Get ready for CAD in the cloud.
At least, that is what graphics chip and subsystem maker NVIDIA Corp. says will be possible thanks to its newly developed VGX platform. With this scheme, a single server carrying one VGX graphics processor board can handle up to 100 users doing heavy duty graphics. NVIDIA says servers equipped with VGX will make it possible to access a cloud server from any device – thin client, laptop, tablet or smartphone – regardless of its operating system.


Even computation-intensive applications such as first-person shooter video games can be virtualized this way, NVIDIA says, with no discernable degradation in performance. And 3D solid models and simulations that characterize sophisticated engineering design work will work equally well when run from a VGX-equipped server.


There have already been attempts to make some kinds of engineering software cloud-based. But response time has been an issue. Users seeing a CAD model called up from a server, for example, might notice a perceptible lag between moving a cursor on a model and seeing the software finally respond. This lag can be just an annoyance or it can bad enough to make real-time server interactions impractical.
NVIDIA says it has eliminated such effects by removing about 100 msec from the chain of events that transpire between generating an image on a server and producing it on a remote PC or tablet. The better performance comes thanks to several fundamental patents by NVDIA researchers that pertain to memory management technology.


The company also wrote a special hypervisor program for virtualized graphics. A hypervisor lets multiple operating systems run concurrently on a host computer. It is so named because it is conceptually one level higher than a supervisory program. It lets multiple instances of a variety of operating systems share virtualized hardware resources. The problem, though, is that hypervisors to date haven’t been able to render graphics particularly well.


To solve this difficulty, NVIDIA devised a hypervisor optimized to work with graphic processing units (GPUs). Memory management techniques then effectively create a mini-dedicated GPU on the server for each virtual user. During interactions between the server and the remote user, graphic data streams directly out of the GPU frame buffer to the appropriate network interface card (NIC) without first having to go to the main CPU.


“We can push pixels or frames directly into system memory,” says Jeff Brown, general manager of the Professional Solutions Group at NVIDIA. “The remoting protocol can grab the data at that point. I can’t imagine a scenario in which there is no available system memory (to slow things down) given the trivial size of the frame buffer.”


NVIDIA also says delivering virtualized desktops this way can minimize the security risks inherent in sharing critical data and intellectual property. For example, source data for CAD models need never leave a secure server, but can be manipulated by users across the globe in real time using the VGX scheme.


Finally, NVIDIA says integrating the VGX platform into the corporate network also lets enterprise IT departments handle “BYOD” computing, that is, employees bringing their own computing device to work. It delivers a remote desktop to these devices, giving users the same access they have on their desktop terminal.


There are three parts to the NVIDIA VGX technology:
VGX Boards – These carry NVIDIA Kepler GPUs. The first NVIDIA VGX board is configured with four GPUs and 16 GB of memory and fits into the industry-standard PCI Express interface in servers. Each Kepler GPU has 192 NVIDIA CUDA architecture cores and 4 GB of frame buffer.
VGX GPU Hypervisor – This software layer integrates into commercial hypervisors, such as the Citrix XenServer, letting multiple users share common hardware and ensures virtual machines running on a single server have protected access to critical resources.
NVIDIA User Selectable Machines – This option lets companies configure the graphics they deliver to individual users in the network, based on their demands. Capabilities range from PC experiences to professional 3D design and engineering experiences.