AMD finally unveils an open-source answer to Nvidia's GameWorks

Welcome to Part II of our coverage of AMD's Sonoma event earlier this month. Previously, we took you through the company's plans for future HDR displays as well as how new technologies would be supported on current and future GPUs. Today, we're diving into the software side of the equation, including a new open source software initiative meant to answer Nvidia's GameWorks.

We first wrote about GameWorks nearly two years ago, and have written a number of follow-up pieces since. Nvidia's GameWorks program allows participating developers to use Nvidia's own middleware libraries for various in-game effects rather than relying on other third-parties or writing such code themselves. The advantage to GameWorks, in theory, is that Nvidia knows its driver code and hardware best, and (again, in theory) creates the best implementation of a given effect that you can achieve in the industry -- provided you own an Nvidia graphics card. The flip side to this is that if you don't own an Nvidia graphics card, you're a bit screwed. There's very little AMD can do to optimize performance for the specific libraries used within a GameWorks title, short of providing their own full-fledged library and hoping the developer is willing to integrate two separate libraries that do the same thing. (Spoiler: Most aren't).

Nvidia's response to AMD's complaints have typically boiled down to "If they want access to custom libraries, they can build their own." And now, AMD has.

One of the areas where companies sometimes try to fudge their facts is whether a product or project is actually open source. We've seen it with Mantle (which never went open source), with GameWorks (the fact that developers can pay for a code license under certain circumstances doesn't make a project open), and in plenty of other scenarios outside the GPU market.

AMD is licensing GPU Open and its libraries under the MIT open source license, which means yes, this is open source. It's not "open source if you squint," or "open code," or "code samples," or any other set of buzzwords. This is one of the most fundamental differences between AMD's new strategy with GPU Open and Nvidia's strategy with GameWorks -- AMD is explicitly inviting developers to contribute not just to code samples, but to the libraries themselves.

We expect to see GPU Open start rolling out in January with an initial set of libraries and capabilities. In addition to TressFX 3.0, AMD will launch new libraries centered on geometry, ambient occlusion, and shadows (we may have seen some of this work in Grand Theft Auto V). The program will also include multiple SDKs and tools, all of which will be collectively governed by the MIT license going forward.

According to AMD, GPU Open is a long-term initiative for the company, not a short-term effort to make a PR splash. AMD has championed a collaborative model of game and driver development for the last few years; with GPU Open, the company is putting its money where its mouth is. The fundamentally open nature of GPU Open will make it impossible for AMD to skew game performance towards its own hardware in the same way that Nvidia is accused of doing -- anyone can contribute code optimizations to GPU Open, which means there's nothing AMD could do to prevent Nvidia or a developer from writing its own optimizations into the code. The license explicitly allows for this type of modification.

One major question in future GPU Open versus GameWorks battle is whether or not AMD's initiative will make much headway against Nvidia's. To understand why this is the case, it's important to understand that GameWorks is often part of an agreement between Nvidia and the game's publisher. In such deals, it's common for membership in a particular program (Gaming Evolved, TWIMTBP) to also include co-branded marketing funds and certain sales guarantees. Nvidia might guarantee to a publisher that if it adopts GameWorks, Nvidia will purchase a certain number of game copies to be distributed along with qualifying GeForce cards. AMD's "Never Settled" program may also have used such considerations; companies don't typically get into the nitty-gritty of these arrangements for obvious reasons.