[Software] Nvidia CUDA: PhysX and General Computing

[WilkerCSBD]

It is known that since the implementation of the unified shading units in DirectX 10, the graphics chips achieved added benefits that can be applied to other types of operations that we call general computing and that is nothing more than to say that we can use them to execute other types of applications not oriented to the generation of graphics on the screen. As it happens, these multi-unit programmable designs are spectacularly effective in all applications that can take advantage of their great capacity for floating-point mathematical computation.

These types of applications tend to be very oriented to the multimedia world, both professional and domestic, as well as to research where a single graphic offers a potential that before were only found in expensive workstations that were more similar to a supercomputer than to a computer. conventional.

General computing is not exclusive to any manufacturer, all of them, when meeting certain Microsoft API requirements, had to adopt this type of architecture and in fact there are currently open platforms for general computing application programming, being Windows 7, with DirectX 11 and DirectCompute. 11, one of the ideal platforms for this type of application. Of course there are also other open APIs, very widespread and with a promising future, such as OpenCL, which is used on any platform, be it Windows, Mac or Linux, and which possibly ends up becoming the OpenGL of general computing applied on chips. graphics.

There are also proprietary platforms, with free SDKs, such as CUDA that belongs to Nvidia and that, except for a miracle, is only supported by the graphics chips of this manufacturer. It is a proprietary solution that hangs more or less established standards in the sector such as PhysX, which is applied to the physics calculation applied to video games. CUDA is de facto the most widespread API today, both in professional, domestic and recreational applications, but it is not the only platform supported by Nvidia since they also have a presence in the other most important APIs already mentioned.

What does CUDA offer us?
What does CUDA offer us?

CUDA, which is present in all graphics from the Geforce 8000 generation onwards, offers us the possibility of using the graphics as a substitute or support for the central CPU in specific and specially designed applications. Not by having an Nvidia graphic with the appropriate drivers will we be able to speed up any application. This must be specifically designed for it by means of one of the standards that supports the graph in question.

Currently, CUDA offers the ordinary user the possibility of accelerating multimedia processes, both creation and reproduction, as well as game modes improved with standards such as PhysX. There is not much else to scratch and there are no added applications that can benefit significantly from this added capability. Currently the processes that require the most hardware from a computer are basically playing games, playing high definition content and creating this type of content.

ION playing 1080 video with less than 20% CPU consumption.

It is true that there are other needs for other users but I think you will agree with me that currently any low-end computer is capable of performing the most common functions, such as office automation functions, without the need for large processing capacities. It is when we play or perform multimedia tasks when we begin to ask for more power from our machine. That is why current developments are mainly focused on three pillars: scientific professional computing, high-definition multimedia reproduction and creation, and video games with high technical requirements.

I do not think that I am wrong if I say that the first of the pillars is for now one of the great beneficiaries and also one of the great markets to be exploited for this type of solutions since the calculation potential of a modern chip, such as the RV870 of ATI or the Nvidia GT300, far exceed the computing power of the best x86 CPUs on the market, and therefore offer at reasonable prices superior computing capacity with easy scaling, since we can mount several graphics chips in the same PC, and a programming landscape with various options, all completely open development without having to pay royalties or use licenses.

Real examples. Our machines. Test of performance.

We have used three platforms to learn about the contribution that CUDA offers us to our daily work. Three levels of performance to know the capabilities of this API. The most powerful of them is a Quad-SLI GTX 295 which is the most powerful, in the domestic market, that we can have with Nvidia graphics chips. On the other hand, a veteran but still fighting 9800GX2 that can still give war, with its two graphics chips, to any GTS 250 or GTX 260 and finally an ION platform embodied in an Acer Aspire Revo 3600.