Kabylake review from [H]. This is really bad :(

[Hulk]

What the hell is Intel thinking? Zero performance increase? A small efficiency gain and some improvements in the media core. And this isn't a die shrink its a... a... I don't know what it is actually. Stalling I guess, marketing.

Perhaps not just all of the low hanging fruit is gone, but all of the fruit is gone.

If ever there was an opportunity for AMD to compete now it is. Intel opened the door and basically said, "we're gonna hold here for a few years so you can catch up."

 

[SPBHM]

maybe the IGP is better?

 

[TheELF]

maybe the IGP is better?

Yeah I hope so too but the chances are very very bad that it will.
Still, being able to encode h265 with minimal CPU load will be fun.

 

[Phynaz]

What the hell is Intel thinking?

What Intel is thinking is the desktop PC is dead.

https://forums.anandtech.com/threads/brian-krzanich-just-killed-the-pc.2471343/

 

[Hulk]

Process -> Architecture -> Optimization?

More like Sleep -> Yawn -> Stretch

Excellent but I think it should be Stretch>Yawn>Sleep
Since they are definitely in a sleep cycle now

 

[LTC8K6]

maybe the IGP is better?

The IGP was 20-30% faster in the mobile chips, possibly due to being able to boost for a lot longer.

 

[Excessi0n]

What the hell is Intel thinking? Zero performance increase? A small efficiency gain and some improvements in the media core. And this isn't a die shrink its a... a... I don't know what it is actually. Stalling I guess, marketing.

There isn't zero performance increase. [H] decided they would compare the 7700K to a 6700K overclocked to match. They didn't test with both at stock and they didn't test with both overclocked. They deliberately tested with settings that would give identical results and then lambasted Intel because they got identical results.

I'm not sure why they felt the need to misrepresent the 7700K when the performance increase is already tepid, but they did.

 

[2is]

Not the same performance, the speeds out of the box are higher. [H] even said that these things will overclock better than the 6700K.

What else were you expecting?

I don't think his expectations are unreasonable at all. An i5 6600 stock clock is higher then an i5 6500 and they both have the same architecture. Why are they both called Skylake? Why is an i7 4770k and i7 4790k both called Haswell? Some replies here, yours included seem to imply that a small clock speed bump is reason enough for an entirely new CPU line, yet this is rather unprecedented.

 

[iamgenius]

I don't think his expectations are unreasonable at all. An i5 6600 stock clock is higher then an i5 6500 and they both have the same architecture. Why are they both called Skylake? Why is an i7 4770k and i7 4790k both called Haswell? Some replies here, yours included seem to imply that a small clock speed bump is reason enough for an entirely new CPU line, yet this is rather unprecedented.

Great post buddy!

 

[Hulk]

They deliberately tested with settings that would give identical results and then lambasted Intel because they got identical results.

I'm not sure why they felt the need to misrepresent the 7700K when the performance increase is already tepid, but they did.

I don't understand? The first thing most people who study/follow the CPU market want to know when a new core is released is the IPC compared to previous generations. That is the only way to really isolate architectural improvements. This is the FIRST Intel new core that shows absolutely no IPC improvement. And honestly clockspeed is a wash. A few hundred MHz up or down. We've been farting around 4GHz for like 10 years now.

No IPC improvement from Skylake. That sucks. And this is coming from someone who would defend Intel when a new core would "only" show a 5% IPC increase from the previous generation. 5% is 5%. 0% is 0%.

I'm sorry but as I see it thus far Kaby Lake for the desktop if DOA. I'm on Haswell now and there is no way I'll consider Kaby Lake upgrade. Intel will sell millions of them in pre-built systems and mobile devices so they don't care.

 

[Excessi0n]

The new cores don't show improvements per clock because they aren't new cores. We've known for a long time now that Kabylake wasn't going to have any changes to the architecture. It was never going to be something that would entice owners of recent-generation chips to upgrade. If you expected otherwise then your expectations weren't low enough.

 

[CHADBOGA]

I don't understand? The first thing most people who study/follow the CPU market want to know when a new core is released is the IPC compared to previous generations. That is the only way to really isolate architectural improvements. This is the FIRST Intel new core that shows absolutely no IPC improvement. And honestly clockspeed is a wash. A few hundred MHz up or down. We've been farting around 4GHz for like 10 years now.

No IPC improvement from Skylake. That sucks. And this is coming from someone who would defend Intel when a new core would "only" show a 5% IPC increase from the previous generation. 5% is 5%. 0% is 0%.

I'm sorry but as I see it thus far Kaby Lake for the desktop if DOA. I'm on Haswell now and there is no way I'll consider Kaby Lake upgrade. Intel will sell millions of them in pre-built systems and mobile devices so they don't care.

The days of people upgrading their CPU's every year or so because the improvements on offer by the latest and greatest, has been over for quite a few years now, only the diehard 1% of the market do this and now even they will surely drop off in great numbers(percentage wise of their numbers).

 

[Vesku]

The IGP was 20-30% faster in the mobile chips, possibly due to being able to boost for a lot longer.

