[Autocrat]

I have heard some people saying that the performance of dual core CPUs is found out by multiplying their rating by 2. Like a dual core processor 2.6 GHz will have a total performance of 5.2 GHz. This sounds s horseshit to me. If that was so, then who has payed me to tell that the performance of my quad core is 2.33 GHz instead of 9.32 GHz?

Is it true that total performance = rating x number of cores?

Moreover, some say that quad core is nothing but a single CPU having abilities equal to four computers. Four what? Four PDP 11's or 4 Pentium 4's?

Why do we see a single CPU usage graph in task manager for single core CPUs, two graphs for dual core and four for quad cores?

[raziel_]

Quote

the performance of my quad core is 2.33 GHz instead of 9.32 GHz?

the performance of your quad core is 4x 2.33 GHz. i mean 4 processors each at 2.33 GHz. you may say it is 9.32 GHz the total power you have for your PC.

This post has been edited by NoBrain: 27 April 2010 - 01:38 AM

[no2pencil]

** Moved to Computer Support & Operating Systems **

That number is how many instructions the system can handle. But there are many many factors that go into how fact the instructions get passed to the processor, such as the Amount of RAM, Front Side Bus speed, & Cache on the CPUs. It's more than just how many processes that the CPUs can handle, but how fast they can get the instructions.

Or at least that's how I understand it.

[calvinthedestroyer]

I saw a talk about multithreaded programing or parallel programming. The talk focused on taking advantage of the multiple cores of a CPU to write faster code. The talk showed comparisons of code that was multi threaded and non-multithreaded and gave examples of performance.

Two main things that I learned from the talk:
1> The hardware on a multi core system will try to multi thread the CPU's instructions. (ie, it will look for instruction that can be executed at the same time)

2>Newer IDE's can use multithreading and parallel programming right in your program (at compile time, I think it puts code that can be multithreaded together).

This is not the talk that I saw, but this video may help you understand more about how fast multi core systems can be.
Tech chat with Microsoft's parallel guru Steve Teixeira

[AdamSpeight2008]

The speed up is never equal to the number of cores, as a small percentage is communication & synchronisation.

See: Amdhal's Law

[JBrace1990]

Multiple core CPUs CAN act as one large CPU, or they can function individually.

For example, on my laptop I have dual cores, and I can limit whaich programs I want to run on which core, meaning I can always have one dedicated solely to photostop, or some other program, instead of everything.

Multiple cores function as a single core if, and only if, the program supports it. Newer applications usually (but not always) do. If it does not, then it only uses 1 core, meaning your dual core processor is just like a single core cpu.

[SpeedisaVirus]

Autocrat, on 26 April 2010 - 11:36 PM, said:

Why do we see a single CPU usage graph in task manager for single core CPUs, two graphs for dual core and four for quad cores?

Well because each CPU core can have different loads. The speedup, like was mentioned is never equal to a multiple of the number of cores.

There are many other factors at play that have been mentioned. Code has to be optimized to take advantage of multiple cores or the system smart enough to divvy the workload. Additionally there is either communication overhead or cache checking overheads associated with splitting computation up so even in an embarrassingly parallelization situation it's still not a perfect comptime/numcores result. Some things simply can't be parallelized. You can't predict the speedup of multiple cores without knowing about the source code. A good start on approximating real speedup would be to read about Gustafson-Barsis's and Amdahl's Laws.

Probably getting outside the scope of the question, oh well.

[Antiokus]

As other have pointed out, it's never clock x # of cores.

The graphs are related to each (logical) core. The graph shows a total amount of load for each core in the cpu. Hence 4 graphs for a quad core. Now, i7 CPUs with HyperThreading enabled will show double the number of graphs as cores. This is because HT makes each core act (for the sake of keeping it short) like 2 cores.

There is always overhead with multicore processors and since it's up to the OS (usually) to spread the load of applications across the cores you never get 100% extra processing from the cores. It's the same way with SLI and CrossFire graphics cards - you never get 100% return from each additional.

Clock speed is mostly an arbitrary number because of the differences in how different architectures handle instructions, memory controllers, and chip set differences. The i7 CPU gains a lot of performance from the triple channel memory controller and the high number of transactions per second (most around 5.8 GT/s) but maintains a much lower clock speed than old P4 chips that came stock at 3.4+ GHz. A lot of people advertise performance with synthetic and real-world benchmarks (SiSandra, 3DMark, SuperPi, etc). For the most part they give an accurate evaluation that can be used for comparing different components.

[PDUNZ]

Antiokus, on 20 May 2010 - 08:34 AM, said:

The graphs are related to each (logical) core.

Dual Core/Quad Core are physical cores, Hyperthreading are logical cores.

[Antiokus]

PDUNZ, on 23 May 2010 - 06:35 PM, said:

Antiokus, on 20 May 2010 - 08:34 AM, said:

The graphs are related to each (logical) core.

Dual Core/Quad Core are physical cores, Hyperthreading are logical cores.

Yes, I know the difference. I was talking about the OS not differentiating between the two and why with intel the number of graphs may be 2x the number of cores.

[PDUNZ]

Antiokus, on 25 May 2010 - 03:05 AM, said:

PDUNZ, on 23 May 2010 - 06:35 PM, said:

Antiokus, on 20 May 2010 - 08:34 AM, said:

The graphs are related to each (logical) core.

Dual Core/Quad Core are physical cores, Hyperthreading are logical cores.

Yes, I know the difference. I was talking about the OS not differentiating between the two and why with intel the number of graphs may be 2x the number of cores.

Oh sorry lol, You are correct. Ignore my post