Was naming a cloud gaming platform after the evil energy corporation from Final Fantasy VII a comic masterstroke, playing on the fact that cloud initiatives are popularly perceived to be the work of the Devil, or deserving of a shower of rotten eggs? Either way, Shinra, Square Enix’s big push into the field of remote processing, is intriguing, if subject to some of the same limitations that have troubled services such as OnLive. Now in beta across Japan (a US beta will follow later this year), it isn’t just a new and device-agnostic way of distributing games but a more elegant, cost-effective approach to how they’re run.
I’m sitting in a presentation room high in Square Enix’s London offices overlooking a glittering sweep of riverside. To my right, Shinra’s senior vice-president of technology Tetsuji Iwasaki rattles through PowerPoint slides. Senior vice-president of business Jacob Navok throws in commentary and the odd bit of translation from a laptop screen—he’s broadcasting from Japan, where it’s well after midnight.
The foundation of Shinra is the idea that a network’s power is a question of smart integration rather than the strength of each individual CPU or GPU. Past a certain point, Iwasaki explains, processing performance doesn’t increase in proportion to cost, which means it’s more sensible to eke out gains by way of fast network cards and, in particular, RDMA/TCP dual protocol interconnection, whereby computers skip the operating system layer and dip directly into each other’s memory banks.
The machine was able to outperform the Earth Simulator, a 60 billion yen government project.
He cites a supercomputer created from off-the-shelf components by Professor Tsuyoshi Hamada of Nagasaki University in 2009—despite its homely origins, the machine was able to outperform the Earth Simulator, a 60 billion yen government project, because its components worked together more efficiently. Shinra’s own server farms are built according to the same principles.
The other piece of the puzzle is, of course, software. Shinra games are broken up into modules—AI, graphics rendering, physics and so forth—that can be run on different machines, rather than having the entire thing sit inside a single computer. This is less wasteful, because it means Shinra can ‘freely’ re-allocate the network’s resources on a module-by-module basis—the physics calculations of a shooter such as Battlefield 4 might get a bigger helping of CPU power, for example. Having all the calculations for several copies of the same game occur in the same place also means that certain results can be duplicated. According to Iwasaki, you could calculate the animation and rendering for a character once per 100 users, rather than once per user.
Greater efficiency aside, a benefit of Shinra is that because all of the key processes happen locally, synchronisation between players is less of a bother. You aren’t, after all, shuttling data about the position of objects from one end of the globe to the other—all these relationships are worked out on Shinra’s servers and a video stream is sent to client devices. This theoretically means that nitty-gritty real-time mechanics and effects are practical in massively multiplayer environments, because there’s comparatively little latency to deal with.
Iwasaki illustrates all this by way of a multiplayer world simulation, created in roughly six months by a team of three, in which a thousand AI-controlled puffball creatures spread out across a forested valley that’s supposedly 17 times the size of Skyrim. Where a game like Skyrim vanishes NPCs when they stray out of view, to save on memory, here everything is rendered and animated no matter how far the animals disperse. That’s possible, Iwasaki claims, because the simulation is able to ramp up its share of the server farm’s RAM to a whopping 100GB. “For me if you can make a world that massive and populate it with all of those fully animated strange little creatures in such a short time, that by itself is a new experience,” he adds.
Shinra features many of the same potential drawbacks as previous cloud services.
Another example of Shinra’s potential is the platform-exclusive Space Sweeper, billed as a “twitch game on a massive scale”. It resembles a top-down shoot-’em-up in screenshots—budding Sweepers must weave a path through thousands of dynamic, fully rendered projectiles—but the difference is that it unfolds on massive open worlds with hundreds of players at once.
It all sounds and looks mightily impressive, but then again we’ve nodded our way through similarly wide‑eyed takes on the notion of server-based processing in the past and it’s important to note that Shinra features many of the same potential drawbacks as previous cloud services.
Latency between server and client is still an obstacle, for one thing, which is why Shinra won’t support all connections to begin with. Square Enix’s big gamble is that internet infrastructure will have advanced sufficiently by the time the service becomes widely available. “Even certain cable users aren’t going to have a great connection, so we’ll start with fibre users and then we’ll roll out from there as far as we can, at a level of quality we’re comfortable with,” concedes Navok.
Shinra’s sheer speed does, however, allow it to compensate a little for the infrastructure’s failings. “When we take what we’re rendering, plus encode, plus decode and internet latency, we end up with a total loop that’s actually half that of a console,” says Navok. “Just by virtue of doing the render loop cut, we give about an additional 40 milliseconds back to the net, so the user experience on a fibre connection feels local.”
The other challenge is simply persuading developers to invest. Shinra already supports a number of highly-fed internal projects, including the monitor‑melting tech demo Agni’s Philosophy, and has enlisted the support of both unspecified larger publishers and smaller outfits such as Camouflaj, developer of the République series. But it seems badly in need of a stunning tentpole exclusive to hammer the possibilities home.
As you’d expect, Navok insists that it’s only a matter of time. “I think there are some developers who were waiting for this,” he concludes. “People who always thought that it was possible, but were waiting for somebody to be crazy enough to do it.”
