Intel Skylake / Kaby Lake

[crashtech]

I don't get it though, if you compare stock-to-stock, 6700K offered what, 10-15% higher perf/clock than 4790K but it regressed in frequency so the performance gain was often sub-10%.

7700K doesn't change IPC, but it gets a solid boost in frequency (10% all core turbo improvement) over the 6700K.

Why does it matter how the performance increase is achieved as long as it is achieved?

We don't disagree on any of the facts, and I'm glad Intel has been able to get some very fast parts out this iteration. I don't have a problem with how it's done, I have a problem with what it's named. I know you won't "get" that, and that's fine. Carry on!

 

[lopri]

i have found that 3866c16 tweaked with a 6700k at stock is faster in game than 3200c14 xmp with the 6700k at 4.7ghz core\cache

That is sort of expected seeing the absolute latency of 3866/CL16 is 8.27ns (16/3866/2 = 0.00827) and 3600/CL14's is 8.75ns( 14/3600/2 = 0.00875). Assuming similar configuration (i.e. tRD, tRCD, tRAS values as usual), 3866/CL16 should be faster than 3600/CL14.

It is true though traditionally Intel CPUs seem to prefer higher bandwidth to lower latency.

i guess i should provide a little of the data so you can see for yourself

That is a rather extreme scaling. I am not sure if I have ever seen anything like that. What I typically remember is:

DDR4 Haswell-E Scaling (No appreciable gain at all)
DDR3 Haswell Scaling (Max 6% improvement)


Even DDR4 v. DDR3 is underwhelming:

Comparing DDR3 to DDR4 on Haswell-E (Mixed results)

Only cases I have seen the kind of performance gains as seen on the quoted graph are:

1. When a task is iGPU-bound (i5 Series and A-Series APU)
2. When the memory controller (or L3) is the bottleneck (Thubamn CPU-NB)

Did things change in Skylake and DDR4?

 

[Arachnotronic]

I have a problem with what it's named. I know you won't "get" that, and that's fine. Carry on!

It's named as it is because that's what the OEMs want, they want to be able to sell systems with new chips. It does nobody in the industry any good to call these chips "6th generation core" because for many segments the performance gains are significant, especially at low power.

Also since there are feature differences (media), it can become a nightmare for customers to know what they're buying if there are sixth gen Core CPUs with different/more features than other sixth gen Core CPUs.

And finally, AMD called Bristol Ridge 7th Gen, so should Intel call their updated CPUs (which were updated just as much as AMD's were) sixth gen, putting them at a marketing disadvantage?

The naming is what it should be.

 

[lopri]

Wait, what is the 6700K's stock freq?

 

[dahorns]

Wait, what is the 6700K's stock freq?

Base 4.0, Turbo 4.2

http://ark.intel.com/products/88195/Intel-Core-i7-6700K-Processor-8M-Cache-up-to-4_20-GHz

 

[IntelUser2000]

Where does it show that these SKUs maintain turbo for longer than skylake? I see some with higher boost clocks scoring higher, but there is not data on what clock speed they can maintain relative to similarly clocked skylake SKUs. That could just be better binning.

It's too simplistic to say its merely binning. I do believe in the amount of work they did on the process to call it a 14nm plus. That said, we are always comparing from a historical perspective, and in those terms, Kaby Lake sucks. There's no reason to go out of the way to disprove Intel's claims that its a new process. I mean, the gains are on par with actual shrinks(sometimes even better) for the last few years. And they seem to do that with less power on all frequency levels(http://www.notebookcheck.net/Kaby-Lake-Core-i7-7500U-Review-Skylake-on-Steroids.172692.0.html).

Cars aren't the best comparison for most, because they've been at a mediocre curve for decades. Computers only recently. As someone said, they(enthusiats/architects) will stumble over themselves trying to get over the curve(or wanting it at least), but fail to do so because its at its limits.

