• Post Reply Bookmark Topic Watch Topic
  • New Topic
programming forums Java Mobile Certification Databases Caching Books Engineering Micro Controllers OS Languages Paradigms IDEs Build Tools Frameworks Application Servers Open Source This Site Careers Other all forums
this forum made possible by our volunteer staff, including ...
Marshals:
  • Campbell Ritchie
  • Liutauras Vilda
  • Bear Bibeault
  • Tim Cooke
  • Junilu Lacar
Sheriffs:
  • Paul Clapham
  • Devaka Cooray
  • Knute Snortum
Saloon Keepers:
  • Ron McLeod
  • Tim Moores
  • Stephan van Hulst
  • Tim Holloway
  • Frits Walraven
Bartenders:
  • Carey Brown
  • salvin francis
  • Claude Moore

Clock Speed vs Multi-Core  RSS feed

 
Greenhorn
Posts: 15
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
During a presentation by a scala engineer, i heard he explained due to the drawbacks of clock speed multi-core processor has been invented to solve the problem.

He also said, even though multi-core processor exist today it is not efficiently used until we do functional programming

Questions

1. What is a clock speed? what is the maximum clock speed exist so far? What are the limitations?
2. How does clock speed matters in terms of processor?
3. Does clock speed MHZ cannot be increased and that's the reason multi-core processor is invented?

Thanks
 
Sheriff
Posts: 24374
55
Eclipse IDE Firefox Browser MySQL Database
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator

Joseph Michael wrote:During a presentation by a scala engineer, i heard he explained due to the drawbacks of clock speed multi-core processor has been invented to solve the problem.



I'm not sure I even agree with this premise.
 
Saloon Keeper
Posts: 10136
214
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Electronic circuits require a pulse of electricity to update the values stored in memory and registers with newly calculated values. Such a pulse is called a 'tick' and the electronic component that generates ticks is called a 'clock'.

When a signal changes (for instance, the inputs or outputs of the electronic component that performs arithmetic), it needs time to stabilize because even though pulses travel fast, they don't travel with unlimited speed. If the clock ticks before the signals in the previous tick have stabilized, then signals can interfere with each other and you get crazy results.

How much time it takes for signals to stabilize depends how big the transistors are that make up the electronic components, how much distance there is between them and how many of them you need to make a component. If you create smaller transistors that are also closer together, you can increase the speed of the clock because signals need less time to travel through and between transistors.

Up until a few years ago, semi-conductor manufacturers managed to create smaller and smaller transistors, which allowed processor manufacturers to double the clock speed every so many years. That era has ended because transistors are now so small that quantum mechanical effects such as electron tunneling become significant enough that they interfere with the normal operation of the transistor. Until we increase our understanding of quantum mechanics and manage to work around or even leverage quantum mechanical effects, processor cores won't become significantly faster.

I think world records for clock speeds are around 8GHz these days. Note that for commercial processors it's important to consider how much energy the processor draws and how much heat it generates. That's why the fastest commercial processors are rated for clock speeds of around 4GHz.
 
Saloon Keeper
Posts: 5412
143
Android Firefox Browser Mac OS X Safari Tomcat Server VI Editor
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator

Joseph Michael wrote:due to the drawbacks of clock speed multi-core processor has been invented to solve the problem.


I think what he meant is that -since clock speeds have not been getting much faster for years now- adding more cores (and to a lesser degree, larger caches) is the main approach to faster CPU hardware.

He also said, even though multi-core processor exist today it is not efficiently used until we do functional programming


It's probably true that it is easier with FP code to get higher parallel speedups for user-written code than with other approaches. But the OS and the JVM have support for utilizing multiple cores for themselves, so the user-written code does not need to do all the parallelization effort. And in server environments where you have multiple concurrent users getting parallel speedups is much easier, since the code run for each user is largely independent of the code for all other users. Compute-intensive code for a single process may be better off offloading work to a GPU rather than try to use the CPU more efficiently.
 
Marshal
Posts: 64166
215
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator

Stephan van Hulst wrote:. . . consider how much energy the processor draws and how much heat it generates. . . . .

