New Nvidia GPUs will support “real C++”

 

Gallery

By Darien Graham-Smith in San Jose

Posted on 1 Oct 2009 at 20:14

Nvidia’s new Fermi graphics architecture, announced yesterday at the company’s GPU Technology Conference, will run “real C++” applications, the company has confirmed.

“The capabilities of C++ are coming to CUDA,” Sanford Russell, general manager for CUDA, assured PC Pro. “We have template support, and the hardware can do pointers now. We really couldn’t do that before.”

But generic C++ code will not magically benefit from being compiled under CUDA.

“You still need a few extensions to handle the parallelism,” he explained. “You need a few extensions that can launch a thousand threads, statements that tell the compiler ‘this heads to the GPU, this heads to the CPU.’ These are the same extensions as we used in our C compiler.”

“But our objective is that a developer can sit down and go ‘it’s C++’. And that a company that has a raft of C++ programmers can be productive right up front.”

No threat from DX11

Russell dismissed suggestions that DirectX 11’s upgraded DirectCompute API might be a threat to CUDA.

“DirectCompute [as implemented in DirectX 10] already runs on 180 million Nvidia GPUs,” he countered.

“But you go out there,” he suggested, gesturing through the window to the streets of Silicon Valley, “and you say ‘everybody who programs in C, stand on this side of the street. And everybody who uses an API, go stand over there.’ There’d be a few people on the API side and a whole bunch on the C side.”

“The market will decide what they want to use, not Nvidia.”