[SOLVED] How much does IPC actually impact performance?

[ethan206]

According to recent rumors, Ryzen 4000/Zen 3 will bring a 15-17% IPC uplift from Ryzen 3000/Zen 2 which already brought a 15% IPC increase from Zen+/Zen and apparently clock speeds are staying the same (remember, this is just a rumor). And Intel's latest Ice Lake chips had an 18% IPC increase over Sky Lake, but their clocks actually went down and it appears that some of their 14nm+++ CPUs are beating their new 10nm Ice Lake parts. I just wanted to know how much instructions-per-clock actually impacts performance. I'm not that much of a techy person, but would higher clock speeds or better instructions-per-clock bring better performance to a CPU? Also, I found it weird how Intel's new Ice Lake CPUs dropped down in clock speeds. Is this just because they've spent years stuck and perfecting 14nm or is just moving to a smaller node/newer architecture?

 

[prophet51]

Clock for clock intel still wins at gaming, even though it has worse ipc. IPC isn't everything but it greatly helps.

 

[Karadjgne]

3rd gen Ryzen and 9th gen Intel are roughly equitable IPC. The only real underlying reason Intel gets higher fps is its ability to bounce around the 5GHz area while Amd is bouncing at 4GHz. But thats only in single thread performance, which is still somewhat valid to multiple thread performance upto the point of maximum thread availability. So it's entirely possible for applications that are strongly written for 9 or more thread usage to see highly favorable results from Ryzen, beating anything upto a 9700k.

Winzip proved this years ago, the R7-1700x getting double the amount of compression done vrs an i7-8700k. Yet other apps like Adobe CC which do not scale over 8 threads, strongly favored Intel single core performance, even the I5-9600k handily beating a 2700x with ease.

Basically it all goes hand in hand, the software, the clock speeds, the IPC. In some things the weight is one way, in others the other way. All of it is important, higher/faster = better. An 8 thread FX 8350 at 5GHz is no comparison in any way vrs an 8 thread i7-9700k at 5GHz, just as an i5-9600k can't keep up with an R9 3950x.

 

[TerryLaze]

3rd gen Ryzen and 9th gen Intel are roughly equitable IPC. The only real underlying reason Intel gets higher fps is its ability to bounce around the 5GHz area while Amd is bouncing at 4GHz.
But thats only in single thread performance, which is still somewhat valid to multiple thread performance upto the point of maximum thread availability.

No that's not it,intel is still faster even at the same clocks, the 25% clock difference is mostly being killed by today's weak GPUs but there is still more than enough reason to go with intel to avoid lag spikes.
Youtube link

 

[TerryLaze]

Also, I found it weird how Intel's new Ice Lake CPUs dropped down in clock speeds. Is this just because they've spent years stuck and perfecting 14nm or is just moving to a smaller node/newer architecture?

Ice lake drops because the previous CPUs hit above 5Ghz easily...if it only got to 4Ghz like ryzen then ice lake would also be the same if not higher clocks.

 

[mdd1963]

BF5 is indeed an important benchmark! (to me anyway...) Certainly more important than Cinebench...(to me, anyway)

Based on the results thus far, it seems doubtful that anything near the top in BF5 (indeed, average FPS is less important than the minimum FPS) is likely to ever be 'slow' in most other tasks, be it manipulating Office 2019 docs, surfing, etc...

 

[ethan206]

3rd gen Ryzen and 9th gen Intel are roughly equitable IPC. The only real underlying reason Intel gets higher fps is its ability to bounce around the 5GHz area while Amd is bouncing at 4GHz. But thats only in single thread performance, which is still somewhat valid to multiple thread performance upto the point of maximum thread availability. So it's entirely possible for applications that are strongly written for 9 or more thread usage to see highly favorable results from Ryzen, beating anything upto a 9700k.

Winzip proved this years ago, the R7-1700x getting double the amount of compression done vrs an i7-8700k. Yet other apps like Adobe CC which do not scale over 8 threads, strongly favored Intel single core performance, even the I5-9600k handily beating a 2700x with ease.

