by Rob Farber
With the outsourcing of animation to the cloud, virtual reality becoming a web service, and hints of HPC becoming cloud-friendly, it is worth looking at how companies are rising to address these markets. One example is the Dallas, Texas company Nimbix that bills itself as, “a pure high performance computing (HPC) cloud” that provides turnkey end user HPC applications in various vertical markets including media and entertainment. In short, Nimbix enables software-as-a-service on HPC hardware so designers and animators can to interact with the Nimbix cloud without having to learn HPC skills or procure expensive infrastructure.
The Nimbix cloud distinguishes itself from AWS (Amazon Web Services) and other “cloud” providers by utilizing a Nimbix built and operated supercomputer that is based on a custom self service platform (called JARVICE) to deliver applications to remote end users. The company claims that there is no virtualization “jitter” that has traditionally been a barrier to acceptance of virtual machines in an HPC environment. More about jitter can be found in the paper. “The case of the missing supercomputer performance“. Instead, all jobs run on baremetal with Infiniband for parallel message passing (like MPI), and GPU acceleration. Nimbix is currently running NVIDIA K40 GPUs but will soon upgrade to the recently announced K80s.
GPU computing is oversubscribed as most “cloud instances” use what are called “grid” GPUs, which are designed for virtualized infrastructure for things like high density remote desktops (where 50 instances run on a given server at a time). While not of interest for HPC, these GPUs are intended to deliver desktop content to users. Nimbix on the other hand dedicates supercomputing GPUs to HPC jobs and claims to have observed a up to a 60x performance increase using a real customer file on Blender as compared to a high end Mac. Such claims demonstrate the power of their HPC solution for animation and rendering. Since Nimbix is a cloud provider, users can quickly and easily verify these claims with a minimal investment.
The answer to the question, “Why Nimbix” is pretty simple: Nimbix believes it provides a solution that is much easier to use than in-house HPC, much faster than commodity cloud for typical HPC use cases, and delivers lower overall cost as a result of less skills required and less “instances” required (when compared to commodity cloud). A big plus is that users get access to a supercomputer rather than a cluster of web services instances or a desktop PC. Nimbix is betting that the deal-maker for the animation industry (and others) is scalabilty as organizations can run a render farm on demand without having to build and manage it. (Please see our interview with Lincoln Wallen, CTO of DreamWorks to see how this fits into the current seismic changes in the animation industry.)
Ray-tracing as a service is available for sites that need real-time photo-realistic rendering. Speed is the issue and the Nimbix CEO Steve Herbert and CTO Leo Reiter plainly state they plan to be the fastest solution.
Steve Hebert CEO Nimbix (image courtesy Nimbix)
Leo Reiter CTO Nimbix (image courtesy Nimbix)
Paul Arden, who runs the company that supports RealityServer, discusses the speeds and feeds when RealityServer runs on Nimbix hardware in this blog post. Nimbix is listed as one of several GPU cloud providers by NVIDIA partners as can be seen here.
When thinking about delivering interactive applications like Blender directly to end user desktops on Nimbix, this of something like Citrix but delivered from a supercomputer instead of IT server farm. In short, work in batch, interactive, or realtime and let Nimbix manage the hardware and hardware investment.