First complete review of Haswell i7-4770K

[Lepton87]

Hmm, well...if those 3dMark11 scores are true I expect GT3 with that(!) super fast RAM to be faster than a gt650. And the GT3e a bit more, no? Unless the GT3e bandwidth is less than that 2600Mhz RAM?

No way GT3e is going to be faster then GT660M in games, It can be 3x faster in 3dmark11 for all I care, I don't play 3dmark. Just take a look how it compares to GT630 in 3dmark11 and how that reflects actual performance in games.

 

[willomz]

The difference must be down to drivers, surely Intel will be able to improve on that?

 

[Vesku]

The difference must be down to drivers, surely Intel will be able to improve on that?

Isn't the most likely culprit memory bandwidth, given the GT 630 they compared it to had DDR5? Question becomes, is 3DMark not exercising the memory as much as actual games? Or is Intel able to overcome slow memory via drivers for 3DMark and, if so, is that method extendable to actual games?

 

[MightyMalus]

No way GT3e is going to be faster then GT660M in games, It can be 3x faster in 3dmark11 for all I care, I don't play 3dmark. Just take a look how it compares to GT630 in 3dmark11 and how that reflects actual performance in games.

Oh, I agree. But, this super high speed cache must do something right?

So, I'm guessing, by scaling, that GT3 will score 3000, and GT3e 10~20% higher than GT3.

I'm gonna make an unintelligent statement, maybe the EU's architecture is optimized for something else? Consider the "leak", RAM's at 2600Mhz. Scores double, yet perform only slightly better? What's holding it back? If it was the memory bandwidth, wouldn't the benchmark scores also be impacted?

Totally clueless, to the details, simply wondering...

 

[colonelciller]

when will the Intel Marketing Executives be giving Anandtech permission to post the Anandtech review on Anandtech.com???

 

[colonelciller]

Bad information is worse than no information.

perhaps, but nobody here is in any position to say that the posted review is inaccurate, unless there is a completed review that is not being released

 

[notty22]

Intel and Anand had a couple early previews of Sandy Bridge and it was impressive. Most notable, I remember was Intel showed a couple towers with high o/c's 4500mhz and one near 5000mhz I believe (think it was a 2500k/2600k on water). Then Anand did a preview at least 4 months before SB launched.
AnandTech | The Sandy Bridge Preview


Not sure , why the silence with newer launches. Different strategy?

 

[ViRGE]

Intel and Anand had a couple early previews of Sandy Bridge and it was impressive. Most notable, I remember was Intel showed a couple towers with high o/c's 4500mhz and one near 5000mhz I believe (think it was a 2500k/2600k on water). Then Anand did a preview at least 4 months before SB launched.
AnandTech | The Sandy Bridge Preview


Not sure , why the silence with newer launches. Different strategy?

I can't imagine Intel was very happy with those early articles. Maybe they've been cracking down?

 

[colonelciller]

sounds like Intel controls the reviews on Anandtech then...

 

[Enigmoid]

HD4600 (4770K) vs HD4000 (3770K)

Is that really 4x and 8x AA?

Intel's igps have never been particularly good at high levels of AA (and really high resolutions). I expect that at tests using no or 2x AA the gt2 will be closer to the 630. Who runs 8x AA on an igp anyway?

 

[IntelUser2000]

Oh, I agree. But, this super high speed cache must do something right?

So, I'm guessing, by scaling, that GT3 will score 3000, and GT3e 10~20% higher than GT3.

Based on that the 3770K is getting 815 points, 2.9x that according to Intel's slide would end up in the 2300 point range, if GPU score closely correlates Overall scores.

 

[Zucker2k]

As we get our hands on the Core i7-4770K is the [Bold]ES[/Bold] project template, so just the appearance of the picture we did not see any type of words. New CPU-Z 1.64 version has been perfectly detected Core i7-4770K models, in addition to the voltage part still exists BUG rest of the various parameters are displayed properly.

The new Intel Core i7-4770K and i7-3770K using the 22nm process quad-core eight thread design design, with 8MB L3 cache. Both default clocked to 3.5GHz maximum Turbo can accelerate to 3.9GHz. The only difference is the i7-4770K can be clocked down to 800MHz (i7-3770K only reduced frequency to 1.6GHz) and joined the the new AVX2 with FMA3 two instruction set.

Still not good due to the current-voltage detection CPU-Z version of Haswell Support section display 0.880V value is not the real value. Measured, Fengyun Core i7-4770K standby voltage of about 0.7V, [Bold]and the full load voltage is approximately 0.976V.[/Bold]

http://translate.google.com/transla...p://www.chinadiy.com.cn/html/24/n-9024-3.html

Should be mentioned that the Fengyun Core i7-4770K and tested before the Xeon E3-1280V3, [Bold]the maximum Turbo amplitude Z87 motherboard 3.9GHz can only achieve the full load of the single-core, core frequency at full load at the same time when the quad-core can be maintained only on the 3.7GHz. And test comparison i7-3770K can be implemented in the MSI Z77M-Power quad-core loaded with 3.9GHz frequency at the same time (most of the Z77 motherboard upgrade can be achieved when the core is fully loaded with four octave,[/Bold] I believe that is the motherboard after optimized results), which is why we want the evaluation part is divided into the default state reasons and the same frequency of the status of the two test items.

http://translate.google.com/transla...p://www.chinadiy.com.cn/html/24/n-9024-3.html

So
*Engineering Sample
*Premature bios
*Gimped Turbo Boost (3.7Ghz)
*Full load of less than a volt (approximated)

Kicks butt nonetheless! Definitely on course to be the most powerful mainstream desktop processor. Bring on the official retail Haswell reviews.