Basically it all goes hand in hand, the software, the clock speeds, the IPC. In some things the weight is one way, in others the other way. All of it is important, higher/faster = better. An 8 thread FX 8350 at 5GHz is no comparison in any way vrs an 8 thread i7-9700k at 5GHz, just as an i5-9600k can't keep up with an R9 3950x.

towards the end of your para when you compared the FX 8350 to an i7-9700K, you were saying that despite both CPUs being clocked at 5GHz, then Intel part wins because of it's better IPC and smaller node right? And when you compared the i5 to the R9, the R9 has double the cores but IPC+clocks plays a huge role if those cores will actually run fast?

 

[ethan206]

Ice lake drops because the previous CPUs hit above 5Ghz easily...if it only got to 4Ghz like ryzen then ice lake would also be the same if not higher clocks.

Ah I see, will Intel be planning multiple stages of 10nm (like they did with 14nm, 14nm+, 14nm++) to better improve their 10nm node or will they revert
back to the Tick-Tock model where one year it's a die shrink and then the next an architecture redesign?

 

[ethan206]

No that's not it,intel is still faster even at the same clocks, the 25% clock difference is mostly being killed by today's weak GPUs but there is still more than enough reason to go with intel to avoid lag spikes.
Youtube link

Will next year AMD catch up with Intel? According to recent leaks, Intel is sticking to 14nm again on desktop while AMD
will be moving in to 7nm+ EUV which will bring a 15-17% IPC uplift, so will AMD catch up with Intel in terms of single-threaded performance?

 

[ethan206]

BF5 is indeed an important benchmark! (to me anyway...) Certainly more important than Cinebench...(to me, anyway)

Based on the results thus far, it seems doubtful that anything near the top in BF5 (indeed, average FPS is less important than the minimum FPS) is likely to ever be 'slow' in most other tasks, be it manipulating Office 2019 docs, surfing, etc...

In games does higher IPC matter or does higher clock speeds matter? Higher IPC would benefit single-threaded performance more right?

 

[InvalidError]

In games does higher IPC matter or does higher clock speeds matter? Higher IPC would benefit single-threaded performance more right?

IPC vs clocks is a ridiculous argument to try to make. Throughput = IPC * clock, you want an OPTIMAL BALANCE of both.

The reason why one architecture gets better performance in one set of benchmarks while it somehow gets beat in others is because IPC depends on the instruction mix and algorithm structures which vary greatly from one piece of software to another. CPU architectures like some instruction mixes and programming styles better than others. Combine the broad spectrum of programming styles and instruction mix with the myriad of design considerations that go into putting a CPU together, you've got innumerable possible outcomes.

If you sacrifice IPC for higher clocks, you get Netbust and Faildozer. Lots of heat, crap performance. It does not work.

 

[TerryLaze]

Will next year AMD catch up with Intel? According to recent leaks, Intel is sticking to 14nm again on desktop while AMD
will be moving in to 7nm+ EUV which will bring a 15-17% IPC uplift, so will AMD catch up with Intel in terms of single-threaded performance?

Zen+ already has more IPC than coffee lake, as you can see from the video at 4Ghz it only helps with 3d render and zip it does not help with games because as invaliderror already said different code uses different parts of a CPU,so even if AMD gets an 17% uplift it will only help with software that can use the additional IPC like 3d render and zip.
We have seen the same thing on intel's 10nm laptop CPUs,they do have more IPC as well and they are also only faster in cinebench and zip.

 

[dorsai]

As InvalidError said above...clockspeed and IPC is a balancing act along with the instruction sets and software architecture of any given program.

Ryzen simply blows Intel away in most applications today because they've better optimized the balance between IPC and clockspeed as well as proving more cores at a given price...Intel still leads verly slightly in gaming with a 2080ti due primarily to the slow development cycle in gaming. Games tend to re-use game engines across multiple generations of games (im looking at you Total War) so the optimization process for those game engines often takes years and simply has not caught up to Ryzens architecture as they remain primarily optimized for quad core Intel.

