Overclocking Kaby Lake
While many will argue that Intel is utilizing some of the overclocking headroom that Kaby Lake provides to increase stock clocks and offer some semblance of performance increases over Skylake, the truth is that the updated 14nm+ process technology allows Intel to do more with the same architecture and design.
With base clock-aware frequency and voltage curves, and a new AVX offset ratio to lower AVX clock speed to avoid the thermal constraints those instructions can create, Intel thinks that the overclocking capability of Kaby Lake should be higher than older architectures. The truth is somewhere in the middle; the overclockability relative to stock speeds looks about the same as Skylake, but with higher starting values – and that still means going above 5.0 GHz with modest cooling.
For me, a pretty basic overclocker that I think represents the vast majority of the consumer market, the process of overclocking Intel CPUs over the last several generational steps has been unchanged when working on unlocked parts:
- Adjust the multiplier to a fixed number on all cores until instability occurs
- Increase voltage until given clock speed is stable
- Adjust multiplier again until higher voltages adversely affect temperature or have no affect
On the ASUS Maximus IX Code motherboard this was dead easy and I quickly found that the limit on my review sample was 5.1 GHz at a fixed voltage of 1.375v.
The screenshot above shows our Core i7-7700K running on all cores at 5.1 GHz at 1.375v during a run of POV-Ray, after completing a solid hour of stability testing. That is 600 MHz faster than the stock Turbo clock rate and a good 300-400 MHz faster than I have been able to reliably run at on my various Core i7-6700K parts over the last year. Temperatures on the CPU (based on readings from the ASUS AI Suite III software) never exceeded 70C while using a Corsair H100i GTX AIO cooler.
What does that net us in terms of usable performance over stock?
- POV-Ray
- Stock: 2080.87
- 5.1 GHz: 2368.04
- Diff: +13.8%
- CineBench 11.5
- Stock: 10.66
- 5.1 GHz: 12.10
- Diff: +13.5%
- Handbrake
- Stock: 296.61
- 5.1 GHz: 345.46
- Diff: +16.4%
Overclocking might be considered a dying art by many, but gaining an additional 13-16% in performance for highly threaded applications with very little effort (and maybe a bit of extra cost for a better cooler) is hard to argue with. Morry will be following up with a more detailed look at overclocking with a higher end water cooling setup next week.
And there's just something about being able to run above that 5.0 GHz mark that makes me smile.
When you specify the bitrate
When you specify the bitrate of something in ‘mbps’ the reader can obviously infer that the ‘m’ stands for ‘mega’ instead of ‘milli’, but has no way to tell if the ‘b’ stands for ‘bits’ or ‘bytes’ seeing as you didn’t bother to capitalize the previous letter even though it was appropriate.
Little ‘b’ is bits.
Little ‘b’ is bits. Always.
Mbps = Megabits/sec
MBps = Megabytes/sec
if you had actually read the
if you had actually read the comment you replied to, you would have felt stupid
People are generally a little
People are generally a little more careful with the b since it can have two meanings. The m doesn’t really, since there is no such thing as a millibit. Bits are not divisible by definition. You see mbps quite often, even when Mbps is meant. You don’t see mBps very often though.
Kaby Lake: It’s mostly not
Kaby Lake: It’s mostly not worse!
Hope you can get your hands on Zen soon. At least that should be different enough to make things interesting.
yay zen… less performance
yay zen… less performance but its AMD so it different
Anything to make Intel
Anything to make Intel actually have to consider their pricing and have some competition would be quite worth it.
Minor nit – the TrueCrypt
Minor nit – the TrueCrypt results on the clock-for-clock page are labeled MB/s at the side but GB/s at the bottom.
Ah thanks!
Ah thanks!
Kaby Lake in summary: womp
Kaby Lake in summary: womp womp
So its pretty much official.
So its pretty much official. Intel has shown they don’t need to do anything with CPU’s anymore. Outside of the HEVC decode Intel took Skylake and raised the CPU number. Just pathetic Intel…
Whats pathetic that they made
Whats pathetic that they made improvements every year? AMD is so impressive with there 40% on 5 year old architecture. This thing is still going to be faster than zen in 95% of the applications everyone runs but its pathetic because its not a huge gain?
NOW it’s official? How many
NOW it’s official? How many years after Sandy Bridge did it take to realize this, lol.
Love the combined video and
Love the combined video and text article – thanks for doing this PCPer!. The 5.1 GHz stable for benching is awesome..
I think the 8-core Zen and the upcoming 6-core Coffeelake will give a really good reason for us 2600K holdouts to upgrade, although 7700K looks like a pretty big out of the box difference now..
