On 7 January 2011 15:56, Carl Jokl <carl.j...@gmail.com> wrote: > There is also a trend towards integrating more of the computer into > the same dye as the CPU. > > The memory controller went in first. The GPU is going in now. Some CPU > (such as the Utrasparc T1 and T2) integrate a network controller. > I could well expect the PCI Express controller to be a hot candidate > for getting integrated. Also an APU (audio processing unit) for > consumer processors. > Some other controllers could also go in such as maybe USB and SATA. > > This kind of increased integration would reduce size and potentially > save power. This also is a definite trend but it is hard to say how > far it will go. > Rather than the CPU dye being part of the computer, this system on a > chip is the computer. > > I have looked at the 1000 core processor article. It does mention > something along the lines of not being a general purpose computer as I > read it and being aimed more at embedded scenarios. I might have > misunderstood. > I think a certain limited amount of use of the 3rd dimension happens > already for modern processors. I don't know that this solves overall > power consumption and power dissipation issues and I could see > potential for expansion that way making chips harder to cool. > >
Also interesting to note is that a significant portion of current processors is taken up by local cache memory implemented in SRAM, it's extremely likely that this will be implemented in the future using a lower-power, higher-density alternative such as MRAM, thus allowing more, smaller, cooler cores. Most predictions of future core counts are based on Moore's law, which refers to the increasing transistor count of CPUs (not increasing clock speeds, as some people think). This has held true throughout the migration to multi-core architectures. It's ironic then that a shift from SRAM (transistors) to MRAM (capacitors) could be the one thing that finally breaks it! > On Jan 7, 3:33 pm, Kevin Wright <kev.lee.wri...@gmail.com> wrote: > > On 7 January 2011 15:24, Carl Jokl <carl.j...@gmail.com> wrote: > > > > > > > > > > > > > I am not sure the 1000s or cores in a desktop machine principle is > > > realistic. This seems like just taking the the current core count > > > increase as being able to keep going up forever. It seems like a > > > casual assumption was also made about processor clock rate until that > > > hit its practical limit about 2003. From what I read it is expected to > > > be unlikely that manufacturing processes can shrink bellow 16 > > > nanometers. With mainstream processors already being produced at 32 > > > nanometers it seems like it may only be 2 die shrink generations till > > > we reach that limit. If a processor can be manufactured today with 8 > > > cores on a 32 nanometer manufacturing process then at 16 nanometers > > > the same processor should fit into more or less a quarter of the size. > > > That would give 32 cores. If a processor today can be manufactured > > > with more simple cores say 32 of them then at 16 nanometers there > > > would be 128 cores. If 64 cores today then that becomes 256. This > > > brings us to between tens of cores to a couple of hundred or so > > > depending on the complexity of each core. The idea of thousands of > > > cores for a client machine at least may not be realistic. The trend is > > > also now to combine CPU and GPU on a single die. The GPU is generally > > > larger than the CPU so such combined processors would have to > > > sacrifice processing cores for GPU cores or other items integrated on > > > the die. In practice 3-4Ghz ended up being a limit. If 16 nanometer > > > ends up being a practical limit too then there comes a cut of point > > > after which more cores cannot be added. This too in the foreseeable > > > future. Perhaps it will be possible to go bellow 16 nanometers but > > > fundamentally the size of atoms is fixed meaning there is a limit to > > > how small things can be shrunk. Increasing the size of the die > > > increases heat, power consumption and potential for timer signal > > > latency issues so there are limits on expanding that way too. > > > > Except... 1000 cores has already been done! > http://lmgtfy.com/?q=1000+core+processor > > Of course, there's also a 3rd dimension available, so these things can be > > stacked in a single package. > > > > There's another issue also at work here, having a greater number of > > less-powerful cores uses less electricity. So there's less heat to > > dissipate, fine-grained power management becomes possible by turning off > > cores one at a time, and carbon footprints go down. I think it's also > > reasonable to expect that CPUs and GPUs will gradually merge back into a > > single unit. > > > > The industry is absolutely headed in this direction... > > > > -- > > Kevin Wright > > > > gtalk / msn : kev.lee.wri...@gmail.com > > <kev.lee.wri...@gmail.com>mail: kevin.wri...@scalatechnology.com > > vibe / skype: kev.lee.wright > > twitter: @thecoda > > -- > You received this message because you are subscribed to the Google Groups > "The Java Posse" group. > To post to this group, send email to javapo...@googlegroups.com. > To unsubscribe from this group, send email to > javaposse+unsubscr...@googlegroups.com<javaposse%2bunsubscr...@googlegroups.com> > . > For more options, visit this group at > http://groups.google.com/group/javaposse?hl=en. > > -- Kevin Wright gtalk / msn : kev.lee.wri...@gmail.com <kev.lee.wri...@gmail.com>mail: kevin.wri...@scalatechnology.com vibe / skype: kev.lee.wright twitter: @thecoda -- You received this message because you are subscribed to the Google Groups "The Java Posse" group. To post to this group, send email to javapo...@googlegroups.com. To unsubscribe from this group, send email to javaposse+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/javaposse?hl=en.
