I7 4770K Benchmarks

[ShintaiDK]

Its pretty simple with turbo. Its 4/3/2/2 depending on threads. Some boards turbooverclock so to say like Asus, MSI etc and does 4/4/4/4 that could potentially make it throttle on the stock heatsink.

A subject we also covered in one of the many temperature threads.

 

[mikk]

I'm well familiar with intel CPUs. MTed Turbo is not guaranteed by intel spec to be always at 3.9Ghz (in case of Haswell or 3770K).

I'm not saying the review is perfect, far from it. But there is no magic pixie dust left, sorry. This is Haswell and it's not that bad. It's just not major step up for SB/IB users unless one uses exclusively iGPU and doesn't care about discrete graphics. For pre-SB users it's a great step up.

That doesn't matter. Multithreaded Turbo is always active. 3770k with turbo enabled clocks 3.9 Ghz (or even more on some Z77 Boards). Some dumb Haswell tester with a pre release Board or pre release Bios possibly doesn't realize its Turbo doesn't work for some reason (not enabled in the Bios or just broken) and you get a flawed comparison between Haswell and Ivy Bridge because he possibly tested Ivy Bridge with a proper working turbo clock against Haswell with a broken or not working turbo clock. I see a big possibility that this happened here. What is your conclusion from this test? Haswell slower than Ivy Bridge?

 

[inf64]

That doesn't matter. Multithreaded Turbo is always active. 3770k with turbo enabled clocks 3.9 Ghz (or even more on some Z77 Boards). Some dumb Haswell tester with a pre release Board or pre release Bios possibly doesn't realize its Turbo doesn't work for some reason (not enabled in the Bios or just broken) and you get a flawed comparison between Haswell and Ivy Bridge because he possibly tested Ivy Bridge with a proper working turbo clock against Haswell with a broken or not working turbo clock. I see a big possibility that this happened here. What is your conclusion from this test? Haswell slower than Ivy Bridge?

Nope I don't see it being slower. Up to 8% IPC increase with some outliers just like in previous leaks. Or you are saying previous leaks were also flawed across the board? I'm just telling you not to expect miracles from Haswell. Intel is not claiming this so we are not to expect them. iGPU,AVX2 and integrated voltage regulators are Haswell's biggest improvements. If you need a very fast and yet very efficient notebook you cannot go wrong with Haswell. If you have 3770K (stock or OCed) in your gaming machine you won't feel a difference at all IMO (unless you happen to code your own AVX2 optimized application).

 

[mikk]

Nope I don't see it being slower. Up to 8% IPC increase with some outliers just like in previous leaks.

Here is the problem. According to this test Haswell slowed down. SuperPi, Fritzchess, Cinebench 11.5 slower on Haswell. That's why I told the turbo might not work properly and you claimed all is fine with the turbo.

 

[inf64]

Here is the problem. According to this test Haswell slowed down. SuperPi, Fritzchess, Cinebench 11.5 slower on Haswell. That's why I told the turbo might not work properly and you claimed all is fine with the turbo.

I'm telling you that Turbo works fine on 4770K since the increase from stock to 4.5Ghz is reflected in the scores. I posted proof in post on 1st page, go back and read it.

The results are like that since 3770K obviously Turbos to near max. value due to board/bios/whatever. Super Pi uses legacy instructions so if it's slower there (by a fraction of percent btw which is within margin of error) it's quite normal. Here and here you can see 3770K scoring slower than 4770K in this Chinese preview, a trend which I will demonstrate happens in other examples you used.

Another clue that 3770K runs at higher than normal Turbo (due to bios/board) is C11.5 score which is higher than normal as can be seen here , here and here. It should score 7.45-7.54pts on normal bios/board which is 7.88/7.54=1.045 or ~5% lower than what Chinese leak shows (again it Turbos to higher bin Vs the normal case). So taking normal score in C11.5 we can see Haswell scoring ~5% more, both at stock.

Last is Fritz chess. Under normal circumstances 3770K scores lower than what Chinese setup does : here and here . Normal score is therefore ~14000 pts or 4.5% lower than what Chinese review shows. Therefore Haswell is again showing higher stock performance vs normal 3770K.

Maybe they did run their 3770K under such a setup to make it more "even" since Haswell will be able ,at stock, to turbo higher and during longer periods of runtime Vs stock 3770K. This might be important for non-OCers but for enthusiasts that don't run stock setup it's basically unimportant.

 

[mikk]

I'm telling you that Turbo works fine on 4770K since the increase from stock to 4.5Ghz is reflected in the scores. I posted proof in post on 1st page, go back and read it.