It depends on the segment
It depends on the segment (little core) where Intel can increase the IPC and also performance levels for these new HEVC encode/decode blocks is set to allow for 4K 120mbps real-time on both the N-series (6 watt) and J-series (6 watt) 14nm Apollo Lake processors (Goldmount CPU microarchitecture).
source: https://www.youtube.com/watch?v=fHa10EbyKTk
Glad to see the 7700k isn’t a
Glad to see the 7700k isn’t a heat monster like the previously leaked info. Still worth picking up if building a new system though. I will hold onto my OCed 4790k for awhile longer….
The temperature problem was
The temperature problem was legit however its not a problem with the cpu but with some of the z270 boards automatically increasing voltage to unreasonable levels out of the box resulting in 90-100 degrees temps.
What do you run on your OC’ed
What do you run on your OC’ed rig? Overclocking just doesn’t seem worth it anymore. Overclocking your GPU would make a bigger difference for most uses.
So does the processor still
So does the processor still do hvec decoding if you have a discrete graphics card or does it just get shut off?
Hmmm…on second
Hmmm…on second thought…good question.
10bit VP9 decode test?
10bit VP9 decode test?
Looks like Intel was only
Looks like Intel was only targeting mostly power usage on Kaby and that HEVC hardware encode/decode block helps there but there is nothing new for the gamer that mostly uses the GPU for gaming and encoding/decoding IP. So this Kaby lake refresh looks to be targeting laptop usage more than desktop usage. I’m not very happy about the market for laptops still being dominated by that Thin and Light nonsense as it’s impossible to find many Business laptops that are using quad core i7 HQ SKUs in some retail(Brick and Mortar) stores. I’d rather have a discrete GPU in my laptop and a quad core CPU but that will change once the Ryzen/Vega APU variants are on the market with much better integrated graphics and HBM2.
Well one good thing with Kaby lake out Skylake SKUs will maybe go down in price! Ryzen is going to eat Intel’s lunch with AMD’s price/performance metrics!
AMD will just eat their lunch
AMD will just eat their lunch and no one can stop them… not even a price adjustment.
Intel has a lot of corporate
Intel has a lot of corporate blubber that has the get paychecks and benefit packages. Also Intel has loads of chip fab infrastructure to maintain at the costs of billions of dollars per quarter that will prevent Intel from having the pricing latitude that AMD has. AMD is so lean that it has no fat or chip facilities to drain its balance sheets with uber expensive physical plant/fab R&D costs or any of the unneeded management Blubber’s pay/benefits/stock options expenses. Intel also has an investor base that will not stand for any disruption(long term) of those mad dividends and such before they drop Intel and look for greener higher margin producing pastures.
AMD can only go up in market share while Intel will find it very painful do go down on its traditionally high margins to defend that market share, especially with the massive size of the non productive Bloat getting a fat paycheck/stock options/costly benefits in the Intel corporate organization!
you should put all your money
you should put all your money in AMD stock and then go fuck yourself
Damn. If you had offered that
Damn. If you had offered that sage advice a year ago I’d have nearly pentupled my money and have had a great deal of joy in the meantime.
Well AMD is a very low
Well AMD is a very low overhead operation and Ryzen has angered you somehow, just look at this review’s Price/Performance(Performance Per Dollar) listing and AMD will look even better with Ryzen! Sounds like you have too much invested in Intel’s stock but that may not be such a wise investment going forward!
One must understand that AMD has been paired down to mostly productive engineers over the last few years and now that lean operation is geared to generate some great revenue growth at an even lower price point than that obese Chipzilla can manage. Look at all that money that old Chipzilla wasted on Contra Revenue and AMD engineered Zen/Ryzen and Polaris and Vega and HBM/HBM2 on not even one tenth of what Intel wasted on Contra Revenue alone! What an obese dinosaur that Chipzilla has become!
YOU’RE SUCH A STUPID FUCKING
YOU'RE SUCH A STUPID FUCKING POOR PEASANT WHAT A FUCKING PATHETIC FUCKING LIFE YOU MUST LIVE…
Leaving this for the sheer
Leaving this for the sheer brilliance of the unintentional irony.
Since we are hitting
Since we are hitting limitations with process technology, I expect the rest of the industry to catch up soon. Part of Intel’s lead has always been process technology, but with everyone hitting the same barriers, I would expect all of them to be approaching a similar point. We might see quite a few competitors with processors to rival Intel’s best. There is going to be a lot of competition from ARM variants and possibly from Power architecture. IBM’s power processors perform quite well, they just have been limited to high end market segments. I wouldn’t be surprised if Apple gets quite close to Intel and AMD x86-64 performance with their mobile ARM processors in in the next year or two.
