Cloud VR Developments

At the Ultra-Broadband Forum (UBBF) in Hangzhou, China in October, Huawei announced the development of a new lab focused on developing the Cloud VR ecosystem. Much of the push behind Cloud VR has been driving the 5G radio network, as an application capable of consuming significant bandwidth and requiring low latency network – in many cases, Cloud VR has referred equally to implementations of AR and VR – with AR generally having a higher mobility requirement and therefore requires a stronger push to mobility and the benefits of 5G.

The VR OpenLab will be a new facility staffed with over 30 full-time people; supporting companies include Huawei, Cyber Cloud, iQiyi, 3Glasses, Hypereal, LetinVR, Nine!, Archiact, 3D, beamVR, High Fidelity, JD.com, Realis, Noitom, Rockchip, BOE, Virzoom, uSens, Pimax, SDMC, and other education and commercial institutions. This will likely act as a combination between an innovation lab – with some multi-party funded projects to be solved through cooperation of the various entities, with each retaining their own intellectual property, and an incubator – where the co-working of multiple party’s speeds innovation through shared expertise and access to funding and talent.

Interestingly, a starting point proof of concept has already been developed. Huawei, Cyber Cloud, and Leap Motion already developed and showed a prototype cloud VR system which leverages a cloud-based game engine (developed by Cyber Cloud) – connecting to an Oculus Rift headset which plugs via an HDMI and USB adapter directly into a broadband modem, sending motion vectors and receiving graphics from the cloud. The system doesn’t use much video compression, sending a remarkable 300 Mbps video stream down to the device, but is functional and low enough latency to play many games. With the pervasive Gbps FTTx based connectivity in Asia, this type of system is viable.

One of the strongest arguments behind cloud VR, and one of the reasons that it is being originated in Asia, is the cost factor.  Today, premium “gaming” PCs cost US$1000-US$1500 while the HMD headset and peripherals are on order of US$500.  Before game content costs, the setup costs roughly US$2000. While individual gamers will vary significantly, let’s assume that this PC will last the gamer three years – the gamer purchases about eight games per year and achieves about 40 hours of gameplay per game. All in, the cost is roughly US$2 per gaming hour. Today, in Amazon Web Services (AWS) within the U.S., the cost of AWS elastic GPU cloud ranges from US$0.05 per hour (medium instance with 1 GB) up to US$0.40 per hour (2 XL instance with 8 GB). All in all, it’s quite possible to see the cost benefit of cloud GPUs, especially if you can fit the workloads within large or extra-large instances at about US$0.10-US$0.20 per hour of compute.

Seeing this early progress and significant investment in lab space around cloud VR, what is required and what is the opportunity?

What is required?

• Like other tethered platforms, wireless tethers especially make sense in cloud VR for a wire-free experience, much like the freedom tablets provide. Ideally, this would be a new 60 GHz WiGig based product, with the broadband modem – or a USB or Ethernet tethered base station – acting as the tether point.
• Lighter weight headsets and lower cost headsets would be beneficial. Products like Oculus Go, at US$199, are a good starting spec for a cloud VR headset. Of course, the goal of Cloud VR is to give more of a 6 degree of freedom (6DOF) experience, thereby requiring inside-out tracking. This will push the price up a little bit, as well as 60 GHz tethering technology, above.  HMD likely require some level of local processing – at least for object detection (safety), simultaneous location and mapping (SLAM) if they are to have any A/R capabilities and/or other more advanced motion. However, Oculus Go has a relatively robust Qualcomm chipset in it with full application capabilities. 
• Application ecosystems must develop around the cloud. Fundamentally, it is quite possible to port the input and output drivers of existing VR systems to render to a remote headset, if latency considerations are met. However, there may be other application development and user experience considerations in the future. Application and platform considerations are large in VR – it would be most straightforward to port a single content platform to the cloud and then drive all applications onto that platform.

What is the opportunity, including consumer and industrial applications? 

• From a consumer perspective, the cost benefit is the primary driver around cloud VR. All the VR opportunities – including video, gaming, education, immersive content, virtual tourism, and social can fit within the cloud VR domain. With the demonstrated latencies in the 8-10 ms range, it is possible to run most or even all computation in the cloud; this is possible at less than 20 ms, depending on frame rate.
• From an industrial perspective, cloud VR may enable easier deployment of services to remote locations. In addition, lower per-unit costs could benefit reach-oriented services – such as virtual bank tellers or remote medicine – in which the cost of deployment may be a significant factor – and the barrier to deploying systems that may or may not be used repeatedly may present significant barriers to system deployment.  Also, like the greater benefits of the cloud, lower start-up costs with more variable costs enables more flexible business models and scaling around service trajectories.

All in all, the computing ecosystem has oscillated between a client-based and server-based approach many times over the years (mainframes versus PCs). Today, a mix of local and enhanced cloud capabilities exists in most domains – save for “embedded” domains in which performance is critical or cloud connectivity cannot be relied upon. Most VR applications will rely to a certain extent on the cloud – from game updates to multiplayer gaming or ad-servers on downloaded gaming content. The question is more a question of howmuch will VR applications rely on local processing and storage, compared to how much they will rely on the cloud.One example in AR is that Microsoft’s Hololens can natively process about 250 K polygons of a CAD image on the device, while industrial designers routinely work on models with millions of polygons. The design choice faced is whether to decimate the model, with some downgrade in visual acuity and a slower model visualization review cycle, or rendering in the cloud. Both make sense for different applications. Ultimately, any technologies developed in support of cloud VR will increase the range of possibility for the entire VR ecosystem.