 

[dorsai]

No that's not it,intel is still faster even at the same clocks, the 25% clock difference is mostly being killed by today's weak GPUs but there is still more than enough reason to go with intel to avoid lag spikes.
Youtube link

The % difference between 156 and 160 is statistically insignificant and falls within the margin of error in game testing...as Karadjgne stated above AMD and Intel are essentially equal in gaming while AMD has a rather solid lead in everything else.

 

[drea.drechsler]

Something that I consider when looking at IPC vs. clockspeed is that modern CPU's increasingly don't hold a constant clockspeed. That brings in other factors to gauge CPU performance, e.g., thermal performance and the boosting algorithms' ability to exploit each processor's performance potential in varying useage scenarios.

For instance, Ryzens hold higher intermediate clock speeds in heavy all-core workloads than do the Intel counterparts, which is a major contributor to their superior performance in that scenario, while also allowing extremely high boost clocks in single core workloads where there's more thermal headroom.

It's no coincidence that Intel prefers Cinebench 15 which runs shorter to showcase crazy-high all core boosts, but can only be held a few seconds before being dialed way back. While AMD prefers Cinebench 20, which runs longer to showcase moderately high intermediate boosts across all cores that are allowed to vary up and down slightly as thermals allow throughout the BM's run, limited only by thermal performance.

IMO, it's not just IPC vs clockspeed, it's IPC vs. clockspeed vs. algorithm.

The % difference between 156 and 160 is statistically insignificant and falls within the margin of error in game testing...as Karadjgne stated above AMD and Intel are essentially equal in gaming while AMD has a rather solid lead in everything else.

Looking at games to compare performance is one thing but to draw conclusions about 'superior architecture' seems pointless to me. That's because so many games are developed with a bias to one processor or another. Sometimes with the assistance of Intel or AMD, sometimes simply because of switches used in compiling game code by the developers. Clearly Intel has been heavily involved in assisting game developers optimize for their processors. But even if not with Intel's help, when a developer chooses some switches that will provide better performance on Intel you can't fault them since, well, there's more Intel users in their customer base and it costs money and time to debug or optimize an alternate code path for AMD processors. So the upshot is: when performance is even close, how can you know the difference has anything to do with the actual processor architecture, whether it be IPC or clockspeed, or simply that one's being gimped a little bit?

 

[TerryLaze]

The % difference between 156 and 160 is statistically insignificant and falls within the margin of error in game testing...as Karadjgne stated above AMD and Intel are essentially equal in gaming while AMD has a rather solid lead in everything else.

The difference between 103 and 112 is not statistically insignificant though, it's 10% and maybe you missed the little detail that intel has another 25% clock potential on top of that.
If you want good minimums with the smallest possibility of dips intel offers you about 35% better chances for that.

AMD has a rather solid lead in everything else.

AMD has a rather solid lead in exactly two things, 3d rendering and compression but only at the same clocks,for a mainstream user both of those are pretty useless,for HEDT/power users it's more relevant.

 

[TerryLaze]

error

 

[RodroX]

According to recent rumors, Ryzen 4000/Zen 3 will bring a 15-17% IPC uplift from Ryzen 3000/Zen 2 which already brought a 15% IPC increase from Zen+/Zen and apparently clock speeds are staying the same (remember, this is just a rumor). And Intel's latest Ice Lake chips had an 18% IPC increase over Sky Lake, but their clocks actually went down and it appears that some of their 14nm+++ CPUs are beating their new 10nm Ice Lake parts. I just wanted to know how much instructions-per-clock actually impacts performance. I'm not that much of a techy person, but would higher clock speeds or better instructions-per-clock bring better performance to a CPU? Also, I found it weird how Intel's new Ice Lake CPUs dropped down in clock speeds. Is this just because they've spent years stuck and perfecting 14nm or is just moving to a smaller node/newer architecture?

Other than what was already said about how Intel and AMD hold diferent grounds and are almost equal on gaming, at this point of time intel may hold the crown as the "best gaming" cpu, but the diference between comparable models is in most case really around 3%, and thats as a top, and thats when Intel wins, which is not all the time.

I think, since july 2019, buyers should forget about "somehow useless" numbers like ipc and frecuency, and focus on more important questions:

What do I need the PC for/What will I use it for?
What level of performance do I want?
How much budget do I have?
Is there a component that I need to prioritize against others?

