nVidia Bags IBM - To Ship Tesla for Datacenters .

[taltamir]

GPU just isn't a good name of itself (which is why we already have extended terms such as GPGPU or CGPU).
Most names describe the function of a processing unit, but a GPU describes an application instead of a function.
FPU is a floating point unit.
DSP is a digital (or dedicated) signal processor.
These terms describe what the processing units do, rather than what you could use them for.
If you were to describe what a GPU does, then it would be something like parallel stream processor, or just SIMD unit
Even DSP would be more or less applicable.

But if we look at how GPUs are designed today (actually GPGPU/CGPU architectures), and where things are likely going with Larrabee, Cuda and Fusion... the designs will remain close to what we call 'GPUs' today, rather than what we know as CPUs today.

I am aware of most of that, but you yourself said "GPU describes an application instead of a function"... which means GPUs (the application of a functionality) will disappear to be replaced with just said functionality rolled into the CPU.
And the application of functionality that is GPU actually has somewhat different functionality today then it did at its inception. (further distancing it).

 

[cbn]

GPU just isn't a good name of itself (which is why we already have extended terms such as GPGPU or CGPU).
Most names describe the function of a processing unit, but a GPU describes an application instead of a function.
FPU is a floating point unit.
DSP is a digital (or dedicated) signal processor.
These terms describe what the processing units do, rather than what you could use them for.
If you were to describe what a GPU does, then it would be something like parallel stream processor, or just SIMD unit
Even DSP would be more or less applicable.

But if we look at how GPUs are designed today (actually GPGPU/CGPU architectures), and where things are likely going with Larrabee, Cuda and Fusion... the designs will remain close to what we call 'GPUs' today, rather than what we know as CPUs today.

Scali, thanks for the explanations.

 

[cbn]

Does anyone know why Nvidia didn't use 295 GTX as the basis for Tesla cards last generation?

 

[Scali]

I am aware of most of that, but you yourself said "GPU describes an application instead of a function"... which means GPUs (the application of a functionality) will disappear to be replaced with just said functionality rolled into the CPU.

A rose by any other name...

And the application of functionality that is GPU actually has somewhat different functionality today then it did at its inception. (further distancing it).

That's just progress... 20 years ago we didn't do the same with our CPUs as we do today either. Just a result of technology becoming more powerful and more flexible.