![[H]ardOCP](/images/blank.gif){kind=small}
- Date:
- Monday , March 20, 2006
- Author:
- Brent Justice
- Editor:
- Kyle Bennett
- Google +1
NVIDIA SLI Physics Tech Preview
Today NVIDIA is announcing some very exciting technology for gamers that is poised to improve your gaming experience in a very real way. Our preview of the technology will tell you what you need to know from the gamer’s perspective.
Havok FX
You have probably heard of the Havok physics engine. This physics engine is used in many games. Here are just a few that immediately come to mind: F.E.A.R., Half Life 2, Age of Empires III, Painkiller, Max Payne 2, and Splinter Cell 3. Here is a more complete list. More interestingly, here is a list of upcoming titles. Some of the more notable games include Alan Wake, Splinter Cell 4, The Elder Scrolls IV: Oblivion (out today in fact), Tom Clancy’s Rainbow Six: Lockdown, Medal of Honor: Airborne and Test Drive Unlimited. The Havok physics engine is very popular; there is no question about that.
The latest incarnation, Havok FX, includes some very enticing features; namely using Shader Model 3.0 GPUs for physics calculations. Notice we didn’t say “NVIDIA GPUs only.” Technically Havok FX will work with any Shader Model 3.0-capable GPU to help game developers utilize those GPUs to help offload physics. ATI has also voiced that they have their own physics bag o’ tricks in relation to their GPU and has a deeper line of products when it comes to supporting SM3.0 than NVIDIA. Keep in mind though, that it is up to the game content developer to develop this support into their games. From what we have been exposed to, Havok FX looks like one lean mean physics machine and should provide developers with a lot of options for using GPUs as physics processors no matter be it Red Team or Green Team’s hardware.
NVIDIA SLI Physics
What NVIDIA is announcing today is an engineering partnership with Havok to develop physics simulation that runs on NVIDIA GPUs and is amplified with SLI. In fact if you are lucky enough to be at GDC today you will find NVIDIA has a live real-time demonstration at their booth of what Havok and NVIDIA SLI can do.
Here is the gist of it; with two GPUs in SLI one GPU can render 3D while the other is dedicated to doing all the physics calculations. This will work with all GeForce 7 series GPUs. NVIDIA recommends a minimum of GeForce 7600 series for this. This is great because no “new” hardware is needed as NVIDIA and Havok’s implementation works with current hardware in the marketplace. This means the install base for the hardware is already there!
It goes further than that even. This physics implementation can also work with a single GPU. When a GPU is rendering a game it may not be using all of its cycles sometimes. The ability is there to use some cycles for physics and the rest for 3D all in the same single GPU. Now, there may in fact be some tradeoff here, as you might lose some 3D performance. The great thing is the game developer can design for that and perhaps program certain areas that need less 3D rendering to have beefed up physics in the same game. It will all come down to the game content developer and how they use the GPU.
All you need to know is that a single NVIDIA GeForce 7 series GPU can help with physics and 3D rendering. SLI would obviously do this better but single GPU implementation is being looked at as well.
How it Works
Since the hardware install base is already there all you need for this to work will be a driver update and the games to support it. Yes, the game must be using the Havok FX physics engine and the game content developer must program this support into their games. It will be completely up to the game content developer to determine how much physics work your GPU(s) will do and what the benefits will be. They could use it just to benefit performance or they could use it to really increase the gameplay experience.
When in SLI, one GPU can be used to do all the physics calculations. The data is then passed on to the DX9 portion of the driver for rendering out on the screen. No transfer back to the CPU is required so the CPU stays out of the loop meaning this is a true hardware dedicated process with the GPUs. In fact technically this could work with GeForce 6 series video cards but the target is for GeForce 7 series. NVIDIA recommends as the minimum a GeForce 7600 for this. Obviously faster video cards will show more benefit.
There can also be user controls built into the driver. For example with SLI you have the option to enable or disable one GPU or force both GPUs in SLI. With SLI Physics NVIDIA will have game profiles so you can force a disable of physics acceleration and use both GPUs for 3D rendering if you want, or if the game supports it one for Physics and one for 3D. Those details will be worked out later when we see this hit our desktops.
In fact that seems to be the catch; this right now is just a technology preview to show you what will be possible and what will be coming down the road. The first gaming tittles that actually benefit from this won’t be out till at least the end of this year. 2006 won’t be the year of physics but it is looking like 2007 may just be the start of something very cool for gamers.