Yes, it does clock quite a bit higher, if you look at mobile reviews. At least for the U parts. Everything they do seems to boost looks to be bringing results on the U chips. I doubt it'll last long though. The mobile chips aren't far from desktop in terms of peak frequency. Sooner or later mobile chips will bring 5% per generation like desktop.

 

[jpiniero]

Yes, it does clock quite a bit higher, if you look at mobile reviews. At least for the U parts. Everything they do seems to boost looks to be bringing results on the U chips. I doubt it'll last long though. The mobile chips aren't far from desktop in terms of peak frequency. Sooner or later mobile chips will bring 5% per generation like desktop.

There's room for improvement though. I'm almost wondering if Intel will focus more on (for lack of a better word) 'voltage-per-clock' and tricks to decrease heat output (to maintain turbo) for the mainstream line as opposed to IPC increases. While the EP line will do both of course.

 

[Arachnotronic]

Base 4.0, Turbo 4.2

http://ark.intel.com/products/88195/Intel-Core-i7-6700K-Processor-8M-Cache-up-to-4_20-GHz

More precisely:

1 core turbo: 4.2ghz
2 core turbo: 4ghz
3 core turbo: 4ghz
4 core turbo: 4ghz

7700K is:

1 core turbo: 4.5ghz
2 core turbo: 4.4ghz
3 core turbo: 4.4ghz
4 core turbo: 4.4ghz

 

[Arachnotronic]

I mean, the gains are on par with actual shrinks(sometimes even better) for the last few years. And they seem to do that with less power on all frequency levels(http://www.notebookcheck.net/Kaby-Lake-Core-i7-7500U-Review-Skylake-on-Steroids.172692.0.html).

yep, and there's a reason for that. Shrinking the metal stack actually runs counter to performance, while improving the transistors helps performance.

14nm+ doesn't shrink the die/metal stack, but it does improve the fin profile and other things in the process, so it's all upside performance wise. None of the "competing forces" that comes with shrinking.

 

[Dasa2]

Nice bm! How is the scaling any different than IB?
Looking at the Fallout graph it seems to me IB 3770 scales aprox the same with faster memory. Its just a very memory dependant bm and the timings and bandwith seems quite different between the modules selected. More illustrative than realistic. Now skl is faster so it obviously will benefit more from faster memory.

Btw a game like bf1 also demands a dual channel 1866 as minimum to keep the 60fps min.

yes sandy\ivy also scale well with faster ram its not new to skylake\karby
most games overflow cpu cache a bit but open world games seem to see the biggest gains in cpu efficiency from higher ram speeds
crysis 3 for example is one of the few games that has its cache usage very well optimized thus cares more about cpu clock speed than ram speed as can be seen in the eurogamer review

i dont own multiple ram kits so for ivy 1600c11 was stock for that ram but it overclocked to 2133c9 however i have tested a winder range of speeds on it in the past https://forums.anandtech.com/threads/ht-vs-ram-in-game-benchmark.2455994/
for sykylake the ram defaults to 2133c15 then the xmp takes it to 3200c14 and beyond that it was mostly done for curiosity to see how it responded and helped me decide where i wanted to leave the 24\7 oc which was 3866 16-16-16 2t 1.425v http://www.overclock.net/t/1569364/...-7-memory-stability-thread/2920#post_25557774

That is sort of expected seeing the absolute latency of 3866/CL16 is 8.27ns (16/3866/2 = 0.00827) and 3600/CL14's is 8.75ns( 14/3600/2 = 0.00875). Assuming similar configuration (i.e. tRD, tRCD, tRAS values as usual), 3866/CL16 should be faster than 3600/CL14.

It is true though traditionally Intel CPUs seem to prefer higher bandwidth to lower latency.


