1

Topic: New way for development of processors

I do not know considered or not, but I will tell. In general, a processor small square/crystal very small. And the case big. It is possible to increase a crystal of times in 1000 minimum and if to make in some layers - that in some millions times. To increase an amount of transistors to trillion and more. What for? And to build in directly the processor of ten cache gigabyte + to implement units for specific algorithms. For example to implement it is hardware all algorithms of cryptography, search, sorting and so forth And such architecture does not lead to magnification of capacity of the processor - the necessary units can be included and disconnected on demand. Why yet did not go on this way?

2

Re: New way for development of processors

S> It is possible to increase a crystal of times in 1000 minimum 1 nanosecond these are 30 sm (driving of an electromagnetic wave). That is, one clock period of the modern processors already faces a barrier of a velocity of light

3

Re: New way for development of processors

Hello, Shmj, you wrote: S> in general, a processor small square/crystal very small. And the case big. It is possible to increase a crystal of times in 1000 minimum and if to make in some layers - that in some millions times. To increase an amount of transistors to trillion and more. S> What for? And to build in directly the processor of ten cache gigabyte + to implement units for specific algorithms. For example to implement it is hardware all algorithms of cryptography, search, sorting and so forth S> And such architecture does not lead to magnification of capacity of the processor - the necessary units can be included and disconnected on demand. S> why yet did not go on this way? I think percent of spoilage such huge  will be nearby 100. Somewhere because of defect disconnect the cryptography unit, somewhere search, and somewhere sortings etc. And each percents turns out not similar on descended from the same pipeline.

4

Re: New way for development of processors

Hello, Shmj, you wrote: S> in general, a processor small square/crystal very small. And the case big. It is possible to increase a crystal of times in 1000 minimum and if to make in some layers - that in some millions times. To increase an amount of transistors to trillion and more. To cool more difficult. S> what for? And to build in directly the processor of ten cache gigabyte + to implement units for specific algorithms. For example to implement it is hardware all algorithms of cryptography, search, sorting and so forth "Thought fresh, and the main thing - original" ((, Strugatsky) Who will use these algorithms from an existing software? And search with sorting at programs where it is important - already the. Take the same .

5

Re: New way for development of processors

Hello, Shmj, you wrote: S> Why yet did not go on this way? ASIC/FPGA

6

Re: New way for development of processors

Hello, Shmj, you wrote: S> Why yet did not go on this way? The basic possibilities to refine productivity within the limits of the same transistor budget full, but all of them break about small nuance: compatibility. For example, for the offered variant a bottleneck - expenses for synchronization. It is necessary to bypass completely to change model of calculations and to pass to actors  erlang/scala/orleans and, most likely, to stm (a quite good educational program from author Functors, Applicatives, And Monads In Pictures and grokking alghorithms). It, in turn, demands a failure from a background-nejmana and heaps of extreme operation under the reinvention __ that is turned out for the last half a century for "traditional" computers. That the most amusing, as a result we do not achieve increase of productivity neither for an individual flow of calculations, nor for  - the law of Amdalla in the help. But there will be a possibility to raise __ system performance for the account more an effective utilization of idle units. And here we with interest think that for easy  things at us already is gpgpu/fpga/etc. For actors - effective enough  implementations up to embedded-systems + the arm-clusters getting in 2u-blade. For more effective utilization of iron - analogs intel speed step / hyperthreading. And at last that productivity so ceased to be a problem that the idea to spread application in the form of containers-VM looks quite sensible idea. And from all it there is a reasonable question: and actually ? P.S. And yes, whether to move to programming philosophy? A subject not strongly , as on me.  hung, support.

7

Re: New way for development of processors

Hello, Shmj, you wrote: S> It is possible to increase a crystal of times in 1000 minimum S> Why yet did not go on this way? The elementary greed. The plate - one to do of it of one percents or seven hats 100 - that is more favourable? For the same reason we sit till now on confused copper when for a long time it was already possible to connect ALL components of a computer an optical fiber. Well and time went such , OPTICAL CALCULATIONS were possible still 20 years ago. Obviously, with a velocity of light. Is not went - to play with processor pinches easier! (Spat in a back of the head of Intel)

8

Re: New way for development of processors

Hello, Shmj, you wrote: S> Why yet did not go on this way? Because the more transistors, the probability there is more that any of them it turns out rejected. If  too difficult, the output of suitable crystals is too small. And as in the price of the suitable the price unsuitable is put also, suitable will cost the expensive.

