The gaming industry has come a long way. In 2022 it played host to an estimated 3.2 billion players worldwide, generating a total revenue of $184.4 billion, according to Newzoo.
One of the most remarkable developments in recent years has been the accessibility and affordability of gaming. Players can now enjoy gameplay on almost any device connected to the Internet via subscription services in addition to traditional PC and console games.
Game publishers have made great strides in adopting the latest graphics and hardware technologies. However, a delay in moving to cloud gaming from console-based approaches could open the door for disruption from subscription video platforms like NETFLIX. Just as NETFLIX disrupted the home entertainment rental ecosystem with their always-available subscription streaming service, they could do the same with gaming.
Cloud gaming platforms operate in a highly competitive environment with narrow margins. In the United States, popular cloud gaming platforms like Amazon Luna start at $4.99 per month. This makes choosing the right GPU for game graphics rendering and video encoder essential for profitability and competitiveness. Cloud gaming platforms specifying video encoders should consider four key factors; CAPEX, OPEX, Quality, and not funding their competitors.
Lowest Cost Per Stream
For a cloud gaming platform, the cost per stream represents the initial investment required to set up the platform, including the cost of servers and encoders. With the cloud, the cost per stream impacts the profitability of a managed service like a cloud gaming platform to the point of making the entire business model viable.
ASICs are the secret to making a cloud gaming service viable. With an ASIC-based encoder like the NETINT Quadra T2 VPU (Video Processing Unit), coupled with a GPU from AMD, a single server can deliver as many as 200 simultaneous 720p60 gameplay sessions. This performance beats the previous high-water mark of 48 game play sessions using eight GPUs in a single server chassis.
Lowest Possible OPEX Per Stream
OPEX (Operating Expense) represents the ongoing costs of running the platform, including electricity, bandwidth, and maintenance. Energy (electricity) costs are a significant part of OPEX, and they are increasing in many regions. This makes power consumption an important and key consideration for choosing an encoder.
Compared to CPU-based encoding with software, the Quadra T2 VPU consumes 10 to 20-times less energy at only 40 watts per hour delivering the same throughput. Depending on the host server configuration, as many as ten VPUs may be installed making each server the functional equivalent of ten to twenty high-end server machines.
Rack space requirements should also be considered. With colocation prices ranging from $50 – $300 per month, the additional servers needed in a software only implementation would cost up to an extra $5,700 per month for 200 gamers (co-location costs only). While costs may be less if housed in your own facility, you still need racks, cooling, and maintenance for 20 servers compared to one.
A long-lingering misconception about ASICs is that the quality cannot match that produced by the software. Obviously, video quality depends upon configuration options and the operational state that the encoder is operated in. Internal tests show that the HEVC output quality of NETINT VPUs is quite competitive to software and other hardware transcoders, especially when run in their lowest latency state. See Table 1.
For example, as compared to x265, the Quadra VPU produced better output quality than NVENC, the popular encoder that is available on NVIDIA’s more recent GPUs and x265 up to the medium preset. x265 using the medium preset produces quality that is close to VOD. But it is an operational mode not commonly used because of the computing power needed.
Most live streaming engineers use the x265 veryfast or superfast presets. When compared to the x265 superfast preset, Quadra VPU produced the same quality and with an additional 25% bitrate reduction, which translates to significant savings.
Table 1. BD-Rate PSNR quality comparisons between Quadra, x265,
and the NVIDIA RTX 3090 encoder in low latency settings.
At the extreme right, you see that Quadra was able to match the quality of the NVIDIA RTX 3090 HEVC encoder at up to an 11.57% bitrate production. ASICs producing quality that rivals software encoding is not unusual. As discussed here, Google has achieved near-software quality with their ASIC-based ARGOS transcoder as well. This shows that clearly, you do not need to compromise on quality to achieve the density and efficiency benefits of ASIC-based transcoding.
Hidden Costs of GPU
Evaluating the cost of hardware is relatively straightforward if the primary factors are easily understood and defined. However, with GPUs, there are hidden costs that are not always recognized or acknowledged. For example, as tech platforms expand their offerings, Cloud gaming platforms could find that they are funding potential competitors.
As an illustration, the US Federal Trade Commission is attempting to block Microsoft’s acquisition of Activision, partly because the Azure cloud platform gives Microsoft a cost advantage over cloud gaming platforms without similar infrastructure.
Presumably, Amazon with AWS, has the same advantage. Similarly, this article describes the cost advantage that NVIDIA derives from other services that buy its GPUs for game rendering.
Another hidden cost can be found in the complexity of the procurement process for GPUs. Due to the supply chain issues triggered by COVID, and the incredible demand spike for GPUs, simply having the opportunity to buy the amount needed was far from certain. Still, your negotiation strength could have significant sway on the price or delivery schedule that you received. Put simply, for anyone needing to buy GPUs in the quantities needed by a cloud gaming platform, it cannot be assumed all that is needed is a P.O.
Finally, there’s a significant loss of negotiating leverage once a gaming platform chooses a GPU vendor, and this is particularly true when the GPU performs double duty in rendering frames and encoding them for streaming. Once a platform chooses a GPU vendor, their technical architecture is essentially locked with that selection, so they can’t switch to another GPU vendor without significant development time and cost. This puts the platform at a disadvantage when negotiating with the selected vendor as they have limited bargaining power.
Often, GPU vendors abuse this leverage by charging expensive license/API costs or refusing to make improvements for their customers. In other cases, this lack of bargaining power could lock platforms into using a GPU-based encoder that delivers uncompetitive quality as compared to third-party options. Some GPU vendors may even refuse to undertake enhancements that would enable the use of third-party transcoders, even if this would improve throughput and quality and reduce OPEX for the game platform.
By implementing a dedicated transcoding unit separate from the GPU, a cloud gaming platform can decouple its design into standalone GPU and VPU modules. This makes it simpler for vendors to switch to different vendors, providing significant leverage to negotiations with all vendors.
The Cloud Gaming Opportunity
According to Newzoo, cloud gaming is one of the fastest-growing gaming industry segments, with a CAGR of 50.9% from 2020 to 2023, accounting for 49% of the global gaming market. Cloud gaming is a benefit to players in all regions and it opens up new entertainment experiences for many people without access to expensive consoles or who cannot afford the newest games.
For others, access to high-quality gaming is a way to extend the entertainment experience outside of the home. Also, it offers a way for mobile gamers to access games that they may be unable to play on their mobile devices due to hardware limitations.