High-performance enterprise VR applications require hardware optimization
In order for enterprise VR applications to operate at peak performance, they require specially-designed compute platforms. Professional-grade HMDs (such as the VIVE Pro, XTAL, and StarVR) must display millions of pixels at high frame rates. Professional applications involve large and complex 3D models consisting of tens of millions of polygons, detailed textures, lighting effects, etc. In order to render at high quality for VR, the hardware platform must generate dual-stereoscopic images at a sustained 90 frames per second; dropped frames are disruptive to the viewer and can result in significant discomfort (motion sickness). Standard off-the-shelf graphic workstations are built with more general requirements and therefore do not supply the best performance for VR.
Silverdraft has designed a series of supercomputers to optimize performance for professional VR applications.
Optimized for Software
Different software packages for VR have different methods for rendering. Silverdraft is intimately familiar with how each application and each class of dataset places unique demands on the hardware. Some applications can only take advantage of a single high-clock CPU, while other applications can scale with many GPUs. Some applications benefit from large amounts of VRAM, while other applications benefit from higher core counts. Silverdraft designs systems specifically optimized for each customer scenario to deliver the highest level of performance at the lowest cost.
High-performance SLI bridge for dual NVIDIA Quadro GPU systems
GPUs offer a high-bandwidth interconnect, which lets them operate in tandem, supporting high-performance stereoscopic rendering in VR headsets. Many mass-market manufacturers have been known to install cheaper SLI bridge hardware that does not allow the GPUs to communicate at full speed. It is important to have a high-quality, high-bandwidth SLI (or NVLink bridge) connecting GPUs, which has been properly validated. Validation includes high-load processing, complex VR scenes with professional-grade HMDs such as Vive Pro, XTAL, or StarVR.
Cooling to maintain high performance
GPUs generate a lot of heat when they are processing at high load. As the temperature climbs, the GPUs will attempt to reduce their clock speeds and restrict performance in order to avoid overheating. Silverdraft’s bespoke systems include a robust, proprietary liquid-cooling mechanism specially mounted on the GPUs. This mechanism has been validated over many years for successfully maintaining peak performance even at high processing loads.
CPU-to-GPU Communication
Single-CPU systems are more efficient than dual-CPU systems for most VR applications. However, a CPU needs to exchange data between GPUs as fast as possible to maintain high frame rates and consistent motion tracking. A high-performance platform architecture generally needs a high-clock speed CPU and a high-bandwidth bus for fast communication to a given GPU. Many ‘off-the-shelf’ workstations are simply not built with that consideration in mind.
VRAM requirements
3D models having minimal textures or minimal geometry complexity may not need large quantities of VRAM on the GPU; performance on lower-cost GPUs may be sufficient as long as they are operating at high-clock speeds. Silverdraft helps their customers identify and achieve the most optimal GPU configuration, providing the best performance at the lowest cost.