That is a rather extreme scaling. I am not sure if I have ever seen anything like that. What I typically remember is:

DDR4 Haswell-E Scaling (No appreciable gain at all)
DDR3 Haswell Scaling (Max 6% improvement)


Even DDR4 v. DDR3 is underwhelming:

Comparing DDR3 to DDR4 on Haswell-E (Mixed results)

Only cases I have seen the kind of performance gains as seen on the quoted graph are:

1. When a task is iGPU-bound (i5 Series and A-Series APU)
2. When the memory controller (or L3) is the bottleneck (Thubamn CPU-NB)

Did things change in Skylake and DDR4?

the tests done by anandtech are heavily gpu bottltecked which represents most games much like my r6 test where the 3770k at stock with the gpu overclocked is faster than the 6700k\ram oc with gpu at stock
faster ram doesnt help the gpu unless you run out of vram which you really dont want to do

 

[IntelUser2000]

There's room for improvement though. I'm almost wondering if Intel will focus more on (for lack of a better word) 'voltage-per-clock' and tricks to decrease heat output (to maintain turbo) for the mainstream line as opposed to IPC increases. While the EP line will do both of course.

Yes, it won't last.

Core i7 7600U is already at 3.9GHz. If we get 10-15% gain on Cannonlake, and 10-15% on Icelake, then we'll be back to 0-5% gains. That's in 2-3 years. Wonder if there's even 0-5% after that.

Just an interesting observation. The multi-threading performance of the 2 core, 4 thread 7500U is fast as a 4 core, 8 thread 940XM chip, a 1st generation Core i7 top of the line.

 

[Sweepr]

OCUK selling guaranteed 5 GHz + delidded Core i7-7700K

The key features of this product are:
-Guaranteed 5g+ Core Overclock in all none AVX applications. All games and most professional applications are non AVX.
- CPUs binned by myself and delidded by our technicians with Liquid Metal replacing original Intel TIM for upto 30c cooler running temps.
- Full 1 year warranty with OCUK
- Free Thermal Grizzly 1g TIM so as to get the best thermal bond with your heat sink and hence great cool running.

Interesting choice if you don't want to take the risks of doing it yourself. But of course it's more expensive than a regular Core i7-7700K.

https://forums.overclockers.co.uk/showthread.php?t=18765975
https://www.overclockers.co.uk/inte...a1151-pre-binned-processor-oem-cp-003-oe.html

 

[mikk]

Sooner or later mobile chips will bring 5% per generation like desktop.

It's really odd what you are saying considering that Intel is preparing the switch from 2 to 4 cores for 15W ULT. Unlike AMD there's a big room for MT improvement in basically all segments. It started with the low end Pentium and HT, mobile and desktop will follow this trend.

 

[jpiniero]

It's really odd what you are saying considering that Intel is preparing the switch from 2 to 4 cores for 15W ULT. Unlike AMD there's a big room for MT improvement in basically all segments. It started with the low end Pentium and HT, mobile and desktop will follow this trend.

Except they're not totally. Intel's going to sell the 4+2 Kaby Lake Refresh U alongside the Cannonlake 2+2 U/Y.

 

[LTC8K6]

Has anybody done the delid treatment to a 7350K?

 

[dullard]

It is really hard to consider this a "generational" change. More like next year's car model, with upgraded upholstery, radio, and some slight engine ECU changes. Same chassis, same engine, same tires, same tranny, same quarter-mile speed (for desktop CPUs at least). Maybe top speed was improved a few MPH due to the ECU changes at high RPMs, but nothing major new.

5% to 10% faster and a few features that if you need it are night/day differences -- all at the same price. Yet you are complaining because the number on a sticker went from 6 to 7. Of all things to complain about, faster processors at the same price shouldn't be one of them.

 

[Sweepr]

First 'proper' Core i3-7350K reviews and tests out there.

ComputerBase - INTEL CORE I3-7350K IN TEST: Overclocked to 5.1 GHz in a duel with real four-core CPUs

1080P Gaming

Applications Windows

Value for Money

www.computerbase.de/2017-01/intel-core-i3-7350k-test-overclocking


Legit Review - Core i3-7350K Tests @ Stock

http://www.legitreviews.com/amd-fx-9590-8-core-cpu-review_190566


PurePC: Test CPU Intel Core i3-7350K - will be cheap overclocking?

