Nvidia releases public OpenCL drivers and SDK

[Final8y]

http://developer.nvidia.com/object/opencl-download.html

 

[deuse]

Thanks for the heads up
Having a read now

 

[Neolink]

Interesting stuff.

Could there become a point where you could run a computer simply from a GPU instead of having both a CPU and GPU? Thinking of the future of course.

 

[deuse]

Interesting stuff.

Could there become a point where you could run a computer simply from a GPU instead of having both a CPU and GPU? Thinking of the future of course.

sooner than you think

"NVIDIA today announced work with Microsoft to promote NVIDIA Tesla graphics processing units (GPUs) for high performance
parallel computing using the Windows HPC Server 2008"

 

[kylew]

Interesting stuff.

Could there become a point where you could run a computer simply from a GPU instead of having both a CPU and GPU? Thinking of the future of course.

Well nVidia don't have an x86 license, so they can't produce anything that will be able to run windows without a traditional CPU.

 

[deuse]

Well nVidia don't have an x86 license, so they can't produce anything that will be able to run windows without a traditional CPU.

They wont need one, they will run CPU and the GPU together in a co-processing model, simples

 

[Lightnix]

They wont need one, they will run CPU and the GPU together in a co-processing model, simples

Of course nobody'd get it because they wouldn't be able to play Windows' version of minesweeper.

 

[hideki]

hopefully nvidia see sense and port physx to opencl

 

[Rroff]

hopefully nvidia see sense and port physx to opencl

They'll have to really... otherwise whoever ports a physics API to open CL first is probably going to clean house... bullet looks like a fore-runner but its not as mature as physx or havok... I suspect nVidia already have an internal port of physx to open CL.

 

[hideki]

at least if that happens then it will be down to which company has the better drivers, and those can always be improved upon.

 

[kylew]

They wont need one, they will run CPU and the GPU together in a co-processing model, simples

But then that's not a PC running CPU calculations on just a GPU as per the question I responded to.

 

[Duff-Man]

Interesting stuff.

Could there become a point where you could run a computer simply from a GPU instead of having both a CPU and GPU? Thinking of the future of course.

No, you'll always need both.

The CPU for fast serial tasks, and the GPU for parallel vector-type computations. They may well end up on the same piece of silicon, but there will always be a distrinction between the CPU-like and a GPU-like part.

The "holy grail" would be a compiler which can analyse user code, and efficiently determine which portions should be sent to what bit of hardware. This is a long long way away though, a long way behind the hardware. As it stands today, it takes a huge amount of skill to manually specify what should be sent to what device.

 

[Andric]

No, you'll always need both.

The CPU for fast serial tasks, and the GPU for parallel vector-type computations. They may well end up on the same piece of silicon, but there will always be a distrinction between the CPU-like and a GPU-like part.

The "holy grail" would be a compiler which can analyse user code, and efficiently determine which portions should be sent to what bit of hardware. This is a long long way away though, a long way behind the hardware. As it stands today, it takes a huge amount of skill to manually specify what should be sent to what device.

exactly - future versions of c# are workign on these areas and I beleive java though I dont follow that. several other languages looking to make working with massive parrarel as well - pretty much need to rethink the traditional programming model though.