What are cuda cores?

[The Dragon]

what are cuda cores?

[Glenwing]

They are GPU cores, same thing as stream processors. NVIDIA just has a fancy name for them.

[The Dragon]

They are GPU cores, same thing as stream processors. NVIDIA just has a fancy name for them.

I don't know what stream processors are

[Glenwing]

I don't know what stream processors are

 

GPU cores. They are all the same thing. 

[AlwaysFSX]

I don't know what stream processors are

Do you know what a cpu core is? Same idea, but used in parallel.

[SiverJohn17]

To summarize a paragraph from one of the two CUDA Programming books in front of me. (Of which I need to return both to their rightful owners.) Cuda is an architecture upon which the Arithmetric logic unit (ALU) can be used for more general processing. "...these ALUs were built to comply with IEEE requirements for single-precision floating-point arithmetic and were designed to use an instruction set tailored for general computation rather than specifically for graphics." The rest is basically giving them access to system memory and a few other things to make them better for computation as well as graphical processing.

 

-Source: CUDA By Example -Sanders Kandrot

[The Dragon]

GPU cores. They are all the same thing. 

Why do GPUs have so many compared to cpus then?

[Glenwing]

Why do GPUs have so many compared to cpus then?

 

Different types of computation tasks benefit from different design. CPUs have a few strong cores, GPUs have loads and loads of really little cores. Rendering graphics is highly parallel, which means it's easy to spread the work among as many cores as you have, and with a lot of small cores it can be done much more quickly and efficiently than on a CPU. But some types of computations can't be done in parallel, for example if each stage of the calculations requires the results from the previous stage. Then they can only be done one at a time, and it's better to have a single strong core that can do them quickly in a series, called "serial" computation.

[The Dragon]

Different types of computation tasks benefit from different design. CPUs have a few strong cores, GPUs have loads and loads of really little cores. Rendering graphics is highly parallel, which means it's easy to spread the work among as many cores as you have, and with a lot of small cores it can be done much more quickly and efficiently than on a CPU. But some types of computations can't be done in parallel, for example if each stage of the calculations requires the results from the previous stage. Then they can only be done one at a time, and it's better to have a single strong core that can do them quickly in a series, called "serial" computation.

Ok so now that I know generally what cuda cores are another question is, is it better to have more or does it only sometimes benefit like with cpus where some things favor fewer faster cores and others work better with more cores.

[Glenwing]

Ok so now that I know generally what cuda cores are another question is, is it better to have more or does it only sometimes benefit like with cpus where some things favor fewer faster cores and others work better with more cores.

 

Pretty much more is always better. GPUs only get tasked with things that really benefit from tons of cores.

[The Dragon]

Pretty much more is always better. GPUs only get tasked with things that really benefit from tons of cores.

Ok cool