Terminology

[Ruger22C]

I had only heard it used as "Clock-to-clock ratio," what is the proper term for the difference in CPU's running at the same MHZ, but with different performance? I.E. as I recall a wolfdale beating the i7 by about 5% in games, but losing by 30-80% in other tasks, encoding decoding etc... yet, at the same MHZ.

Also, a chart comparing the Phenom 2 to the i7 would be appreciated.

 

[Cogman]

The term? Somewhat meaningless. I've only ever heard something along the lines of clock for clock. The reason it is meaningless is that some chips are capable of higher clock speeds then others. You could have a stellar performer at a low clock speed and one that outperforms it at a higher clock speed, but both are in the same price range.

A better ratio would be price/performance, yet performance is extremely elusive.

 

[Ruger22C]

Cogman, some operate either at stock (if you are scared to overclock) of the same GHZ, and some OC to about the same GHZ. Therefore, it does make a significant difference to me - ESPECIALLY considering the "vast" difference between wolfdale and i7. It was very extreme, depending on task at hand.

Any way, you heard basically the same term, then. Any idea where I can find that? I had it at one time, before the phenom 2 was tested iirc.

 

[Idontcare]

The term you are looking for is IPC and it stand for instructions per clock.

Comparing IPC between processors is helpful as a performance reference point and is useful in highlighting microarchitectural differences between processors.

 

[Ruger22C]

This is an example of what I meant, even though the phenom is at 3.8, the i7 is beating it.

Idontcare.. Does IPC take into account specific tasks, i.e. encoding vs games?

Thanks.

 

[Idontcare]

fixed link to image: http://media.bestofmicro.com/D.../original/image015.png

Originally posted by: Ruger22C
Idontcare.. Does IPC take into account specific tasks, i.e. encoding vs games?

Thanks.

Well technically it goes a level deeper than that...the keyword in the acronym IPC is instruction...and there are hundreds of instructions in the ISA of a modern x86 compatible processor.

Each instruction has its own execution speed, called instruction set latency which can be easily measured with the EVEREST Ultimate Edition for using its Instruction Latency routine.

So getting back to your question, first we have to understand that benchmarking involves the use of many many differing instructions, so we don't get a true "instruction per clock" result, but rather a fuzzy average of the aggregate performance of all those various instructions that were executed during the benching.

Now to the question of encoding vs. games. These applications use different sets of instructions, so an IPC in encoding will be different than an IPC in gaming even though the clockspeed is the same because the instructions used by the respective apps have different intrinsic instruction set latency.

 

[Ruger22C]

Appreciated.

 

[heyheybooboo]

Your linked chart has less to do with IPC and more to do with SIMD (I think). iTunes is a type of convoluted metric designed around the PowerPC arch and rebuilt for the transition to Intel.

AMD for the most part has had their arse handed to them in SIMD from 3DNow to SSE5 (which was recently dropped in favor of 'AVX' - the next generation instruction set for Intel's 32nm Sandy Bridge processors).

AMD will try to duplicate the functionality of SSE5 in a combined AVX they call AVX/FMA.

I'm thinking of all the bones AMD wants to pick with Intel these bones would most likely be the ones that really get panties bunched in Sunnyvale.

Here is more information than you will ever want to know about how K10 handles instructions with a comparison to Yorkfield. The Intel 'execution engine' simply handles the range of instructions better than AMD.

This may help you in forming questions to ask about i7.

 

[Cogman]

Originally posted by: Ruger22C
Cogman, some operate either at stock (if you are scared to overclock) of the same GHZ, and some OC to about the same GHZ. Therefore, it does make a significant difference to me - ESPECIALLY considering the "vast" difference between wolfdale and i7. It was very extreme, depending on task at hand.

Any way, you heard basically the same term, then. Any idea where I can find that? I had it at one time, before the phenom 2 was tested iirc.

My point was more relivent to the stock situation. If an i7 at 2.4 GHz is equivalent in performance to a Phenom at 3.0Ghz. But they are the exact same price, whats the point of comparing clock/clock performance? (it is somewhat funny, because essentially AMD and Intel have suffered a role reversal. Intel used to be the company with the high clocks low IPC and AMD the low clocks high IPC).

Like IDC said, IPC is hard to measure, because the question becomes "what instructions do you measure for IPC and how much weight do you give them?" Some instructions can completely remove the need for several other instructions, some instructions, while they replace several instructions, are slower then the several instructions. Some instructions are never used even though they exist. (3dNOW comes to mind, though, didn't it become a part of MMX?)

Hence my statement, for me, Price/performance is a better indicator of a processors value then Price per clock. For some, Performance/Clock is useful because at times you can get a low clocked CPU and bump it up to your desired speed.