News Intel unveils 32nm process technology

http://www.bit-tech.net/news/hardware/2009/02/10/intel-unveils-32nm-process-technology/1

Intel has demonstrated the world's first 32nm processors and announed massive plans for its latest process technology.

 

Since the most intensive tasks are slowly moved towards GPGPU and CPUs will never even get close to GPU in efficiency and power, designing faster CPUs is just a waste of time. Even if they get to 11nm, the CPUs will be too slow for anything. Although it will be useful for GPUs.

 

For any highly-threadable tasks you're right. But if it's not breakable into many threads, CPUs still have a purpose.

 

Until of course they integrate the GPU into the CPU.... oh wait.
Still, maybe Intel should work on a better GPU to put in with the CPU... oh wait they are doing that as well aren't they.

 

Except the CPU's will still be needed to do the tasks that can't be put onto the GPU. After all not all types of processing can be pushed to the GPU's.

Hence any speed advantage in CPU's will still be able to benefit what remains on them.

 

I'm just excited that they are putting $7 billion in to the U.S. economy.

 

Not always. As soon as the Linux kernel will be able to compile on a GPU architecture, the CPU will be useless. GPUs are, in the worst case, faster than CPU by a factor of ~15 and can easily do anything a CPU can - faster. Well, the will be eventually. CUDA is, fr now, only a frontend, the GPU architecture needs some changes, not to mention Micro$oft's inability to adapt to anything new. So, for now we will use slow, outdated x86, while wasting all the real power on stupid, repetitive games. But some day they will reach 11nm, the core clock won't go any further and no more cores will fit on the die. And than maybe someone will say "x86 was outdated and underperforming anyway, let's do something new" It's like with the oil - the less there is left, the more money and interest goes into alternatives.

 

As mentioned above, for highly parallel and regular data structures, your statement makes sense, but irregular data structures, branched logic, and integer performance the GPU is the waste of time and power. The GPU also has to provide for full double precision, which I believe they have made massive progress, but it still not fully IEEE-754 compliant -- I need to check that, not sure if this is true anymore.

Anyway, each has strengths and weaknesses. As we are witnessing, computing is evolving to bring the two together. In the near term, nVidia has a good argument with CUDA and some real killer applications/results, but they have no CPU to pair with it to round it out -- i.e. you cannot currently run an OS on a GPU for example. Intel, on the other hand, as well as AMD with the ATI acquisition, are have the two major pieces to bring the best of both together.

In the long term nVidia is gonna get squeezed (my prediction)...

 

Amen to that brother.