 

[OBLAMA2009]

dayum, this is like the third or fourth straight generation of chips to show little or negligible performance improvement. no wonder the pc industry is in a slump

 

[ViRGE]

sounds like Intel controls the reviews on Anandtech then...

Controls when AT reviews, perhaps. After all, how is AT supposed to review something if they don't have the product?

 

[georgec84]

dayum, this is like the third or fourth straight generation of chips to show little or negligible performance improvement. no wonder the pc industry is in a slump

Shrinking down to 22 nm and less, it seems the performance gains level off a bit. The challenge seems to be more about efficiency and cooling as we see smaller and smaller transistors.

 

[Smartazz]

How many times are we going to hear that Intel hasn't made big strides lately? Yes, they've slowed down a bit but Conroe to Penryn and Penryn to Nehalem weren't THAT much faster at release. It took time for Nehalem and Sandy Bridge to start crushing the older chips.

 

[jpiniero]

dayum, this is like the third or fourth straight generation of chips to show little or negligible performance improvement. no wonder the pc industry is in a slump

Now you see why Intel isn't going to bother releasing Broadwell LGA. There is no point.

 

[A5]

sounds like Intel controls the reviews on Anandtech then...

You must be new to the whole PC hardware scene if you don't know what NDAs are and why sites follow them.

Between the NDA and distributing review samples, Intel (and every other computer company - AMD, Nvidia, Apple, etc all do the exact same thing) controls the timing of the review cycle. They don't exercise any editorial control over the review.

 

[Charles Kozierok]

perhaps, but nobody here is in any position to say that the posted review is inaccurate, unless there is a completed review that is not being released

You're free to take that approach if you like. Mine is that all sources are suspect until they establish a credible reputation. If I've never heard of someone, I don't even read their reviews unless I know someone who can assure me that they have a clue. Too much bogus crap floating around.

Now you see why Intel isn't going to bother releasing Broadwell LGA. There is no point.

I thought that had been debunked months ago?

As for Anand, I bet he has hardware right now, but he's not going to say anything until the NDA is lifted.

 

[jpiniero]

I thought that had been debunked months ago?

Intel only said they are not killing off LGA completely. IOW, they will do LGA Skylake, skip Skymont and let's see where the market is a year or two from now.

 

[StrangerGuy]

dayum, this is like the third or fourth straight generation of chips to show little or negligible performance improvement. no wonder the pc industry is in a slump

It's in a slump because the PC industry thinks having a faster CPU that loads web pages on a lousy 768p screen milliseconds faster is much more important than typical tablet things like instant standby.

 

[Charles Kozierok]

Intel only said they are not killing off LGA completely. IOW, they will do LGA Skylake, skip Skymont and let's see where the market is a year or two from now.

This is far from iron-clad, but Xbit has a pretty good track record IME.

 

[krumme]

I still think there is a market for faster hardware, that gives unique experiences.

Intel have the finest engineering teams in the world, with competences out of this world, and they are working on improving IPC 2-10% with each generation.

What is the meaning of this nonsense?

What do we do our selves to promote faster hardware?

Several reviewers and especially Anand have this eye on IPC, and its excactly that focus that is hindering performance where it could really matter - > multicore the next few years then avx2/hsa/?? full throttle (like inf64 wrote).

All, and that includes us, got their eyes on this IPC. If we had moved that years ago, all that valuable competence could have focused more productively elsewhere. Its next to useless. Imagine what could have been invented for the PC if the competences had looked elsewhere.

The effect is a monopoly market that gives average consumers no benefit, and therefore they are going away fx. to mobile -> Arm. Typically Otellinie strategy, and we were dancing to his tune, but the long time eroding of the pc market was one effect. Otellini was a killer of all innovation, no wonder JHH called Intel a money machine.

After all the market reacts to this nonsense. And now Quallcomm have same market cap as Intel. No wonder.

You guys remember what adding a fpu to x86 meant for performance? OoO? Integrated memory controller? Accerated 2D or 3D? We have the tech and the programming today to give us the same huge jump, we just have to stop looking at that idiotic IPC. Its a mental prison.

 

[ShintaiDK]

sounds like Intel controls the reviews on Anandtech then...

Previewing/Reviewing with stolen ES samples is a punishable offense by law.

AMD does exactly the same.

 

[mikk]

Is that really 4x and 8x AA?

Intel's igps have never been particularly good at high levels of AA (and really high resolutions). I expect that at tests using no or 2x AA the gt2 will be closer to the 630. Who runs 8x AA on an igp anyway?

Yes this is true with 4xMSAA/8xMSAA Intel is worse than without. I don't know if this is a driver or hardware thing. 2xMSAA is even worse because Ivy Bridge (and maybe Haswell) don't support this in hardware. I guess having a dedicated 1GB GDDR5 memory helps the GT630 a lot in Benchmarks with MSAA because the bandwidth requirements and memory footprint is much higher. 3dmark in default state isn't using MSAA, so this can be a completely different case. In almost every current game MSAA with integrated graphics is not usable. I wonder what games they used.