Article Thumbnail

Haswell vs. Skylake-S: i7 4790K vs i7 6700K

Written on August 5, 2015 by Matt Bach
Share:
Table of Contents:
  1. Introduction
  2. CPU Specifications
  3. Test Setup
  4. Thermal Output
  5. Power Draw
  6. CPU Performance - Unigine Heaven Pro 4.0
  7. CPU Performance - PCMark 8 Professional
  8. CPU Performance - Geekbench 3
  9. CPU Performance - Cinebench R15
  10. CPU Performance - POV-RAY / Linpack
  11. CPU Performance - Lightroom CC
  12. CPU Performance - Photoshop CC
  13. CPU Performance - Premiere Pro CC
  14. Conclusion

Introduction

If you are not familiar with Intel's CPU launch cycles, they operate on a "Tick-Tock" schedule that designates whether a new CPU launch either has a die shrink (with a smaller manufacturing process) or a completely new microarchitecture. Skylake-S is a "Tock" which means that it uses a new microarchitecture which in turn means a different physical socket from the previous generation. Because of the new socket, if you want to use a Skylake-S CPU you will require a new motherboard (with a compatible chipset). At launch, the only chipset available is the Intel Z170 although more models as expected in the future. 

Skylake-S introduces a number of changes compared to Haswell that makes it fairly attractive as a platform. Probably the biggest improvement is that both the CPU and Z170 chipset now support DDR4-2133 RAM. DDR4 is still more expensive than DDR3 but it is slightly faster and should allow for twice the density. This means that instead of being limited to 32GB of RAM, with Skylake-S you can now have 64GB of RAM. There are also a number of smaller changes in the Z170 chipset (more information can be found in our Z97 vs Z170: What is the Difference? article) but none are quite as significant as the move to DDR4.

In this article we want to go over the new Core i7 6700K CPU and look at how it is different from the Core i7 4790K. To do this, we will first examine the specifications of Haswell and Skylake-S CPUs in general then run a wide variety of benchmarks to find out what applications benefit from the move to DDR4 RAM and the other architecture changes.

If you do not want to read this entire article, we suggest reading through the CPU Specifications below then skipping to the Conclusion section.

CPU Specifications

Since only the K-series CPUs have been launched so far, there are some gaps in what we officially know about the specifications for all the Skylake-S CPUs. However, we can make a few reasonable assumptions about the non K-series CPUs such as all i5/i7 CPUs having 4 cores and only the i7 CPUs having Hyperthreading support. Other specifications like the TDP (or thermal output) we cannot make a reasonable assumption so we will simply leave those blank for now.

In addition, we are only going to compare the i5 and i7 CPUs. The Core i3 CPUs are simply so different in terms of specs that it is really better to think of them as an entirely different class of CPUs altogether.

  Haswell i5/i7 Skylake-S i5/i7
Socket: 1150 1151
Manufacturing Process: 22nm 14nm
Cores: 4 4
Hyperthreading: i7 only i7 only
Smart Cache: 6MB (i5)
8MB (i7)
6MB (i5)
8MB (i7)
Bus Type: DMI 2.0 (5GT/s) DMI 3.0 (8GT/s)
Integrated Graphics: HD Graphics 4600 HD Graphics 530
TDP: 84 W (Standard i5/i7)
88 W (K-series i5/i7)
? (Standard i5/i7)
95 W (K-series i5/i7)
Memory Type: DDR3/DDR3L 1333/1600 DDR4 2133 / DDR3L 1600
Max Memory: 32GB 64GB

As you can see, there is quite a bit different in the new Skylake-S CPUs. Starting from the top, the major changes are:

