NVIDIA GT300 Has 512 Cores and Supports C++


The upcoming NVIDIA GPU, the GT300 codenamed Fermi will have 512 cores, up to 6GB GDDR5 memory (I hope 16kx16k textures will be supported 😉 ) and interesting news, the GT300 will support natively the C++ language:

Fermi architecture natively supports C [CUDA], C++, DirectCompute, DirectX 11, Fortran, OpenCL, OpenGL 3.1 and OpenGL 3.2. Now, you’ve read that correctly – Fermi comes with a support for native execution of C++. For the first time in history, a GPU can run C++ code with no major issues or performance penalties and when you add Fortran or C to that, it is easy to see that GPGPU-wise, nVidia did a huge job.

What does it mean ? No need CUDA anymore ?
Then this simple loop

for (i=0; i<1000000; ++i)

will be spread on GT300 cores ?



  • Does this mean that our daily software will use this GPU without been optimzed for it ?

  • Wiz

    Don’t hope for a magic task repartition, it’s the programmers job to optimize his software to make it able to run on hundreds of parallel tasks.

    I think “C++ Support” means somehow that cuda (or this particular technology does have another name ?) is now able to compile standard c++ code.

  • The question is how small a C++ subset that is available. Probably not anything like C++0x

  • martinsm

    I believe C++ support means just that CUDA will support C++ syntax. Nothing groundbreaking. Just additional feature to CUDA programs.

  • Keldor

    The thing they mean is that Fermi will support virtual functions and recursion. This means that, theoretically at least, it’ll be able to handle any C++ construct.

    In other words, C++0x is doable, though they might not have that out the door just yet.

    This also means that it will now be possible to write a interpreter to run on the GPU. The only barrier to having say Java or Python run on the GPU is now just getting someone to code it up.

  • Pierre

    I think that the C++ support will only enable for some part. For exemple I don’t think that the STL will be run over the GT300.

    I just think that the C++ support will be the object programming in cuda or directX compute support…

  • Interesting. Even if only the subset of real C++ will be supported it’ll still be a step forward and another proof that one day CPU will be useless.

  • Pingback: NVIDIA Fermi: the Nuclear GPU for Scientific Applications | The Geeks Of 3D - 3D Tech News