Why IPC is not measured?

[JuztBe]

IPC has became such a big thing, that even clock rate is some cases are not a factor when comparing CPUs. While clock rate is listed as one of the main specs, there are not a single word about IPC in it.
Why are we not measuring it then?  Instructions per clock sounds like a measurement.   

[Bob Jim]

I want to know this too, followed.

[arvark]

Because you can't say "x% ipc" and no one would understand "SEE2 SKODSKL3WJKLJ12 FFU2 FML45 SSE" and such names when it comes to performance!

[Enderman]

there are way more factors than just ipc

read about them here:

https://en.wikipedia.org/wiki/Instructions_per_cycle

in the "computer speed" section

 

the best way is to actually use the CPUs in the real world and look at benchmarks for the program/game you want to know

[juretrn]

It's likely also the kind of thing that is not meant for the public and their competitors.

[Sakkura]

IPC is task-dependent, unlike clock frequency. So there's no standard to measure by. Also, vendors are more interested in just marketing based on clocks.

[JuztBe]

Because you can't say "x% ipc" and no one would understand "SEE2 SKODSKL3WJKLJ12 FFU2 FML45 SSE" and such names when it comes to performance!

But a lot of people are saying x% IPC improvement. "AMD announces ZEN, 40% IPC improvement"

 

[AresKrieger]

IPC has became such a big thing, that even clock rate is some cases are not a factor when comparing CPUs. While clock rate is listed as one of the main specs, there are not a single word about IPC in it.

Why are we not measuring it then?  Instructions per clock sounds like a measurement.   

 

Isn't that just part of MFLOPS calculation Millions of Floating-Point Operations Per Second

 

[App4that]

But a lot of people are saying x% IPC improvement. "AMD announces ZEN, 40% IPC improvement"

. You can buy "ionized water" too, it sounds cool and sells things to use words few people understand.

[JuztBe]

IPC is task-dependent, unlike clock frequency. So there's no standard to measure by. Also, vendors are more interested in just marketing based on clocks.

If I understood correctly IPC is not static, that's why you can't measure it?

_______________________________

No clear answer yet. More thoughts are always helpful. 

[typographie]

But a lot of people are saying x% IPC improvement. "AMD announces ZEN, 40% IPC improvement"

 

 

It might be more precise to say up to X%.

 

A percentage can be used to boil things down into a more easily understood form for the sake of comparison. It's not the same as actually quantifying IPC, though.

[i_build_nanosuits]

If I understood correctly IPC is not static, that's why you can't measure it?

_______________________________

No clear answer yet. More thoughts are always helpful. 

yes, the amount of instructions the CPU can process within each clock cycle at any given time will vary a lot depending on the nature of the task and it's requirements. That's why it's not possible to come up with a defenitive number or stat for this.

To compare CPU's unfortunately you have to rely on benchmarks, a wide variety of benchmarks results will give you an accurate representation of a CPU performance.

[Glenwing]

IPC isn't a specific thing and thus can't be measured. It's just a vague term.

 

It's a bit like asking why we don't have a neat table of values for the "strength" of materials. Surely granite is stronger than paper, because paper can be easily ripped apart while granite can't. But you can crack granite with a hammer, but no matter how hard you hit paper it won't shatter into pieces. So surely paper is stronger than granite. Similarly Kevlar is surely stronger than 2mm of glass, since Kevlar can stop bullets; but Kevlar is more flexible so can be easily bent while glass can't. So surely glass is stronger.

 

In reality, there is no such thing as a "strength" of a material. That's just a general term, it doesn't have any actual measurable value itself. Instead we have tensile strength, shear strength, hardness, stiffness, etc. And depending on what exact situation the material will be used for and what types of stresses the material will encounter, they will perform differently based on how strong it is for that type of stress.

 

The same goes for processor IPC. There is no "IPC" number. Instead you have floating point performance (FLOPs) at various levels of precision, integer performance, and hundreds of other instruction types, all of which have their own completely independent "IPC" number. How it will fare in an application depends on what types of instructions that application uses and how good the CPU is at those specific instructions.

[Nena Trinity]

But a lot of people are saying x% IPC improvement. "AMD announces ZEN, 40% IPC improvement"

 

When AMD says DX12 will boost FPS with 50% I expect 25%...

 

When AMD says Zen will have 40% IPC improvement I expect 20%...