  • New Socket 1151 - Although the socket name is only one number higher than the Haswell Socket 1150, these two sockets are not cross-compatible. In other words, you cannot use a Haswell CPU in a socket 1151 motherboard or a Skylake-S CPU in a socket 1150 motherboard. However, the heatsink mounting is the same so any socket 1150 (or socket 1155/1156 for that matter) heatsink will work just fine on socket 1151.
  • Smaller manufacturing process (14nm) - Technically, the Broadwell CPUs were Intel's first 14nm CPUs, but Intel only ever released a handful of desktop Broadwell CPUs and they were honestly very underwhelming. However, this does mean that this manufacturing process is already a bit established which should help with both supply and reliability. 
  • Faster connection between the CPU and motherboard - With Skylake-S, the bus connection between the CPU and motherboard has been improved to DMI 3.0 which is theoretically about 40% faster than DMI 2.0. While this means that the CPU can communicate with the chipset (and all the USB, LAN, SATA, etc. ports that go through the chipset) faster, the main advantage it brings is simply in being able to put more devices on the chipset without creating a bottleneck. This is the main reason why the Z170 chipset has 20 PCI-E 3.0 lanes versus the 8 PCI-E 2.0 lanes that are on Z97. More information on this is available in our Z97 vs Z170: What is the Difference? article.
  • Improved integrated graphics - If you don't need a discrete video card, the Skylake-S CPUs have much improved onboard graphics compared to Haswell. 530 may be a smaller number than 4600, but we saw more than a 25% increase in 3D performance when using the onboard graphics. In addition, the HD Graphics 530 supports newer feature sets such as DX12.
  • Higher TDP for K-series CPUs - While we don't know yet what the TDP will be for the standard Skylake-S CPUs, we do know that the K-series are rated to need 7W more than the Haswell K-series CPUs. This higher power draw should result in hotter CPU temperatures, although as we will show later we actually found that Skylake-S CPUs actually run cooler than Haswell CPUs. We have a few theories as to why which you can find in the Thermal Output section of this article.
  • DDR4 support - Probably one of the biggest changes in Skylake-S is the addition of DDR4 support. DDR4 memory is slightly faster and uses less power than DDR3, but probably the biggest advantage to DDR4 is that it support twice the capacity (or density) as DDR3. So while both Haswell and Skylake-S can only support four physical sticks of RAM, Skylake-S supports twice the amount of memory as Haswell (64GB vs 32GB).

Test Setup

In order to get an accurate comparison between Haswell and Skylake-S, we decided to test the top CPU from each family. The Core i7 4790K and the Core i7 6700K are very similar in terms of specifications although there are a few small differences:

  Core i7 4790K Core i7-6700K
Cores(Threads): 4(8) 4(8)
Base Frequency: 4.0 GHz 4.0 GHz
Single Core Turbo Frequency: 4.4 GHz 4.2 GHz
All-Core Turbo Frequency: 4.2 GHz 4.0 GHz
Smart Cache: 8 MB 8 MB
Integrated Graphics: HD Graphics 4600 Graphics 530
TDP: 88 W 95 W

Technically, the Core i7 4790K Haswell CPU is faster in terms of frequency with a maximum Turbo frequency of either 4.4 or 4.2 GHz depending on if you only have a single core or all the cores active. This is only .2 GHz higher than the Core i7 6700K CPU but it is enough that in terms of frequency the i7 4790K should be about 4.5% faster than the i7 6700K. However, the addition of DDR4 support and other small changes to the CPU architecture (such as the smaller manufacturing process and DMI 3.0) may be enough to make the i7 6700K faster than its predecessor.

For the rest of our test systems, we used the following hardware:

This is a pretty standard setup that we use for a lot of our testing and should be great for showing the performance, thermal, and power draw differences between the two CPUs. You will notice, however, that we did not use the freshly launched Windows 10 as our test OS. We actually started our testing on Windows 10, but we saw some very strange performance inconsistencies (which we are attributing to minor software and driver bugs) that made us revert to the older, more established Windows 8.1 operating system. 

We will be doing limited testing using the onboard graphics to see how it performs but for the majority of our testing we will be using a GeForce GTX 980. However, we did find that for CPU intensive tasks using onboard graphics did not significantly change the performance, temperature, or power draw of the CPU.

Thermal Output

To start our testing, we decided to see how much hotter (or cooler) the i7 6700K run compared to the i7 4790K. We performed this testing in two ways: once with the CPU fan set to run at 100% and once with it set to Asus' Standard QFan profile which ramps the fan based on temperature.

 

 

Even though the i7 6700K is rated to use more power than the i7 4790K, we saw anywhere from a three to seven degree drop in temperature depending on the benchmark we ran. Honestly, this is exactly the opposite of what we expected. There has been some changes to how power is supplied to the CPU which may be what is causing the lower temperature (such as the majority of the voltage regulation being moved from the CPU onto the motherboard) but no matter the reason this is a great surprise.

Due to the lower temperature, we also saw a small - but measurable - drop in the CPU fan speed when using QFan. It was only a drop of 50-100 RPM, but if you want a silent system like our Serenity or Serenity Pro that small drop in fan speed is actually fairly significant.

Power Draw

Since the i7 6700K runs significantly cooler than the i7 4790K you may expect power draw to be lower as well but we found that not to be the case in all situations. Power draw at idle and during POV-Ray was indeed 4-10 watts lower but Prime95 was actually 2 watts higher. In 3D graphics such as Unigine Heaven, we saw lower power draw when using the GeForce GTX 980, but a very large 15 watt increase when using onboard graphics.

Overall, it appears that Skylake-S has a lower power draw compared to Haswell in most situations unless you are utilizing the onboard graphics. In that case, power draw can be significantly higher during 3D applications. 