If I remember correctly, power consumption is proportional to the clock speed multiplied by the voltage squared. Smaller transistors will run at a lower voltage, but even so the heat production from a computer chip per unit volume is greater than an electric cooker. As well as those resistive losses, there are two other losses mandated by the laws of thermodynamics. One is that power is unavoidably required to make the computer run at a particular speed. The other is that each bit of information deleted releases kTlog2 of heat, which is 3×10⁻²¹J at about 40.5℃. 3×10⁻²¹J might not seem a lot, but when you multiply it by clock speed and number of transistors, it actually constitutes an appreciable part of the power cost, maybe as much as 0.1% nowadays. The amount of power for cooling a chip is similar to the chip's power consumption itself.
 
Stephan van Hulst
Saloon Keeper
Posts: 10136
214
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator

Campbell Ritchie wrote:The other is that each bit of information deleted releases kTlog2 of heat


I'm not an expert, but I would think that most information is deleted by directing it to the ground pin, and not from friction. That's not to say that information lost through friction is not significant, as demonstrated by CPUs running really hot.
 
Campbell Ritchie
Marshal
Posts: 64166
215
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
It doesn't make any difference how the information is deleted. Either the current fills a previous 0 or current runs out of a previous 1. There are some instances where you can overwrite memory without deleting information, e.g. filling previously empty memory, which contained no information to delete. In the case of something like the common AND gate, you have two inputs and one output. You cannot always tell from the output what the inputs were; there are three possible inputs producing a 0 output. On average you put in 2 bits of information and get out something like 1.26 bits, and in some circumstances you have more current running through the output which heats up more than usual.
 
Stephan van Hulst
Saloon Keeper
Posts: 10136
214
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Except that's not how most gates operate. An AND gate is constructed from two NAND gates, and currents flowing through a NAND gate are collected in the ground wire.



Let's say that both A and B have a high potential. This means there is a low potential difference between Vcc and A/B, so VT1 is in the off-state and there will be a high potential difference between the base of VT2 and GND, meaning VT2 is in the on-state. Because of this, a current flows from Q to GND and Q will have a low potential.

When we measure that Q is low, we know that both A and B are high. When the circuit switches and Q is high, we know that A is low, B is low, or both are low. We have lost information. This information loss is represented by a current flowing from A or B back to the GND connection of the previous NAND gate.

So yes, the information loss DOES cause heat-up, but most of that is from friction in the ground wire.

Computer processors heat up because of parasitic capacitances between conductors that are close together. If you put two transistors in close proximity, the magnetic field between them will create a capacitor that draws energy. When the circuit switches, that energy is dissipated in resistors, creating heat. So heat is generated every tick, regardless of whether any information was lost during that tick.
 
Campbell Ritchie
Marshal
Posts: 64166
215
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
The latter is resistive loss, which might be reduced by reducing voltage, current, etc., or by moving to different materials, e.g. carbon nanotubes. The thermodynamic losses are fixed. That is why there is interest in reversible computing, which doesn't lose information, but that interest is obscure and only a few people know about it.
 
Stephan van Hulst
Saloon Keeper
Posts: 10136
214
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator

Campbell Ritchie wrote:The latter is resistive loss, which might be reduced by reducing voltage, current, etc., or by moving to different materials, e.g. carbon nanotubes. The thermodynamic losses are fixed.


Reducing voltage decreases the speed at which circuits can operate. Using nano-tubes does not eliminate the generation of magnetic fields when a current flows, as far as I know, so parasitic capacitances will still present themselves. As a matter of fact, I believe that nano-tubes can be used to create super-capacitors. I'm not sure if that has any bearing on parasitic capacitances, but it doesn't sound promising.

I don't believe reversible computing will have a significant impact on heat generation, and that other sources of heat will always dwarf the heat generated by loss of information.
 
When you have exhausted all possibilities, remember this: you haven't - Edison. Tiny ad:
Create Edit Print & Convert PDF Using Free API with Java
https://coderanch.com/wiki/703735/Create-Convert-PDF-Free-Spire
  • Post Reply Bookmark Topic Watch Topic
  • New Topic
Boost this thread!