Those are, for me, the most important things someone need to ask before making a purchase. Once you have those 4 question very clear, then you can ask other stuff.

As for Ryzen 4xxx, no one can be sure of how much that "%" gain will compare with oldest/contemporary parts until expert reviewers like TomsHardware, LTT, HardwareUnboxed, Guru3D, Gamer Nexus, etc. get thier hands o the real product and test it.

Cheers

 

[RodroX]

The difference between 103 and 112 is not statistically insignificant though, it's 10% and maybe you missed the little detail that intel has another 25% clock potential on top of that.
If you want good minimums with the smallest possibility of dips intel offers you about 35% better chances for that.

Actually no, you are forgetting that The Ryzen 9 3900X has another 4 cores and another extra 4 SMT threads to hold thier ground.
And the .1% low 103 vs 112 FPS does not scale or compares the same way as AVG FPS do (sorry for my english if I can not express the best way). At this point even if the "math" between those two numbers tells you "hey is a 10% diference!", yeah, of course it is, but can you really feel it, will you have a worst performance playing the game?, My honest guess is no, you wont feel the diference.

Sorry for double posting.

PD, and no again, actually right now Intel only have very few cpu that can hold the 1% low crwon against AMD, those are the Core i7 8700K, 9700K, and i9 9900K/KS, in most games and in most case, even low budget Ryzen CPU will show better 1% low numbers than more expensive intel parts.

 

[ethan206]

so this thread sort of went all over the place, but the conclusion I'm drawing from all of this is that there are multiple factors to a CPU's performance. Cores, clock speeds, process node, TDP, IPC, architecture and cooling all contribute, but from what I'm gathering, IPC does matter, but is not the most important gain to a CPU's performance? Because with Zen 2 advancements, AMD and Intel are roughly on par with instructions per clock, but the reason why AMD outperforms Intel in multi-threaded workloads is because of higher cores and SMT, while Intel outperforms (though by slightly) AMD in single-threaded workloads/gaming because of their higher clocks? But if AMD keeps on driving their IPC higher (with Zen 3, Zen 4), they could beat Intel in single-threaded performance despite having lower clocks right? And if Intel continues to stay on 14nm, their only choice to increase performance would be to drive up TDP right (because they've been refining 14nm for like 4-5 years now)? And also, happy new year everyone

 

[RodroX]

so this thread sort of went all over the place, but the conclusion I'm drawing from all of this is that there are multiple factors to a CPU's performance. Cores, clock speeds, process node, TDP, IPC, architecture and cooling all contribute, but from what I'm gathering, IPC does matter, but is not the most important gain to a CPU's performance? Because with Zen 2 advancements, AMD and Intel are roughly on par with instructions per clock, but the reason why AMD outperforms Intel in multi-threaded workloads is because of higher cores and SMT, while Intel outperforms (though by slightly) AMD in single-threaded workloads/gaming because of their higher clocks? But if AMD keeps on driving their IPC higher (with Zen 3, Zen 4), they could beat Intel in single-threaded performance despite having lower clocks right? And if Intel continues to stay on 14nm, their only choice to increase performance would be to drive up TDP right (because they've been refining 14nm for like 4-5 years now)? And also, happy new year everyone

Basically yes.
In short if you are a gamer and need a new Motherboard and CPU, you have some very nice and competitive choices that weren't here two years ago. No matter what your budget is, any of the Ryzen 5, 7, 9 (first, second or third gen) or to the 8th gen i7 (the one with 6 cores / 12 threads, if you can find one) to the latest gen Core i7/i9 9xxx will do the trick. Sadly the Core i5 is out of the arena for now, not worth to get one unless you already own a compatible motherboard and you are upgrading from a Pentium or Core i3.
If you're a basic office user, then any decent cpu will do the trick.
If you are a power/producer user, depending on the software there could be a chance you may still want and be better with an intel cpu, but in most cases AMD zen 2 cpus have the most competitive products (Ryzen or Threadripper).

And of course, happy new year to you and everyone else!