a look at (ry)Zen's IPC - from Excavator to KabyLake

[Imglidinhere]

Just now, zMeul said:

mate, the title is IPC and you expect gaming benchmarks?! 

 

 

no it's not!

when you benchmark CPUs you don't benchmark games 

Why not? That's the best way to find out how bad of a bottleneck there actually is.

 

Just play a suite of games at 720p with a 1080 and run the same games at the same clock speeds with the same number of cores and threads. This guy can obviously disable cores and threads in these tests. It's a surefire way to eliminate ANY other factors... and it's highlighted a processor bottleneck really really rapidly in the past too.

[zMeul]

6 minutes ago, djdwosk97 said:

That's exactly what that graph implies and, I'm sure why you chose to use it -- because it exacerbates the lead of Haswell/Kaby Lake over Zen. 

 

That graph implies that Kaby Lake's IPC is 17% faster than Haswell. That's not exactly the case. In terms of raw IPC it's nowhere near that much of an improvement. 

no, NO

that's not what the graph implies! please the the title! it sais relative to Excavator - Excavator is the baseline Zen, Haswell. KabyLake are compared against

 

 

[Bouzoo]

1 minute ago, zMeul said:

mate, what are we discussing?! what is way off?!

Nvm me

[Imglidinhere]

Just now, zMeul said:

no, NO

that's not what the graph implies! please the the title! it sais relative to Excavator - Excavator is the baseline Zen, Haswell. KabyLake is compared against

 

 

What he's saying is essentially the same thing I was trying to convey, while it's not what the graph is openly trying to show, it DOES convey that Kaby Lake has a 17% lead over Haswell in these benchmarks, if you look at it subjectively, in an overall average here. Essentially an unintentional revealing of otherwise unknown benefits from Haswell to Kaby Lake.

[79wjd]

2 minutes ago, zMeul said:

no, NO

that's not what the graph implies! please the the title! it sais relative to Excavator - Excavator is the baseline Zen, Haswell. KabyLake is compared against

 

 

Yes, and if you convert it to using Haswell as the baseline instead of Excavator, then it indicates a 17% increase from Haswell to Kaby Lake.

[zMeul]

1 minute ago, Imglidinhere said:

Why not? That's the best way to find out how bad of a bottleneck there actually is.

 

Just play a suite of games at 720p with a 1080 and run the same games at the same clock speeds with the same number of cores and threads. This guy can obviously disable cores and threads in these tests. It's a surefire way to eliminate ANY other factors... and it's highlighted a processor bottleneck really really rapidly in the past too.

 

[zMeul]

1 minute ago, djdwosk97 said:

Yes, and if you convert it to using Haswell as the baseline instead of Excavator, then it indicates a 17% increase from Haswell to Kaby Lake.

I'd like you to show me how

[79wjd]

Just now, zMeul said:

I'd like you to show me how

 

7 hours ago, Glenwing said:

 

No, the scores here show Kaby Lake to be about 17.3% higher than Haswell. It's just because they're percentages, not absolute numbers, and percentages depend on reference point.

 

For example:

 

A = 50

B = 100

C = 150

 

If we compare these as a percentage of A:

 

A = 100% of A

B = 200% of A

C = 300% of A

 

Notice that C is 100 percentage points above B when compared like this, even though C (150) is only 50% higher than B (100).

 

In this case, yes Kaby Lake is 180% and Haswell is 150%. However, both of these percentages are "percent of Excavator IPC". So Kaby Lake is 30%-of-Excavator's-IPC above Haswell. But when people ask what the IPC of Kaby Lake is compared to Haswell, usually they want to know how many Haswells-worth of IPC higher it is, not how many Excavators-worth of IPC it is. For that, you would need to convert from the numbers in this chart (just divide the Kaby Lake percentage by the Haswell percentage to get the percentage of Kaby Lake above Haswell).

 

[zMeul]

1 minute ago, djdwosk97 said:

 

no, don't quote me glenwing, show me the actual data that shows a 17% IPC increase for Kaby Lake over Haswell, where Haswell is the baseline

[Imglidinhere]

4 minutes ago, zMeul said:

 

This is not at all what I mean... he's running the i7 at 5GHz and the 1700 at 3.9GHz. Run BOTH processors at the same clock frequencies to eliminate any real differences. To be fair, what I'm asking for will only happen when the R3 1400X comes out.

[Glenwing]

10 minutes ago, zMeul said:

I'd like you to show me how

Algebra, systems of 2 equations:

 

H = 1.55 × E

K = 1.82 × E

 

If both of these equations are true, it follows that K = 1.17 × H, at least according to the information provided.

[Imglidinhere]

4 minutes ago, zMeul said:

no, don't quote me glenwing, show me the actual data that shows a 17% IPC increase for Kaby Lake over Haswell, where Haswell is the baseline

. . . . . . Dude, @Glenwing used math to subjectively pull a 17.3% improvement from that list, from Haswell going to Kaby Lake. It's not hard to understand what he did. It's actually Algebra 1...  

