoDesk. IBM have also been collaborating closely with Citrix, NVIDIA and PLM organisations.

What I particularly like about the case studies in the reference architecture (do read them, here!) is that they are hypervisor neutral (as is XenDesktop and HDX 3D Pro – we support AWS (including GRID GPU EC2 support), Citrix CloudPlatform, VMware  vSphere, Microsoft Hyper-V and Citrix XenServer, as well as via a large number of Citrix Cloud Service Providers) and as such cover full market choice and the avoidance of vendor lock-in. HDX 3D Pro is the enterprise collection of technologies that focus on enterprise ready user experience, graphics, compression, 3D space mouse and printing and we offer it on the widest range of platforms as it can be the key differentiator in making VDI and graphics work.

The case studies covered by IBM include an aircraft manufacturer:

  • “An aircraft company has a large team of CAD designers. The cockpit design is being done by a partner in a remote location. The aircraft company decides to use remote virtual desktops for the cockpit design company so their 20 designers can access the total design. Note that the design files stay within the data center of the aircraft company thus reducing the risk of intellectual property theft. In this case, it has been determined to use dedicated GPUs and the NVIDIA GRID K2 to ensure the best performance for the CAD application. The cockpit designers in the remote location have a dedicated fast connection into the aircraft company data center”
  • This is a fairly common deployment scenario we have seen for GPU pass-through and vGPU, high-end CAD with demanding designers as end users who need performance equivalent or better than on a local workstation but with the flexibility of virtualisation, mobility and management.

The case studies also covered a publishing company:

  • “A publishing company would like all of their 250 digital content editors and magazine editors to use a VDI solution so that they can work from home or in the field using a laptop or tablet. All users are required to use stateless (non-persistent) virtual desktops for easier management and each virtual desktop is 4GB. In this case it has been determined to use hardware virtualized GPUs and the NVIDIA GRID K1. The K140Q profile with up to 32 users per server provides the best performance. In order to allow for failover, 10 compute nodes are used for a capacity of up to 320 users. Up to two of these compute nodes can fail and that still leaves 8 which can service up to 256 users (8 x 32).”
  • This is a significant market for GPU acceleration and sharing technologies such as Citrix-NVIDIA unique vGPU and XenApp GPU sharing; bringing GPU access at a lower price point to companies using less demanding 3D or traditional office applications. These applications could always be virtualised but GPU accelerations enables wider possibilities cost-effectively for mobility, lighter more mobile end points (iPads, smart phones etc) and with the benefits of secure VDI management and disaster provision. There is lots of research such as that from CIMdata who IBM and Citrix have worked with on the business case for such customers, see here.


These reference architectures don’t just appear out of thin air. We spend months, jointly stress-testing servers, their BIOS configurations and compatibility with vendors such as IBM whilst these technologies are in development. This process allows both IBM, NVIDIA and us the opportunity to modify and optimise architecture from the hardware up. Our hardware certification guys often blog about these projects and new servers, as they have done with the IBM servers for this reference architecture. I’d recommend reading their blog on the IBM servers, here, for details on the server but also the testing process. Their QA methodology and culture is also an interesting read, see here.

IBM join other vendors who have recently invested in reference architectures, particularly with a graphics focus for Citrix XenDesktop and XenApp such as:

  • Dell, read about the reference architectures here and about Dell’s heavy investment in professional graphical platforms, here
  • HP reference architectures for graphical deployments are a good read, available now here, for anyone trying to decide upon a solution to deliver anything graphical or CAD related including remotely. Although the white paper refers to their WS460c Gen8 platform most of the information is relevant to anyone considering GPU usage. n particular the give a really nice overview of the technologies that complements our own, if you are looking for a vendor agnostic description of what all those VMWare/Microsoft/Citrix vSGA, vGPU, vDGA acronyms mean this helps!
  • Nutanix is proving to be a very popular choice for use in conjunction with graphical workloads. The converged architecture and IOPs handling capabilities for VDI deployments are proving popular with customers because of the ease of management, IOPs handling and scalability. Nutanix announced reference architectures this year for both Hyper-V and vSphere. Making the solution one worth looking at for those interested in low cost software-GPU acceleration, XenApp sharing on vSphere or high-end design usage of XenDesktop on vSphere with vDGA.