9

Re: New way for development of processors

S> in general, a processor small square/crystal very small. And the case big. It is possible to increase a crystal of times in 1000 minimum and if to make in some layers - that in some millions times. To increase an amount of transistors to trillion and more. Here there is a struggle for that that the size of the chip to reduce by some percent, and at all because of consumption. And it is simple because it is possible to cut more, the price is less. And here I suggest to increase in 1000 times. Well and it will be a million times more expensive. One thousand for a size, yes one thousand for that that suitable an output decreases

10

Re: New way for development of processors

Hello, Pzz, you wrote: Pzz> Because the more transistors, the probability there is more that any of them it turns out rejected. If  too difficult, the output of suitable crystals is too small. And as in the price of the suitable the price unsuitable is put also, suitable will cost the expensive. So... It is possible hardly to increase the size of the transistor, not to try to make the smallest. Well if the crystal is 100*100*5 mm - unless it much? Into any notebook gets. To make a reserve - if one unit it did not turn out - it is disconnected it and the reserve is connected.

11

Re: New way for development of processors

Hello, Sinix, you wrote:> the idea to spread application in the form of containers-VM looks quite sensible idea you, a companion, is a lot of  in the post, but about the most scandalous I all the same will hint. Containers just also appeared because productivity  a problem. But that the most interesting, containers have no relation to virtual machines.

12

Re: New way for development of processors

Hello, antropolog, you wrote:>> the idea to spread application in the form of containers-VM looks quite sensible idea A> But that the most interesting, containers have no relation to virtual machines. Aha, . There on an idea was "in VM in a cloud", formulated crookedly. There is a variant of the answer on "and ?" - .

13

Re: New way for development of processors

Hello, Shmj, you wrote: S> What for? And to build in directly the processor of ten cache gigabyte + to implement units for specific algorithms. For example to implement it is hardware all algorithms of cryptography, search, sorting and so forth Spherical algorithms in vacuum happen only in books. In real life algorithms are nonseparable from data structures.

14

Re: New way for development of processors

Hello, Kolesiki, you wrote: K> Well and time went such , OPTICAL CALCULATIONS were possible still 20 years ago. Obviously, with a velocity of light. Is not went - to play with processor pinches easier! (Spat in a back of the head of Intel) Optical __ calculations. Small such nuance.

15

Re: New way for development of processors

Hello, Shmj, you wrote: S> Hello, Pzz, you wrote: Pzz>> Because the more transistors, the probability there is more that any of them it turns out rejected. If  too difficult, the output of suitable crystals is too small. And as in the price of the suitable the price unsuitable is put also, suitable will cost the expensive. S> so... It is possible hardly to increase the size of the transistor, not to try to make the smallest. Well if the crystal is 100*100*5 mm - unless it much? Into any notebook gets. There will be more area of heating up, more system of cooling, etc. will be necessary

16

Re: New way for development of processors

Hello, Pzz, you wrote: Pzz> Hello, Shmj, you wrote: S>> Why yet did not go on this way? Pzz> because the more transistors, the probability there is more that any of them it turns out rejected. If  too difficult, the output of suitable crystals is too small. And as in the price of the suitable the price unsuitable is put also, suitable will cost the expensive. It is thought, as  all this business on one crystal it becomes essential more expensively, with  probability .

17

Re: New way for development of processors

Hello, Sharov, you wrote: S> It is thought, as  all this business on one crystal it becomes essential more expensively, with  probability . Nobody speaks that at once. Gradually to increase the size of a crystal, to raise accuracy of the press. The main thing that the big perspectives for development.

18

Re: New way for development of processors

Hello, Shmj, you wrote: S> Why yet did not go on this way? The output of the suitable aspires to zero.