My 3DMark11 score on i3-7100U seemed very close to the listed results for i3-6100U, low single digit % variation.

 

[StrangerGuy]

The days of people upgrading their CPU's every year or so because the improvements on offer by the latest and greatest, has been over for quite a few years now, only the diehard 1% of the market do this and now even they will surely drop off in great numbers(percentage wise of their numbers).

Yup, it's exceedingly hard to justify a new CPU over my 4790K, even if Zen is good. Same goes for NVMe SSDs over SATA.

 

[.vodka]

I'm sorry but as I see it thus far Kaby Lake for the desktop if DOA. I'm on Haswell now and there is no way I'll consider Kaby Lake upgrade. Intel will sell millions of them in pre-built systems and mobile devices so they don't care.

Of course it's DOA. That's what happens thanks to process -> architecture -> optimization. We've already had that before, process with IVB, architecture with HSW, optimization with HSW refresh / devil's canyon. The sequence then repeats with BW/SKL/KBL. Thanks physics, you're making CPUs boring after decades of improvements and exponential $ needed to overcome you. It also doesn't help that Nehalem (stretching it..) or Sandy still feel mostly fine for daily and most demanding tasks. Put an SSD in there and you probably couldn't tell them apart from a new machine.

HSW/BW -> SKL/KBL is only worth it if you're willing to pay for DDR4 above 3500-4000MHz to extract every drop of potential out of it. Then you get the exact same cores with a better GPU in KBL. meh for the desktop, who cares? Optimization as shown in the 7700k (in particular RichUK's as an example) only got Core to a stable 5GHz thanks to delidding, if it weren't for that he'd hit a brick wall due to temperatures. On top of that it's a single sample, is it a dud, the norm, or KBL's best? Mobile gets the efficiency improvements from optimization, we don't.

Remember the i7 9xx line getting a C0 -> D0 stepping change back then, being less of a furnace and better overclockers overall? It didn't warrant the i7 2xxx name... KBL somehow gets bumped from 6xxx to 7xxx. The only redeeming point in KBL's release is Z270 and Optane support... but then, it'll cost an arm and a leg, and a NVMe PCIe SSD is already pretty freaking amazing on its own. We've already got that. SKL and its identical refinement KBL is as of today a 1.5 years old architecture. Coffeelake gets us 6c mainstream CPUs, Cannonlake is a KBL shrink? Probably a few tweaks here and there like Broadwell. Icelake is the new architecture in 2018, and by that point we've been on SKL for about 3 years. Intel just pulled a Netburst again (2001-2006) or a Bulldozer if you will (2011-2016). Again, thanks physics and increased costs.


I think the real SKL architecture is what HEDT is getting in Skylake-X... mainstream SKL seems watered down, always seemed mostly unimpressive vs HSW and especially BW+L4 cache. It doesn't help that Intel doesn't do architecture deep dives anymore, Skylake is more of a black box than anything else... when was the last time they disclosed any significant information to the press? Haswell? At least AMD got the decency to let the public know what makes Zen tick, more or less.

Maybe Zen makes Intel wake up somewhat, get some exciting products out instead of the tried and true formula (4c8t mainstream / HEDT overlapping) they've been doing since Nehalem. It's stale, boring. Time to get BW's cool L4 cache across the board or in some select K parts... don't you think?



http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation

In the past Intel has historically run a 1:1 policy whereby a 1% performance gain must come at a maximum of a 1% power penalty – this was adjust to 2:1 for Broadwell, and we should assume that Skylake had similar requirements during the planning stage.

Also probably time to try other tricks that now can't be done thanks to that policy, even if it means going over 100w TDP rating again for the typical 4C8T CPU. It's not like they're gonna end up doing a Prescott or Tejas...

 

[frozentundra123456]

There isn't zero performance increase. [H] decided they would compare the 7700K to a 6700K overclocked to match. They didn't test with both at stock and they didn't test with both overclocked. They deliberately tested with settings that would give identical results and then lambasted Intel because they got identical results.

I'm not sure why they felt the need to misrepresent the 7700K when the performance increase is already tepid, but they did.

Well said. We still have to see max overclock to see if there is an ultimate performance gain on the desktop, but there is definitely a gain in mobile and at stock clocks on the desktop. In fact, the gains in mobile were pretty much in line with that seen in a new generation.

 

[Hulk]

The new cores don't show improvements per clock because they aren't new cores. We've known for a long time now that Kabylake wasn't going to have any changes to the architecture. It was never going to be something that would entice owners of recent-generation chips to upgrade. If you expected otherwise then your expectations weren't low enough.

We are going to have to agree to disagree. This is not a new architecture and it's not a new process (die shrink). So what is it? Just a rebranding of Skylake to 7xxx? Rationale-wise Intel has lost me on this one. Seems like this is more just a new stepping of Skylake rather than something that should have a new name.

For 40 years now new releases have either had IPC improvements or process size reduction to allow for more transistors and faster speeds. Sometimes both actually. This release has neither.

I'm not dismissing the efficiency gains. There is SOMETHING there I suppose. But like I wrote above it's so minimal that this used to be simply a new stepping of the current process.