This is plain wrong. I told you why. A 15% higher frequency can't translate into a 21% better Fritzchess result considering that Haswell should have the better IPC. If the Turbo didn't work on Haswell the clock difference is 28,5% for the 4,5 Ghz Ivy Bridge. 21% better score with 28% higher frequency makes sense with the IPC disadvantage from Ivy Bridge.

 

[inf64]

This is plain wrong. I told you why. A 15% higher frequency can't translate into a 21% better Fritzchess result considering that Haswell should have the better IPC. If the Turbo didn't work on Haswell the clock difference is 28,5% for the 4,5 Ghz Ivy Bridge. 21% better score with 28% higher frequency makes sense with the IPC disadvantage from Ivy Bridge.

I'm getting tired of this back and forth discussion. How do you know their stock 4770K setup ran at 3.9Ghz Turbo clock in that test? Exactly you don't and data shows us this. Stock 4770K in that test didn't reach 3.9Ghz, it was sitting between 3.7 and 3.8Ghz. Stock is 3.5Ghz and this is between 2 and 3 bins more, which is perfectly in line with fp intensive workload.

We have the score of 4.5Ghz, it's fixed clock and it will score this much when it launches in a month and you personally OC it to 4.5Ghz and run the same test (give or take 1-3% due to different bios,board,ambient temp and memory used).

 

[BTRY B 529th FA BN]

I'm sorry do more people want stock results, or baseline results that have no CPU features on like Turbo, Speed Step, Power savings features, etc? I would think baseline results would be more appealing.

 

[WhoBeDaPlaya]

I'm sorry do more people want stock results, or baseline results that have no CPU features on like Turbo, Speed Step, Power savings features, etc? I would think baseline results would be more appealing.

Baseline! No CPU-equivalent of the "highest playable", non apples-to-apples nonsense of [ H ] GPU reviews.

 

[cmdrdredd]

I'd say best to manually set clock speed to the max stock turbo frequency and compare them. Then again at the same clock speed across the board. Avoiding using any turbo, speed step, etc etc. Just set it to 3.8Ghz or whatever and run the bench.

It looks like I'll be waiting until Z97 for my upgrade though. Then I can just build a whole new box that takes advantage of SATA express.

 

[BTRY B 529th FA BN]

What will take advantage of SATAe?

 

[Idontcare]

I'd say best to manually set clock speed to the max stock turbo frequency and compare them. Then again at the same clock speed across the board. Avoiding using any turbo, speed step, etc etc. Just set it to 3.8Ghz or whatever and run the bench.

It looks like I'll be waiting until Z97 for my upgrade though. Then I can just build a whole new box that takes advantage of SATA express.

If you are trying to isolate the performance aspects of the CPU then you really must do this. Otherwise you are not really testing the CPU, rather you are testing the platform (of which the CPU is one variable).

The problem with relying on automated turbo-core/boost (for either AMD or Intel) is (1) the deterministic nature of the power-control algo's (making the turbo dependent on things like ambient temperatures, HSF thermal conductivity efficiency, case airflow specifics), and (2) mobo bios dependent as the manufacturer of any given mobo can (and do) monkey around with the current & temperature thresholds involved in the turbo-bin de-activation algos.

All of which is relevant to the end-user from a "platform performance" perspective, but none of which is relevant from a "processor comparison" analysis. One must isolate the processor from all the other variables in order to create non-convoluted data for an apples-to-apples comparison.

 

[Headfoot]

Is it just me or does that processor look awfully beat up?

 

[Idontcare]

Is it just me or does that processor look awfully beat up?

You don't want to know where they had to hide that thing whilst trying to smuggle it out of Intel's secret labs :ninja:

 

[Headfoot]

You don't want to know where they had to hide that thing whilst trying to smuggle it out of Intel's secret labs :ninja:

Well that certainly explains the smell

 

[cmdrdredd]

What will take advantage of SATAe?

SSD drives. As they become more mainstream and faster they are going to hit a wall on the transfer speed of current SATA versions. The new SATAe will allow 8Gbps and 16Gbps transfer speeds. Up from the current 6. It is a worthy upgrade for a market that is growing rapidly. Devices may not entirely saturate it in a solo configuration but in a RAID it becomes a problem and a new interface is needed to keep up.

 

[guskline]

You don't want to know where they had to hide that thing whilst trying to smuggle it out of Intel's secret labs :ninja:

IDC: What does the "e4" marking mean?

Surely Intel can trace this cpu back to the "tester" or more likely future "non-tester".

 

[ShintaiDK]

IDC: What does the "e4" marking mean?

Surely Intel can trace this cpu back to the "tester" or more likely future "non-tester".

Its a retail CPU, not an engineering sample. Distributors and OEMs already got loads stored just waiting for release.

 

[guskline]

ShintaiDK, Thank you for the explanation.