19

Re: New way for development of processors

Hello, iZEN, you wrote: ZEN> the Output of the suitable aspires to zero. Nothing hinders hardly to increase the size of transistors, is not mandatory 21 nanometer. It is possible and 35 nanometers - the big difference are not present. Besides it is possible to make reservation, that is on one crystal a store of units.

20

Re: New way for development of processors

Hello, Shmj, you wrote: S> And such architecture does not lead to magnification of capacity of the processor - the necessary units can be included and disconnected on demand. S> why yet did not go on this way? Because it is necessary for nobody. The cryptography, everyones SSE and virtualization is for a long time. And outside of it the formula turns out "to accelerate average productivity on 0.1 % by magnification of cost at 10 % and reliability lowerings at 1 %". Nafig-nafig.

21

Re: New way for development of processors

Hello, bazis1, you wrote: B> Because it is necessary for nobody. The cryptography, everyones SSE and virtualization is for a long time. And outside of it the formula turns out "to accelerate average productivity on 0.1 % by magnification of cost at 10 % and reliability lowerings at 1 %". Nafig-nafig.  in tens all changes gigabyte. Algorithms work with data structures. You resulted only AES and why is not present SHA256, ECDSA and ? Only cryptography, but after all at big  it is possible to add it and search/sorting. It is possible to make hardware video-codecs and so forth

22

Re: New way for development of processors

Hello, Shmj, you wrote: S> in tens gigabyte all changes Kesh. Algorithms work with data structures. Aha, accelerates in 10 times of 0.01 % of the program. S> you resulted only AES and why is not present SHA256, ECDSA and ? Because AES is real bottleneck by operation with SSL and in everyones encrypted storage. Hash coding bottleneck th is not. S> only cryptography, but after all at big  it is possible to add it and search/sorting. The strong acceleration of search/sorting actually only for  data-centers. And for them can be cheaper more gland, than  with hardware sorting and the strange caches is simple to buy. S> it is possible to make hardware video-codecs and so forth What for? It accelerates what tasks? Well will be at users at a video playback not 1 % CPU usage, and 0.1 %. And? Really from it benefit only any video hostings or studios, but besides it to buy in addition gland easier.

23

Re: New way for development of processors

Hello, Sinix, you wrote: S> the Basic possibilities to refine productivity within the limits of the same transistor budget full, but all of them break about small nuance: compatibility. Not compatibility, and physics laws. For example one only heat removal hinders to put in a crystal at once all successively as offers the HARDWARE. S> for example, for the offered variant a bottleneck - expenses for synchronization. It is necessary to bypass completely to change model of calculations and to pass to actors  erlang/scala/orleans and, most likely, to stm (a quite good educational program from author Functors, Applicatives, And Monads In Pictures and grokking alghorithms). It, in turn, demands a failure from a background-nejmana and heaps of extreme operation under the reinvention __ that is turned out for the last half a century for "traditional" computers. I how many looked this "a quite good educational program", and did not understand: 1 as it works STM 2 why passage from pictures-children at once to Haskelju not bad

24

Re: New way for development of processors

Hello, Shmj, you wrote: S> And such architecture does not lead to magnification of capacity of the processor - the necessary units can be included and disconnected on demand. S> why yet did not go on this way? There are in essence unremovable problems 1 heat removal - __ a non-linear problem 2 amount of outputs a crystal - one more strongly non-linear problem 3 time of switching of the transistor is restricted by properties of the semiconductor 4 reduction of the sizes of the transistor restricted by the size of atom Roughly speaking, in volume unit you can  only  an amount of devices. It means that different devices will communicate all the same on the general bus. From here it is clear that this bus and will be a bottleneck. Actually, partly progress also went on that way that you offered, only is more prosy - processor  incorporates the most different devices, only sortings of everyones there are not present. If the sorting device works through  the bus to us it is absolute on a drum how fast the sorter works - all the same the limit will be equal to throughput of the bus, and it means that the normal code consults not worse.

25

Re: New way for development of processors

I> Roughly speaking, in volume unit you can  only  an amount of devices. It means that different devices will communicate all the same on the general bus. From here it is clear that this bus and will be a bottleneck. The remark. The general buses already are not present. In rough approximation a matrix.