8,464
edits
Restless Souls/Technology: Difference between revisions
thoughts on processors, more to come
Paradox-01 (talk | contribs) m (typo) |
Paradox-01 (talk | contribs) (thoughts on processors, more to come) |
||
| Line 1,435: | Line 1,435: | ||
The companies will push forward '''multi-core processors''' and '''vertical stacking'''. For example, Samsung already uses vertical stacking in flash memory, by now you should have heard of their V-NAND SSD. The idea to use the third dimension should soon find its way into processor production to fulfill Moore's law. | The companies will push forward '''multi-core processors''' and '''vertical stacking'''. For example, Samsung already uses vertical stacking in flash memory, by now you should have heard of their V-NAND SSD. The idea to use the third dimension should soon find its way into processor production to fulfill Moore's law. | ||
A computer is meant to be a universal problem solver. But special problems require special solutions. So, CPUs got company - most notably - by coprocessors for networking, sound (DSP), graphics (GPU) and physics (PPU). Those coprocessors were - and most still are - available as add in cards. There's a tendency that over time coprocessors are combined on one product for weaker customers. | |||
Harnessing the magnetic property of electrons will not only result in less power consumption but also high clock rates. In the past years the clock speed stagnated between 4 and 5 GHz due to overheating. Processors based on spintronics won't heat up that fast, so can operate at higher frequencies. And higher frequencies means more calculations per second. | * mainboards having "onboard" chips: networking, sound, graphics | ||
* combined CPU and GPU: AMD's APU and Intel's CPUs with integrated HD Graphics | |||
* GPU with PPUs: GPU with CUDA (after Nvidia bought PhysX), general purpose GPU (GPGPU) | |||
Like in organic evolution, after a phase of specialization, a phase of socialization follows. ^_^ | |||
Progress on the hardware level is only one side of the coin. You also need software to utilize it. A Russian team demonstrated that [http://phys.org/news/2016-06-scientists-pc-complex-problems-tens.html some scenarios requiring supercomputing can actually by handled on a desk PC] after installing a new graphics card [http://arxiv.org/pdf/1508.07441.pdf (GTX-670, CUDA supported) and self-written software.] | |||
As for 2016, still many PC programs can't draw advantage from multi-cores processors. To be fair, not every program need multiple cores - just as a simple text editor. Yeah, for what? | |||
Anyway, after these tricks (stacking, multi-core architecture, parallelism) has been fully exploited, the companies will be forced to turn to new technologies such as spintronics in 2025. | |||
Harnessing the magnetic property of electrons will not only result in less power consumption but also high clock rates. In the past years the clock speed stagnated between 4 and 5 GHz due to overheating. Processors based on spintronics won't heat up that fast, so they can operate at higher frequencies. And higher frequencies means more calculations per second. | |||
Next, we will see photonic processors. If we give spintronics 15 years to become fully developed, photonics should arrive at 2040. These appear quite late in this technological vision since individual light particles (photons) are even more difficult to handle than electrons. | Next, we will see photonic processors. If we give spintronics 15 years to become fully developed, photonics should arrive at 2040. These appear quite late in this technological vision since individual light particles (photons) are even more difficult to handle than electrons. | ||
Oh yeah, and of course quantum computers... You often hear of "breakthroughs" but still | Oh yeah, and of course quantum computers... You often hear of "breakthroughs" but still practical examples are nowhere to be found. That's because the ToDo-lists for all its parts are huge. Most parts have to be constructed from scratch and the underlying logic rests on fragile particle states like entanglement. While quantum computer might just gain traction by 2050, we should see more and more hybridization of processor technologies in the meantime. | ||
As you already guessed these numbers (2025, 2040, 2050) are just rough estimates. All I can do here is explaining my estimates in some more detail. | |||
Let's make also an approach via economy. A company will only invest a serious amount of money into a new R&D field if it doesn't have another option. If they try it anyway, there's a high risk to fail. Latest example is HP's the machine, a still-hypothetical computer driven by memristors. The technological gap is still too wide. | |||
"We have silicon wafers. Find something to continue with that." Spintronics... | |||
[...] | |||
Chance of niche products. | |||
[...] | [...] | ||