Ipc question

[evilmeister13]

Does manufacturing process directly determine ipc? For instance, will a 4th gen i7 have a much greater ipc than say, a 2nd gen? And therefore perform much better with the same core count and clock speed? And if thats the case, would a newer amd apu perform better in emulation than a chip in the fx series? I know it'll still suck compared to intel, but it would still be better than older chips right?

[qwertywarrior]

http://www.tomshardware.com/answers/id-1802534/ipc-intel-core-3330-ghz.html#11543027

[Guest]

IPC is always tricky because it is just an average and highly depends on the kind of calculation that is being done. Floating point calculation will take way longer than simple addition, for example. I don't actually know if there is a standardized way to determine the IPC to be honest. With that out of the way, yes it is usually the idea behind a new CPU generation to raise the IPC compared to the previous one. That's why Ivy Bridge was a bit faster than Sandy Bridge on the same clock speeds and that's why haswell was a bit faster than Ivy Bridge as well. With AMD and their APUs it probably gets more complicated since they're focusing on different things, I don't know if raising the IPC is their number one concern right now.

[TheRunningOtaku]

Does manufacturing process directly determine ipc? For instance, will a 4th gen i7 have a much greater ipc than say, a 2nd gen? And therefore perform much better with the same core count and clock speed? And if thats the case, would a newer amd apu perform better in emulation than a chip in the fx series? I know it'll still suck compared to intel, but it would still be better than older chips right?

I'm not sure what you're use case is entirely, but I think you should try to search for benchmarks, reviews or comments made by people with the same software as you on the performance of the chip, that way you can have a much better idea of the real-world performance of the chip.

[harrynowl]

Well manufacturing process shrinks allow more transistors to be added without increasing the silicone die size, and generally speaking that extra complexity could mean an increase in IPC.

 

Intel do a tick-tock thing, where they have a tick as a new manufacturing process (e.g. Westmere core (32nm)) then a tock would be a new micro architecture such as the Sandybridge core (also 32nm). However Ivy was a little different as we got both a ~10% IPC improvement and a die shrink.

 

On AMDs side things are a little different, right now it seems to be a case of 1 step forward 2 steps back, they have K10 which uses a true core design, then they moved to CMT (clustered multi-threading) to try get more cores in the same die space without shrinking the process. Unfortunately this kind of failed as the FX8150 ended up being slower than the old K10 1100T. Piledriver added a ~10% IPC bump but mainly the switch to hard edge flip-flops allowed them to get better power consumption- which they then invested into more clock speed.

 

On the APU side the 1st generation Llano APUs which were K10 derived, then Trinity/Richland were based off the Piledriver core- so no real improvement in CPU power between two generations. Now Kaveri is based on Steamroller which has a 15-20% IPC improvement over Piledriver which is nice, but the manufacturing process used (28nm SHP) doesn't clock as high as the old silicone-on-insulator they've used since forever. So CPU performance is roughly the same.