As I know for now because of laws of Physics there will be not any tangible improvements in CPU cycles per second for the nearest future. However because of Von Neumann bottleneck it seems to not be an issue for non-server applications. So what about RAM, is there any upcoming technologies that promise to improve memory speed or we are stack with the current situation till quantum computers will come out from labs?
Tell me more
×
Programmers Stack Exchange is a question and answer site for
professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.
|
|
We are not stuck yet (at least not to the point of waiting for quantum computer). On the memory front: DDR4. JEDEC expects the DDR4 specification will be finalized in 2011 with commercial production beginning 2012. DDR4 will hit the server market first and full scale transition is expected ~2015. Samsung has already started trials of a DDR4 module that can achieve data transfer rates of 2.133 Gbps (with ~4.2 Gbps expected by 2015). As for CPUs an physics: Intel currently predicts they might hit the physics barrier in atomic structures or power density by 2020. With 22nm fabrication ramping up and new transistor structures entering general use, increases in CPU performance should continue for a bit longer. GPGPU computing can also offer significant increases in bulk calculation performance but, as Martin Beckett mentioned in his answer, slow bus transfer speed might limit use in general applications. Of course, you never know what might show up next. There is a lot of money on the table. |
|||
|
|
|
I don't think we are so stuck really. Main memory bus speeds keep getting faster with every new generation of processors 122, 300, 500MHz ... Also expensive models of processors getting more and more L2 cache space etc. Registers are also memory - number of available registers keep going up as well. However you need to keep updating your OS and compiles to get benefits from all that, but that is different issue. |
|||
|
|
|
One thing I can see is more of the memory moving onto the graphics card. Given that it's the screen that is normally performance sensitive, and moving more of the general purpose calculations to Cuda/OpenCL it seems to make more sense to have the fast RAM local to the graphics and copied to the slower CPU when needed, rather than having it near the CPU and moving it over a slow bus to the fast GPU |
|||
|
|