 

Awwwww... I was too late... :c

Edited March 4, 2017 by Imglidinhere
Awwww.... I was too late. :C

[79wjd]

5 minutes ago, zMeul said:

no, don't quote me glenwing, show me the actual data that shows a 17% IPC increase for Kaby Lake over Haswell, where Haswell is the baseline

Take the absolute IPC graph where Kaby Lake is at 180.33% and use Haswell (154.85%) as the baseline instead of 100% that is Excavator. 

 

Zen has 146.62% IPC of Excavator, as admitted in your own (original post), which, again, according to you, means that Zen is 46.6% above Excavator (i.e. 146.6%/100% = 1.466 = 46.6% increase). Now, let's use Haswell as the baseline, 180.33%/154.85% = 1.1645, which is a 16.5% increase. 

 

17 hours ago, zMeul said:

back in 2015, AMD estimated a 40% IPC increase over Excavator, they achieved that and 6.6% over it

 

[zMeul]

4 minutes ago, Glenwing said:

Algebra, systems of 2 equations:

 

H = 1.55 × E

K = 1.82 × E

 

If both of these equations are true, it follows that K = 1.17 × H, at least according to the information provided.

I'm going to manually sort through all of his results and try to make the average

the problem is that all his individual results are still compared in performance to Excavator

[Glenwing]

1 minute ago, zMeul said:

I'm going to manually sort through all of his results and try to make the average

It's just the information that's on the chart in the OP...

[79wjd]

14 minutes ago, zMeul said:

I'm going to manually sort through all of his results and try to make the average

the problem is that all his individual results are still compared in performance to Excavator

I can save you some trouble, most are around a 0-10% increase from Haswell to Kaby Lake with Cinebench at around 14% and a bunch of random other ones around 20% and one or two around 50%. 

 

And you'll end up with (more or less) the same results regardless of when you convert to using Haswell as the base (before averaging all the graphs together, or after).

[zMeul]

Just now, Glenwing said:

It's just the information that's on the chart in the OP...

no, I'm going to manually take all the individual results from the source and average them - it will take me some time and I will probably make a mess but hey .. in the name of science 

[Glenwing]

Just now, zMeul said:

no, I'm going to manually take all the individual results from the source and average them - it will take me some time and I will probably make a mess but hey .. in the name of science 

Oh sorry, I thought you said "I'm not going to sort through..." and I was like uhh no one said you needed to do that...

[Kinda Bottlenecked]

Ho...

 

Wouldn't that put Zen real close to Haswell ipc? 

[tlink]

2 hours ago, zMeul said:

that's a lie

programs do not optimise for nothing, the same compilers that show poor gaming performance for Zen are used when shown good multithreaded workload performance for Zen

compilers generate exes that work on the x86 arch, nothing more, nothing less - you can instruct compilers to use / not use a particular instruction set

but those instruction sets are already established - if AMD hasn't implemented them correctly, that's on AMD

no thats not a lie. you have two platforms, lets just call them AMD and intel. intel has a way bigger marketshare and pretty far superior products. with what platform in mind are you going to write code? intel ofcourse. never did i say every program out there did this, but processors are complex things even if they use standardized instruction sets, how you write code still makes a difference in performance. 

[zMeul]

22 minutes ago, tlink said:

no thats not a lie. you have two platforms, lets just call them AMD and intel. intel has a way bigger marketshare and pretty far superior products. with what platform in mind are you going to write code? intel ofcourse. never did i say every program out there did this, but processors are complex things even if they use standardized instruction sets, how you write code still makes a difference in performance. 

that's a complete and utter fallacy

and you can easily inspect each and every executable to see with what compiler it was created with

 

now .. go away, I have math problems to do 

[zMeul]

@Glenwing + @djdwosk97 + @Imglidinhere  plus the rest ...

 

I completed the calculation relative to Hawsell, kaby Lake has a 15.24% IPC increase, and with extremes removed has 14.29% IPC increase

like it or not, that's what the data shows; or at least my math 

 

Haswell - Broadwell - Skylake - KabyLake

if you take ~5% increase from each generation over, that's what it adds up to

isn't that what it's said, 5% IPC increase with each generation? well ...

[Sharkyx1]

18 hours ago, KOMTechAndGaming said:

ok so rip the 1800x

Well just like any X intel chip or an nvidia titan. 

[Bouzoo]

16 minutes ago, zMeul said:

@Glenwing + @djdwosk97 + @Imglidinhere  plus the rest ...

 

I completed the calculation relative to Hawsell, kaby Lake has a 15.24% IPC increase, and with extremes removed has 14.29% IPC increase

like it or not, that's what the data shows; or at least my math 

 

Haswell - Broadwell - Skylake - KakyLake

if you take ~5% increase from each generation over, that's what it adds up to

isn't that what it's said, 5% IPC increase with each generation? well ...

In that case the graphs are wrong. Skylake to kaby Lake has 0% improvement, and the article from anandtech that I linked shows 5.3% from Haswell to Skylake iirc? On phone atm. 

[zMeul]

2 minutes ago, Bouzoo said:

In that case the graphs are wrong. Skylake to kaby Lake has 0% improvement, and the article from anandtech that I linked shows 5.3% from Haswell to Skylake iirc? On phone atm. 

Anand did only 4-5 tests (single threaded!), the dude on the forums did 26