Intel Core i3-7350K is generally a very interesting proposal and a breath of fresh air in the ossified world of processors. Performance and overclocking potential provides a truly satisfying. Unfortunately, making full use of its features need relatively expensive motherboard.

Core i3-7350K @ 5 GHz was faster than FX-8350 @ 4.5 GHz in Battlefield 1, Dishonored 2, Deux Ex: Mankind Divided, Fallout 4, Hitman, Rise of the Tomb Raider, Totaw War Warhammer, Watchs Dogs 2 and The Witcher 3: Wild Hunt and slower in Crysis 3.

www.purepc.pl/procesory/test_procesora_intel_core_i3_7350k_bedzie_tanie_podkrecanie


Low power consumption, overclocks like mad and although expensive relative to Kaby Lake Penium and entry level Core i5, it's still the cheapest way to get the best single thread performance possible. Fun little chip, I only wish it was a little bit less expensive.

 

[crashtech]

It's a little sad that retailers are getting away with charging $190 for the 7350K. I have to avoid it just on principle for the time being.

 

[Arachnotronic]

It's a little sad that retailers are getting away with charging $190 for the 7350K. I have to avoid it just on principle for the time being.

At $190 it's an awful deal, to be avoided for sure. 7600K is the better bet at that price, same thread count but four actual cores.

 

[LTC8K6]

Looking at those charts and the voltage, you probably wouldn't get much more out of a 7350K with a delid job.

 

[Justinbaileyman]

So with Zen or Ryzen coming the end of next month do we know yet who has the the stronger per core strength as far as Ryzen Vs Kaby Lake goes? Or is it still all hush hush due to NDA?

 

[VirtualLarry]

Ah, i3-7350K, what a love-hate relationship I have with thee.

Great overclocking, great single-threaded speed, competent in gaming, what's not to love about it?

But ouch, that price, plus retailer gouging, plus having to spend big $$$ on overclocking mobos and fast RAM. Which, once purchased, why not go whole-hog on an unlocked "K" quad-core CPU?

If only my overclocking dreams could be realized cheaply. If ASRock released their H110-DS/Hyper board, and updated it so that it could BCLK OC the G4560 Pentium with HT, I think that I would be in Budget PC Overclocker's Heaven.

Edit: Oh, BTW, my G4600 KBL CPU came in the mail today. Haven't decided where I'm going to put it yet. Was thinking of the DeskMini, but both of mine are stuck in single-channel mode, until I can get another DDR4 DC SO-DIMM RAM kit.

 

[nvgpu]

http://www.fool.com/investing/2017/01/19/intel-corporation-looking-to-build-revolutionary-p.aspx

Our goal: to build a revolutionary microprocessor core to power the next decade of computing and create experiences we have yet to dream up. We're looking for micro-architecture, logic design, and high-speed circuit design talent to help us reinvent the Core IP.

 

[thesmokingman]

http://www.fool.com/investing/2017/01/19/intel-corporation-looking-to-build-revolutionary-p.aspx

Not sure that is a good sign lol. Wtf have they been doing the last decade?

 

[StrangerGuy]

Ah, i3-7350K, what a love-hate relationship I have with thee.

Great overclocking, great single-threaded speed, competent in gaming, what's not to love about it?

But ouch, that price, plus retailer gouging, plus having to spend big $$$ on overclocking mobos and fast RAM. Which, once purchased, why not go whole-hog on an unlocked "K" quad-core CPU?

I'm mostly negative about it because Intel has made a mockery of what old school OCing once stood for: cost effectiveness. I rather just have a G4560 and a RX470 for almost the price of a 7350K with very dubious CPU advantages, while a 470 will pay itself off in 6 months through mining.