I am waiting to see what AMD has the offer with Zen. The rummors are making it sound quite good, but I am waiting for real benchmarks. For the average consumer, even enthusiast, the processor is just not that important anymore though. What specialized hardware is included is more important important. A higher end Core 2 Duo will still perform very well for most task, especially the 4 core variants with larger caches. These old CPUs will not be able to play 4K video, HEVC, or other compute intensive codecs. The inclusion of new codecs might be enough for people to upgrade eventually, but it isn’t really compelling.
I still have an old i7 920 with 24 GB of memory and an intel DC 3500 SSD (old Dell 435t/9000 I picked up when my laptop died). It works quite well for most of the things I use it for. It can compile quite fast with 8 threads and quite a bit of cache. What it does need is a new video card; it has an old 8600 GT or something with only 256 MB of memory. It can play video okay, but software support is going away. It was working fine with the previous version of vlc (I run Linux), but it crashes with the latest due to lack of video memory. I might just add a modern video card and see how it does. Flash video in Firefox plays terrible on it, but a new video card might not fix that. HTML5 from YouTube does fine though. If Zen is a good deal, then I may just build a completely new system. They shouldn’t be too cheap though. AMD needs the make some profits. I might be willing to go up to $500 if it performs well.
Are there any issues running
Are there any issues running Windows 7?
Good question, I almost
Good question, I almost forgot about Windows 7 and Kaby Lake.
Comparing how kaby performs
Comparing how kaby performs on Win 7 vs Win 10 would be a great article! Hint, hint.
What about thermals?
What about thermals?
How did you manage 70C temps,
How did you manage 70C temps, when other sites are reaching 90C+ when overclocked?
Because just stating voltage
Because just stating voltage is too vague. It matters what program you’re running on the chip.
Motherboard UEFI / BIOS
Motherboard UEFI / BIOS revisions matter due to auto voltage tuning.
So a GPU update that had to
So a GPU update that had to be in hardware and less errors in bit by bit transmissions? That is kinda weak to send out a new CPU but fixing leakage is good for when they get smaller any leakage will be a even bigger issue. I just hope that Intel gets their prices for desktop CPUs under control so the next time i buy a CPU it does not cost more than the rest of the computer combined.
We really need graphene to
We really need graphene to start showing some promise now, silicone has hit a brickwall.
In case you didn’t know, it’s
In case you didn’t know, it’s silicon, not silicone. Those are two quite different materials.
Silicone Valley is a totally
Silicone Valley is a totally different place also.
While AMD always overhype and
While AMD always overhype and underdeliver I find it hard to believe any AMD solution until we see them in real life. That they will impact the market, such when the original 64 bit opteron hit the market, is not likely in my opinion.
Having seen Samsung Laptops with A8? or was it A10?-level processor does not make me confident their next product will do better than intel unless for a very particular workload.
Intel has clearly hit a brick wall for the past years. Even my 7 year old T400 “toy” laptop has a 2.8GHz dual core mobile processor, and yes it may draw more electricity than a current CPU, but all this 9-15W focus has made many laptops unusable after Ivy Bridge. You need macbook pro with i7 to get a higher clocked dual core processor in 2016/2017!
Yes, embedded graphics have improved, this might make me upgrade eventually when my next TV is fully capable of 4k input signals. AMD clearly has a lead over Intel here. But for CPU.. blah..
I find it strange the delays facing Intel the past years in getting Broadwell and Skylake products onto the market. Hopefully they will have full availability with Kaby Lake.
Question: What is the 14+nm
Question: What is the 14+nm fabrication node? I’m unable to find information about it online. I’m curious how it differs from 14nm (non plus).
“Other than the
“Other than the aforementioned Z270 chipset, Kaby Lake will be the first platform compatible with Intel Optane memory. (To be extra clear, I was told that previous processors will NOT be able to utilize Optane in its M.2 form factor.)”
So it is an m.2 form factor that isn’t compatible with m.2 slots? Or is it just that you can’t use it as a cache without the latest and greatest? I guess a 16 GB drive isn’t that useful without utilizing a caching system to make it invisible. A hybrid drive seems like a better way to go, but that may not fit on a single m.2 easily. It would need to be u.2 or an actual pci-e card.
Great review. I would like
Great review. I would like to see some Adobe Photoshop CC and Bridge CC performance numbers, including the clock for clock comparison. The slow stuff measured would help, such as Smart Sharpen, Unsharp Mask, opening large files in RAW, and generating full screen size previews for hundreds of large files in a folder. With the clock for clock testing you could gauge whether serious overclocking of a 6850X or 6900X might be the way to go for a new Photoshop/Bridge rig would be the way to go, as opposed to the new 7700K.
I think Photoshop is a great test for high end CPUs. It’s just a shame that Intel’s extreme chips with more than 4 cores are one or more generations behind.