CPU Speeds (IT Coursework)

[majorgibly]

Hello Tom's Hardware,

Been a long time since I posted onto the site asking and answering questions. Exams are over yay But I still have to do ICT assignments - Now I have to write a report and it asks to do a report on each component in the system easy right? But is what I have done is confused myself, I go on about speeds not just being dependent of CPU Clock or amount of cores. So what does indicate speed? CPU Architecture so the size of the die? cycles per second?

I knew this at one point but I have confused myself somehow! I'm a little confused if somebody could clear up with me what indicates CPU speed then that would be really helpful :0)

Thanks!

 

[$hawn]

For each single core, ignoring memory bandwidth and cache etc, u can roughly state that , CPU speed= Clock speed x IPC.

A good architecture like IB will have a high IPC, along with good clock speeds to give outstanding performance.

On the other hand, u can ramp up the clock as high as sanely possible, but with a weak IPC, u are gonna suffer, for examples, the Pentium4 Netburst and the AMD Bulldozer architectures.

Now, if ur software is multi-threaded, adding more cores will increase performance, but the increase is seldom linear. Same goes with cache sizes, bigger cache sizes will produce benefits only upto a certain extent; after that more cache is simply useless, or worse, it may actually affect performance due to longer latencies!!

Die size doesn't affect performance in any direct way.

 

[majorgibly]

For each single core, ignoring memory bandwidth and cache etc, u can roughly state that , CPU speed= Clock speed x IPC.

A good architecture like IB will have a high IPC, along with good clock speeds to give outstanding performance.

On the other hand, u can ramp up the clock as high as sanely possible, but with a weak IPC, u are gonna suffer, for examples, the Pentium4 Netburst and the AMD Bulldozer architectures.

Now, if ur software is multi-threaded, adding more cores will increase performance, but the increase is seldom linear. Same goes with cache sizes, bigger cache sizes will produce benefits only upto a certain extent; after that more cache is simply useless, or worse, it may actually affect performance due to longer latencies!!

Die size doesn't affect performance in any direct way.

Thanks for clearing up some of those issues I had! What does IPC stand for?

Thanks dude means a lot!

 

[$hawn]

Instructions per clock....in layman terms how many instructions a processor can crunch per clock frequency.

Example, an IB core can probably do twice as much as work as a Pentium4 core, if both are clocked at the same freq.

Happy to help cheers

 

[majorgibly]

Instructions per clock....in layman terms how many instructions a processor can crunch per clock frequency.

Example, an IB core can probably do twice as much as work as a Pentium4 core, if both are clocked at the same freq.

Happy to help cheers

Thanks fella

 

[$hawn]

u noe, i wudn't mind a Best Answer badge

 

[majorgibly]

u noe, i wudn't mind a Best Answer badge

Now you are pushing my limits of knowledge on a forum lol.

 

[majorgibly]

u noe, i wudn't mind a Best Answer badge

Okay how do I do it lol

 

[$hawn]

Now you are pushing my limits of knowledge on a forum lol.

lol, there shud be a 'Select as best answer' option somewhere just below my posts

 

[majorgibly]

lol, there shud be a 'Select as best answer' option somewhere just below my posts

Nope I think there is a period of time before the option appears otherwise the thread would close to quickly. I will bookmark the page and check again in the next few days if not I will mail an admin!

:

 

[gamerk316]

The above is basically correct. For a single core, CPU Speed can be measured by a simple formula:

Speed = (Clock Frequency * Instructions Per Clock Cycle)

When you have a multicore chip, however, the math gets a bit more difficult. In theory, a 4 Core CPU should have, as a best case, 4x the performance of a single core CPU. But shared resources on the CPU die will likely yield less then this. Still, you can make a generic formula for a multi-core chip:

Speed = (Clock Frequency * Instructions Per Clock Cycle) * Number of Cores

Throw in logical CPU cores, and the equation again gets complicated, since logical cores are slower then physical cores, so you need to throw in a scaling factor. Same logic applies though: Performance = Frequency * IPC * Number of Cores * Scaling Factor (if applicable)

 

[popatim]

You guys are confusing speed with processing capability/performance. A 3ghz I7 does not have a speed of 6ghz because it can process 2 instructions per cycle...

The speed of a cpu is how fast its clock is; which is usually a multiple of the front side bus. In the case of the new intel chips where the fsb is not really around anymore we have a baseclock (BCLK) which controls your memory speed, qpi speed, and core speed (CPU speed=bclk*multiplier)

We havent even touched on Turbo boost which increases clock speed depending on core usage and workload, nor power saving features which reduce the processors speed and may shut off cores.

 

[viridiancrystal]

The above is basically correct. For a single core, CPU Speed can be measured by a simple formula:

Speed = (Clock Frequency * Instructions Per Clock Cycle)

When you have a multicore chip, however, the math gets a bit more difficult. In theory, a 4 Core CPU should have, as a best case, 4x the performance of a single core CPU. But shared resources on the CPU die will likely yield less then this. Still, you can make a generic formula for a multi-core chip:

Speed = (Clock Frequency * Instructions Per Clock Cycle) * Number of Cores

Throw in logical CPU cores, and the equation again gets complicated, since logical cores are slower then physical cores, so you need to throw in a scaling factor. Same logic applies though: Performance = Frequency * IPC * Number of Cores * Scaling Factor (if applicable)

Also note that # of cores only applies if there are multiple cores being used. Many applications still only use one core. Some can use 16, others only 2.

 

[DJDeCiBeL]

@ OP This thread was opened as a discussion, so you can't select a Best Answer (ever).

 

[majorgibly]

Okay thanks for clearing it up. I talked about clock rates and number of cores. I will go more in-depth and do some more research with the stuff you have all suggested.

Thanks a ton guys means a lot!