In the pursuit of FPS: how AMD and Nvidia are tackling PC gaming optimization
30th Dec 2013 | 17:00
Cold, hard math vs adding a human touch
Over the last few years PC gaming has become significantly more approachable to the everyday gamer, thanks to a combination of more affordable hardware and the availability of discounted digital game libraries through places like Steam and Green Man Gaming.
While the price has gone down for both PC hardware and titles, more recently Nvidia and AMD have tried to alleviate even more gamer headaches with optimization software. To this end, Nvidia opened its GeForce Experience beta to the public in January, and AMD launched its Raptr-powered Gaming Evolved application in September.
On the surface, the two applications almost seem cut from the same cloth. Both simplify the process of downloading new graphics card drivers to a one-click software update. Similarly, the software packages also set all your graphics settings following some basic guidelines that prioritize a faster frame rate or greater visual fidelity.
Continuing the similarities, the pair of apps were developed with the explicit purpose to help fine tune visual settings for less tech savvy gamers. After all, not everyone knows the difference between tessellation (the splitting of polygons) and ambient occlusion (the way light radiates and reflects) to save their overheating GPU as the FPS craters.
While the two apps are much the same in nature and purpose, that's where their commonalities end. We recently spoke with Nvidia and Raptr, the online gaming partner that powers AMD's Gaming Evolved app, to uncover how each are figuring out PC optimization and what they might have in store for the gaming masses in the future.
The numbers game
Raptr is a 7-year old game tracking service that helps users keep tabs on how long they've been playing games. Beyond clocking hours it's also an online community site that ties together users' Xbox, PlayStation Network, and PC gaming accounts for those achievement/trophy obsessed.
Raptr CEO Dennis Fong explained that although the optimization piece is new to his company's offerings, it's not actually that much further than what Raptr was already doing. The company itself tracks over 2,000 computer games with more than 20 million users. It also uses technology that can tell whether gamers were playing a single or multiplayer title, what games they've installed and what hardware makes up their rigs.
In a micro-sized version of the Raptr service, AMD's Gaming Evolved pulls crowdsourced data from a user base edging over one million.
"The way we designed the system is we actually know what types of [PC hardware] configurations people are using," Fong explained. "We know what frame rates they are actually getting and we capture all that data every single time you play."
He continued: "Every single time you play, it records a FPS histogram of your game session. We can tell if you're playing a single player or multiplayer session. When we record these FPS histograms, there's a lot of noise; as an example if you're watching a cut scene that frame rate is capped, so we sort that out."
These cleaned-up histograms are then churned into Raptr's machine learning system, which crunches all the data. The system is designed to zone in on PC gaming experts who get the optimal performance while using high-quality visual settings.
Using an example of expert users who turned off tessellation Fong explained, "what we're seeing a big boost in performance while the rest of the high quality settings are extremely high. Once our machine learns that, it then rolls that [recommendation] out to everyone else. "
"What we're trying to do is find experts," he went on. "The gamers out there that go to all the tech sites and tweak their settings. The beauty of our system is we capture all of that knowledge automatically."
The human touch
In a similar approach, Nvidia also leverages the expertise of hardcore PC gamers, except it uses real-life people in its testing labs around the world.
James Wang, GeForce Experience product manager, was happy to tell us that Nvidia does its optimization work in-house with labs in Moscow, Santa Clara, Calif. and Shenzhen, China.
Each lab employs a team of expert testers to figure out which settings are most important and contribute the greatest to the gameplay experience, all the while leaving out those that simply eat up performance.
"We figured it out pretty early that you can't just run scripts and not have anyone look at it because in the end, when you say to people [that] we recommend these settings are the best, that really requires a subjective call," Wang expounded. "For example, if anti-aliasing or texture quality is more important in a game, a computer can't tell you which one is more preferable."
To do this, Nvidia uses an optimization that tests different combinations of CPU and GPU hardware. While Nvidia doesn't test every single CPU release, Wang was quick to note that every GPU released by Nvidia is covered. Plus, the team constructs a performance index of all Nvidia's GPUs on the market today.
This drawn-up list of recommendations is fed into Nvidia's own algorithm and loaded onto a server of different PC part setups. From there the lab team takes the priority list and figures out how to turn on as many graphics settings as possible.
No perfect system
Both Nvidia and AMD have developed similar systems using two different approaches, but each has its own set of flaws. Wang specifically criticized AMD and Raptr's crowdsourced approach as a privacy risk as well as being ineffective.
"If we just did pure data mining from the users, it's a lot of data being mined and not everyone is comfortable with that," Wang broached. "The other thing is that when you mine the common case, the average answer is not necessarily the correct answer."
These privacy issues were misconceptions that Fong addressed by explicitly saying, "all of this data is used exclusively for optimization. We don't sell any user data [belonging to] individuals or otherwise, it's just to make the optimizations better."
He elaborated that "when a lot of people think about crowd sourcing, then they think whatever is popular, but that does not really work in this particular case because popular is the lowest common denominator of default settings."
Nvidia, meanwhile, has a system that arguably cannot account for every possible PC configuration out there. Wang contended that "the dominant performance impacting parts are the GPU, CPU and of course the resolution of the monitor."
"The more important number [to gamers] - is their PC covered?" he posed. "We cover pretty much all the CPUs today and all of our GPUs, both desktop and notebook."
In the future, Wang and Fong each told us to expect more games to make it under their respective optimization trees. Currently Gaming Evolved serves 65 games with roughly seven being added every week, while GeForce Experience supports 130 games since starting with a meager 30-plus titles.
Features on the horizon
Beyond optimization Nvidia and AMD have implemented other features for the streaming and YouTube gaming communities. The Nvidia GeForce Experience app brought a built-in game capture tool called ShadowPlay, which allows gamers to record and upload their own gameplay using only their graphics card.
Gaming Evolved, on the other hand, implements an easy way for users to start streaming their games on Twitch. Nvidia also recently added Twitch integration.
Both Wang and Fong agreed that the emergence of YouTube and eSports have taken off as a huge part of the gaming community, but just like optimization, Nvidia and AMD have adopted different approaches to game capture and streaming features.
Whereas Nvidia does more work behind the scenes, AMD utilizes an overlay that goes on top of full-screen games to give gamers access to their web browser without having to Alt-Tab.
Looking towards the future, Wang teased that Nvidia's ability to stream games through its graphics cards beyond using just the Nvidia Shield.
"We haven't talked about that right now, but the core technology of streaming does not require the end point has to be a Shield," Wang said. "It is possible that you can stream to another device but we haven't announced any direction to go there."
For AMD, Fong said that Raptr wants to implement more tools into its screen overlay, but we'll hear more about this as we head into January along with more announcements at CES 2014.