 

[thesmokingman]

What else were you expecting?

What the five? I keep seeing these defensive replies and it is hilarious. Talk about anti everything we're supposed to be about.

 

[arandomguy]

The problem is the 6700k real market value now is really around $300 US or lower now. The 7700k (Kaby) effectively is just to raise ASP (and margins) back up via other "improvements" and the issue here is those improvements don't really matter to certain market segments, hence they are majorly disappointed.

If Intel also announced the 7700k's MSRP would actually be lower (in preparation for the market shift with Coffee Lake), than people might have been excited. Otherwise what now? You're over paying compared to what you get over Skylake. You're also one year out from likely when Intel themselves will move to more cores mainstream (never mind AMD with Zen) at possibly similar price points, which will crater the actual value of a 4c/8t processor.

If they dropped MSRP along with the minor performance bump like they do with GPU refreshes reaction would've been different. There would've more hype had they officially just dropped 6700k MSRP to well below $300 without this new release at all.

 

[coercitiv]

If Intel also announced the 7700k's MSRP would actually be lower (in preparation for the market shift with Coffee Lake), than people might have been excited. Otherwise what now? You're over paying compared to what you get over Skylake. You're also one year out from likely when Intel themselves will move to more cores mainstream (never mind AMD with Zen) at possibly similar price points, which will crater the actual value of a 4c/8t processor.

If they dropped MSRP along with the minor performance bump like they do with GPU refreshes reaction would've been different. There would've more hype had they officially just dropped 6700k MSRP to well below $300 without this new release at all.

What makes you think that Intel would offer mainstream 6c/12t at the same MSRP as 6700K/7700K? Except for competition (which would/will bend prices no matter the number of cores), there would be no real reason for Intel to not shift the entire product stack upwards. Keep in mind that as far as we know, the performance boost would be really big across the entire SKU inventory, with i3 actually getting a bigger jump than i5/i7.

In the absence of a feisty AMD I would not be surprised to see all Core products jump $50 up once Coffee Lake arrives. Reviewers would tell you the 50% jump in MT performance is all but worth it, and you could actually get lower prices than before if you stay at the same number of cores (CFL i3 vs. KBL i3). Everybody would be happy, even the people who don't realize right now that Kaby Lake comes with no extra cost to the consumer, who can actually buy whenever and whatever he wants at the best price possible - SKL is by no means obsolete tech, but will be obsolete brand.

January, when all merchants will be very interested to attract customers who already spent their $$$ on the winter holidays, Kaby Lake will offer them the best reason to discount Skylake.

 

[biostud]

The good thing with the lack of i provemnt is that it makes no sense to upgrade your cpu and motherboard and thus saving you lots of $$$, you can spend on other parts.

 

[Vaporizer]

A cheap 6 or 8 core that had nearly the same IPC might look more compelling against this quad cores.

 

[frozentundra123456]

What the five? I keep seeing these defensive replies and it is hilarious. Talk about anti everything we're supposed to be about.

Perhaps some of the replies are defensive, but some are also justified, because the article cited by the OP, while technically correct for the conditions of his test, did not tell the whole story.

 

[Hulk]

Perhaps some of the replies are defensive, but some are also justified, because the article cited by the OP, while technically correct for the conditions of his test, did not tell the whole story.

For me the article referenced by the OP actually does tell the whole story. If clock speed isn't isolated IPC performance can't be evaluated. If IPC is the same but clocks are say 5% faster for one core over another I think most people around here realize that means, with equal IPC, that the 5% faster core will perform 5% better (about, in reality a bit less). It's the reviews that only compare one part to another when we don't know clocks during testing (or other test specifics), especially these days with variable frequencies, where you can't fully understand what is going on. Those kinds of tests are for non-enthusiasts in my opinion. They don't want to know about architecture, just which is faster from a performance point of view. I can get that information from a crappy Intel propaganda slide.

Equalizing clocks as this review does is like stripping the clothes of the Emperor!

 

[LTC8K6]

My 3DMark11 score on i3-7100U seemed very close to the listed results for i3-6100U, low single digit % variation.

Notebookcheck had the increase with a 7500U, I believe.

The MSI CX72-7QL is equipped with 16 GB of the fast DDR4-2133 memory in a dual-channel configuration, which is a great way to utilize the maximum performance of the HD Graphics 620. Even though the HD Graphics 520 successor is still equipped with just 24 Execution units (EUs) and the clocks did not change, either (i7-6500U/i7-7500U: up to 1050 MHz), Intel was able to improve the performance considerably: The gaming performance is 20-30 percent higher on average, so the performance gains are even higher than the synthetic benchmarks indicated. One reason for the comparatively good results is once again the excellent Turbo Boost utilization: Depending on the game, the GPU runs at core clocks between 1000-1050 MHz.

All in all, the HD Graphics 620 is roughly on par with a dedicated Nvidia GeForce 920M or AMD Radeon R7 M440. Upcoming Kaby Lake chips with the more powerful Iris Graphics GPU (GT3e, 48 EUs, eDRAM) should even beat the GeForce 940M(X).