The Entire Game Should Move Onto the GPU, says Rev Lebaredian
As the computing functionality and horsepower of GPUs has grown over the last few years, the role of the GPU is rapidly expanding to game tasks beyond rasterization-based graphics. Now GPU computing can be used for solving a variety of problems in game computing including game physics, artificial intelligence, animation, post-processing effects and others. Rev Lebaredian, Director of Engineering in Nvidia, says that as GPUs grow and become more general purpose, more and more of the game is going to be shifting over to the GPU. The final goal is to have the whole game run on the GPU.
Rev Lebaredian leads a team of engineers developing APEX - a middleware library and tools for integrating realistic and scalable physics simulation into games. APEX is a new technology that empowers artists to quickly create fully interactive in-game clothing, destruction, particles and vegetation. APEX is a middleware library and artist-oriented tools, built on top of Nvidia's PhysX software. The APEX framework streamlines the integration of new effects into games.
Saltmarch Media spoke to Rev Lebaredian during his recent visit to Bangalore, India. The discussion began with a topic close to Rev's heart – APEX. "APEX is trying to handover the power of game physics to the artists directly as they are the biggest section of employees in a game development company. By achieving this, an artist need not depend on a programmer to achieve every task and this in turn will enable the creation of a lot of creative content which the user of today has come to expect from games," says Rev.
Read on to know Rev's thoughts on moving the entire game to the GPU starting with graphics and physics, how video applications are tapping the increasing computing power of GPUs, optimal Direct X 11 support, and advancements in the graphics and digital media processors industry. Rev also shares his experiences with using Optimus technology that addresses the performance/battery life balance by automatically selecting the right graphics processor-between an Nvidia discrete GPU or an Intel integrated GPU
Q. Please introduce yourself to our readership and talk in a bit about your role at Nvidia.
Rev Lebaredian: I am the Director of Engineering within the Content and Technology group within Nvidia. This is the group that deals primarily with application developers. My responsibility right now is leading the group that develops a product called APEX.
APEX is a middleware solution we have been creating and implementing in actual games that enables game developers to put hi-fidelity simulation and dynamics into their games with minimal amount of work for the quality that they get out of it.
The goal is to make APEX scalable across platforms and allow game developers to put as much of it as possible in their games by leveraging their artists, instead of just a few programmers in their group who know game physics.
Q. One of your talks at India Game Developer Summit specifically revolved around APEX, a middleware library and tools for integrating realistic and scalable physics simulation into games. Can you explain APEX in a bit?
RL: What we found while integrating physics with the GPU was that for many years, the bottleneck in most game productions was the fact that most of the actual employees within any game studio are artists and the ratio of artist to other employees remains big because games are getting bigger themselves.
So in order for us to get more content in games we need to address the issue at the artist level and allow and enable them to create physical content without having to depend on a programmer to do everything.
Currently within development companies there are around one or two artists who are experts at doing physics programming and that is not sufficient for us to really enable games with the kind of content everybody wants.
Q. APEX is built on top of Nvidia's PhysX software. PhysX was designed by a company called AGEIA, which was acquired by Nvidia in 2008. What do you see as the benefits of the acquisition?
RL: AEGIA was a great acquisition for us. We knew that as GPUs grow and become more general purpose, more and more of the game is going to be shifting over to the GPU. The final goal is to have the whole game run on the GPU.
The next natural step after graphics is physics. It is very parallelizable and it is scaleable. There is lots of computation so it is in many ways a good fit for GPUs. So we evaluated and found out that it would take a lot of time to build up a physics library just at the rudimentary level and that it was much better to do an acquisition.
We were lucky as AEGIA were willing to do this and they have also had a lot of experience with accelerating physics on the hardware. So in many ways it was a perfect match for us. What they were trying to do with their products matched our business model and hence the expertise that they had gained matched us perfectly. This also has ramifications in the future.
What we learn from moving physics over to the GPU will apply to everything else we try to move over to the GPU as well. Physics is a great real-time application to try to move to the GPU so it is critical for us to have technologies like this in-house that we can use as beacons for where our architectures need to go in the future on the whole.
Q. More and more processing is now being done on the GPU. And video is the killer application for the GPU. What are the new kinds of video applications that are tapping the increasing computing power of GPUs?
RL: Encoding and transcoding are already big on the GPU and we are seeing many compute based applications that take advantage of the raw horse power of the GPU. As and when people create more and more digital media, there is an insatiable appetite for computing power in order to process this and archive it and store it in ways that are convenient for people. So we definitely have a big need for this sort of computation abilities.
In addition there are lots of problems with video that are naturally suited towards massively parallel architectures. Motion estimation, doing optimal flow, and these sort of image processing problems map very nicely onto the GPU.
Q. Nvidia had earlier pioneered a technology called switchable graphics to address the performance/battery life balance. But it was cumbersome to use since users had to manually switch between the two display adapters. Recently Nvidia released the Optimus technology that automatically selects the right graphics processor-between an Nvidia discrete GPU or an Intel integrated GPU. Have you used the Optimus technology and what is your feedback about it?
RL: I think its fantastic and it is much better than the switchable stuff that we had before. Optimus is extremely elegant and I think what you will find in the reviews and in the press will confirm this. Everybody loves this and I think it is the ideal solution to this particular problem and it is the sort of thing that we are actually good at.
It has taken a lot more than just developing hardware and the driver in order to produce something like this. We have to go through all the applications people are running, create a profile and invest a lot in QA and there are things we have to do in an ecosystem to create a product like this. So it makes me proud when we actually deliver something as elegant as this as a company.
Q. Will optimal Direct X 11 support be a deciding factor in terms of GPU sales in the next few years?
RL: History has always shown that to be true with respect to GPUs. We are not done with graphics. There is obviously a long way to go before we get to photo realism. And with games we are not even close to where film has been for many years now. DirectX 11 is a large step towards that.
We have had a lot of texture complexity for a long time and we have hit the limits of it and it looks kind of weird when the geometric complexities of games do not match the shading and texture complexity that we have become accustomed to. So as soon as we have some games out there that take advantage of DirectX 11 features, it is going to become obvious that all games need to increase their geometric quality to match this.
Q. What are the advancements you see in the graphics and digital media processors industry in the coming years?
RL: We are at an inflexion point in this industry. The whole parallelization thing was something the world was forced to accept when CPUs started going to dual core. We are not able to depend on the same speed ups that we were getting year after year with traditional CPU architectures. So GPUs happen to be at the right place at the right time.
With the level of programmability we have now, we are slowly converging towards the limits of what CPUs will eventually become. So now it is a race to see who can create the massively parallelizable architecture for the future. My hope is to see applications as complex as games completely running on the GPU.
Q. What are Nvidia’s ambitions for the next decade?
RL: In the coming years we have a lot of work to do in integrating physics properly onto the GPU. Then we need to ensure that all of it is running in there and it is optimal as possible and coexisting with all other parts of the game that we want to run on the GPU as well – including artifical intelligence (AI) and various sorts of scripting and all the various processes that happen within a modern gaming engine.