How high of an IPC improvement is considered "good" when a new CPU generation releases?

[IAEInferno]

Hello, how high of an IPC improvement is considered "good" when a new CPU generation releases? Like for example:

 

i7-6700k --------------------->i7-7700k

 

At what exact IPC improvement would be considered "good" enough or is enough to be called a huge upgrade?

 

More than 15%? or?

[Godlygamer23]

This is subjective, and you will likely get different answers, but I'd say at least 20%.

[SpaceGhostC2C]

Is considered "good" by whom? The things with considerations is that they are subjective and context-specific. Would you change your CPU to enjoy an x% increase an IPC? Or perhaps the relevant question is: how much would you spend to improve an x%? Then, would you be positively or negatively surprised if the next intel release is x% above the previous one?

 

For example, if Intel would deliver 10-15% increases in IPC with Kaby Lake (unlikely), many people would call it a good improvement. But if Zen turned out to be 10-15% superior to Excavator in IPC, most people would say it failed and didn't deliver on the hype.

 

In the end, the objective reality is that processor A has characteristics X, and processor B has characteristics Y. Giving adjectives to A and B, or to the difference between A and B based on release dates, is a subjective matter.

[IAEInferno]

Just now, SpaceGhostC2C said:

Is considered "good" by whom? The things with considerations is that they are subjective and context-specific. Would you change your CPU to enjoy an x% increase an IPC? Or perhaps the relevant question is: how much would you spend to improve an x%? Then, would you be positively or negatively surprised if the next intel release is x% above the previous one?

 

For example, if Intel would deliver 10-15% increases in IPC with Kaby Lake (unlikely), many people would call it a good improvement. But if Zen turned out to be 10-15% superior to Excavator in IPC, most people would say it failed and didn't deliver on the hype.

 

In the end, the objective reality is that processor A has characteristics X, and processor B has characteristics Y. Giving adjectives to A and B, or to the difference between A and B based on release dates, is a subjective matter.

 

I see, so how much would be considered a good IPC improvement from excavator to zen in your own opinion?

[SpaceGhostC2C]

1 minute ago, IAEInferno said:

I see, so how much would be considered a good IPC improvement from excavator to zen in your own opinion?

AMD promised/aimed at a 40% increase, so I guess it makes sense to judge it on that basis.

[ARikozuM]

As much as possible without lowering clock speed (looking at you Intel).

[IAEInferno]

Just now, SpaceGhostC2C said:

AMD promised/aimed at a 40% increase, so I guess it makes sense to judge it on that basis.

Do you think it would be worth it to buy the Zen one instead of a 7700k?

 

I know they aren't released yet till next year, but what do you think is the best option since I'm going to upgrade my CPU, I will just be gaming.

[meenmeen1103]

For me: +X% IPC equal to or greater than previous gen freq Y, where X*Y=P and P/$=V is equal to or greater than previous gen V. That would be good improvement.

[SpaceGhostC2C]

Just now, IAEInferno said:

Do you think it would be worth it to buy the Zen one instead of a 7700k?

 

I know they aren't released yet till next year, but what do you think is the best option since I'm going to upgrade my CPU, I will just be gaming.

I tend to side with Linus and suggest i5 for just gaming, but i5s are the ones most likely to be influenced by Zen.

 

But, sticking to your question: AMD's own claims on Zen's performance so far imply they will get closer to Intel, but they probably won't have a CPU at 6700K or 7700K level. Because I consider a 7700K to be overspending for gaming, I would tend to say that is better to see what AMD has and at what prices, then go for the cost-effective solution. But if you are dead set on getting an i7-7700K, and although we can't tell for sure until both are out, chances there won't be a 7700K Zen equivalent, and that prices of 7700K are not as affected by a successful Zen release as other CPUs of Intel's lineup. So in that case, probably you'll end up with the 7700K one way or another, and at a similar price.

 

 

[SpaceGhostC2C]

5 minutes ago, meenmeen1103 said:

For me: +X% IPC equal to or greater than previous gen freq Y, where X*Y=P and P/$=V is equal to or lower than previous gen V. That would be good improvement.

If V=P/$, and V equal or lower than previous generation, are you saying you want lower performance per dollar compared to the previous generation?

[meenmeen1103]

17 minutes ago, SpaceGhostC2C said:

If V=P/$, and V equal or lower than previous generation, are you saying you want lower performance per dollar compared to the previous generation?

Shit, edited lol

[DELTAprime]

6 minutes ago, SpaceGhostC2C said:

I tend to side with Linus and suggest i5 for just gaming, but i5s are the ones most likely to be influenced by Zen.

After going with a 6600k this generation and seeing how many games are now multi-threaded and will take advantage of all threads together with the fact some poorly optimised games are maxing out threads or even whole CPUs at 100% I'm leaning more towards 4 cores 8 threads is the way to go.

 

Not sure if it's a bug, poor optimisation or the game really needs all that horse power, but I get 100% CPU utilisation on vanilla GTA V at max settings at 1440p. BF1 is too close for comfort to 100% utilisation. FH3 is poorly optimised but people on 6700k's and 4790k's are having less issues.

 

There are just enough examples like this that make me recommend going for 6700k or the upcoming 7700k.

[SpaceGhostC2C]

3 minutes ago, DELTAprime said:

After going with a 6600k this generation and seeing how many games are now multi-threaded and will take advantage of all threads together with the fact some poorly optimised games are maxing out threads or even whole CPUs at 100% I'm leaning more towards 4 cores 8 threads is the way to go.

If games make use of many threads, you need more cores, not HT (CPUs don't have threads, they have cores. Programs have threads. When manufacturers report "4C/8T" they just mean that their 4 cores can accept 8 threads at once - and with some luck, do at least some work on them simultaneously, but not necessarily).

 

Having said that, I'm sure you are going to obtain better performance from a 6700K than from a 6700K. You are also going to obtain better performance from a Titan XP than a GTX 1080. So? You can always go higher end and obtain better performance, but at ~60% higher price I can't recommend a 6700K purely for gaming, because even performing better it's terrible value.

[Space Reptile]

id say anything above 1% , so kaby lake failed already

 

but no to be serious 10%+ atleast

[Glenwing]

There is no answer. IPC change by itself is neither good nor bad, the only thing that matters is the performance, which is the combination of frequency and IPC.

If you design a CPU that only has half the IPC of last generation, but due to your innovative design achieves 10x higher frequencies with the same power and cooling, your performance has just increased 5x. That is a very good design improvement. Does it matter if IPC decreased? Not really, if that design allows you to achieve much higher frequencies (somehow).

 

Likewise if you design a heavy core that has a much higher IPC, but due to its complexity is forced to operate at much lower frequencies, then the design may not be much of an improvement over the last generation design.

 

So how much performance increase is considered decent? Ehh... I'd say 20%, at the same price point / die size.

[porina]

IPC is often a single number representing lots of different things. You might see improvements in some areas and not others. Thus, how much difference you see will depend on what you use it for.

 

For example, a lot of areas have been inside CPUs for a long time, and they're going to be pretty well optimised so don't expect much more to come from them in future unless there's a radical idea for improvement. Offsetting that, there are often new instructions to do a particular thing really well, but is of little or no benefit to others.

 

Some might say there isn't much difference from say, Sandy Bridge to Skylake, but in my interest area which is floating point heavy, I've seen some big improvements. SB to Haswell saw a 50% improvement from added AVX2 and FMA instructions. Haswell to Skylake saw 14% increase. While nothing new was added, Intel did trim a clock cycle out of many of those instructions which likely contribute to the improvement. And this is before you look at ram interface changes.

[Matias_Chambers]

I would say 10% at least.