SAN FRANCISCO—Back in 2016, when Google first approached id Software about bringing some games to a potential new streaming service, the game developer was skeptical to say the least. "The proposal immediately bumped against our main bias," id Senior Programmer Dustin Land said during a talk at this week's Game Developers Conference. "Streaming adds latency to the thing we desperately want to remove latency from."
Fast forward more than two years, and id was proudly on stage this week showing a version of Doom Eternal running on Google's newly announced Stadia streaming platform. But getting from that initial skepticism to that grand unveiling wasn't always an easy process, Land said.
Getting to yes
For years, Land said, Google had been watching their YouTube analytics, waiting for a big enough group of users to reach the point where their connections would be able to handle game streaming. By September of 2016, Google thought the broadband market was mature enough to give it a try, and the company approached id for some real-world help with game testing.
Porting the 2016 Doom to that very early version of Stadia took just three weeks of full-time work by two people, Land said. That process was made easier because the game already ran on Vulkan graphics with Linux support, things that Land said made it a "good time [for developers] to adopt" after two years of improvements from Khronos.
But that first streaming version of Doom was "lackluster," he said, with apparent lag even when running on a local network. So after a bit of iteration, Google came back to id that November with a new version that ran on a nearby cloud instance using a wireless router, a Chromebook, and an Android phone for input.
What about the speed of light?
While users can't be sure where the Google data center serving their Stadia experience will be, Land said that server will always be within a 64 km radius. With data traveling at around 200,000 km per second along fiber cable, that means a direct signal could be sent in under 320 microseconds (or just 0.32 ms).
"We know those [latency] speeds aren't reality for the open Internet," Land continued, citing myriad issues related to switching, congestion, packet loss, video encode/decode, last-mile delivery, and consumer AV hardware. But these are "all things that can be addressed and are continuously being addressed by the companies involved in building and maintaining the Internet," he added. In any case, the speed of light does not seem to be at the core of the latency problem.
"Needless to say, it was a night and day difference," Land said. "We were stunned by how much things had improved… It [only felt] like someone just forgot to enable game mode on the TV." That was enough to convince id to move forward with the project, even though there were still some kinks to work out.
More recently, Land said Google came to the id offices to set up a "Pepsi Challenge"-style blind test between Stadia and local hardware "to keep themselves honest and really drill down on eliminating perceivable differences in the play experience. They also wanted to demonstrate that Stadia could be superior to a local experience in certain eyes."
In that blind test, you could "hardly tell what was local and what was remote," according to Land. The results were strong enough to get non-gamers in the company excited, he added, and convincing enough to get Google's executives to greenlight the whole project for production, leading to this week's announcement.
Building a platform
Aside from the streaming quality improvement, Google's first entry into the gaming platform world has had some other growing pains, Land said. The company's first version of a Stadia development toolchain, for instance, was just "bad," with lots of bugs and a complicated sequence steps for setting up and migrating developer environments.
[contf] [contfnew] 
Ars Technica
[contfnewc] [contfnewc]