CPU Performance - Unigine Heaven Pro 4.0

To start our performance comparison we are going to begin with Unigine Heaven Pro. Every game is different so if you are concerned about performance on a specific title you should try to find a benchmark for that title, but we have found Unigine Heaven Pro to be a great indicator of 3D graphical performance.

With a GeForce GTX 980 4GB video card, we saw absolutely no difference in performance between the i7 6700K and the i7 4790K. Most modern games are GPU limited, however, so this does not really mean that neither CPU is better - simply that both CPUs are fast enough to not be the bottleneck in our test system.

When we switched to onboard graphics, however, we saw over a 25% increase in performance. There are a number of different reasons for this, but it is mainly due to the new graphics architecture (HD 530 versus HD 4000) and the move to DDR4 memory. Since onboard graphics uses the system RAM instead of dedicated VRAM like you would find on a video card, using the faster DDR4 RAM results in an increase in onboard graphics performance.

If nothing else, this benchmark shows that onboard graphics is continuing to get better and better. It is still nowhere near the performance of most video cards, but if you were considering purchasing a <$100 video card to pair with Skylake-S you may opt to simply use onboard graphics instead.

CPU Performance - PCMark 8 Professional

PCMark is one of the standard benchmarks that is used in most reviews, but you need to keep in mind that no matter how realistic it tries to be, it is still a synthetic benchmark. With that said, PCMark shows some of the least exciting results from any of our testing. We did still see a small increase in performance with the i7 6700K, but never higher than a tiny .9% increase. 

CPU Performance - Geekbench 3

Geekbench 3 is another synthetic benchmark, although it focuses more on CPU-intensive benchmarks rather than full-system benchmarks. Overall, Geekbench 3 showed an increase in performance of almost 5% with the i7 6700K which is fairly good.

For the individual categories, integer performance saw about a 5% increase in performance while floating point performance only saw about a 3% increase. Memory performance, however, saw a huge 22.2% increase due to the move to DDR4 memory. Most applications are not limited by the speed of your memory, however, so don't expect to see a 20% increase in performance except in rare situations.

One interesting thing we found with Geekbench is that the single core integer and floating point performance of the i7 6700K was actually worse than the i7 4790K. It was not until we looked at the multi core benchmark results that we saw an increase in performance. This suggests that the i7 6700K is not actually any faster than the i7 4790K per clock, but is simply more efficient at multi-threading which results in higher performance in most applications.

CPU Performance - Cinebench R15


Similar to Geekbench, we saw very little increase in performance when only using a single core in CineBench R15. However, even with all the CPU cores active we only say about a 1% increase in performance which is similar to what we saw in PCMark 8. Interestingly, we actually saw about a 1.5% decrease in performance during the OpenGL portion of Cinebench R15. This isn't a very large decrease but it is outside of what we consider to be our margin of error.

CPU Performance - POV-RAY / Linpack

For the last of our synthetic benchmarks we ran POV-Ray and Linpack. Both of these benchmarks are widely used in the scientific community and are usually very good at showing raw CPU performance.

Starting with POV-RAY we saw a very large 7.7% increase in performance compared to the i7 4790K. This is a significant increase in performance and is an indicator that ray tracing in general should be much better with the i7 6700K.

Linpack, however, is a bit of an aberration and to be honest we almost didn't include it because we believe there to be a problem currently between Linpack and Skylake-S. Simply put, we saw a 20% drop in performance with the i7-6700K. We are still investigating why the i7 6700K is giving such low performance (we should be seeing at least 205 GFLOPs, not 173 GFLOPs) but for now all we can say is that the i7 6700K is giving very low performance in Linpack.

CPU Performance - Lightroom CC

Starting off our real-world benchmarks we tested a number of actions in Lightroom CC 2015. If you read our recent     
Adobe Lightroom CC/6 CPU Multi-threading Performance article these actions will be very familiar to you. For importing images, generating smart or 1:1 previews, and converting the images to DNG we saw about a 1.5-2% increase in performance. This isn't great, but it is still noteworthy. Where the i7-6700K really impressed us was when it came to exporting images where we saw a 15% increase in performance. This seemed way too high at first, but we tested and re-tested it and we are confident that the i7 6700K can indeed export images in Lightroom 15% faster than the i7 4790K.

In addition to basic image handling in Lightroom, we also tested the creation of HDR and panorama images. Neither was quite as exciting as the "Export as JPG" results, but we still saw a 1.7% increase in performance while creating panorama images and a 3.4% increase in performance when creating HDR images.

CPU Performance - Photoshop CC

Photoshop saw some of the largest increases in performance out of all the software we tested. Overall, we saw about a 8.5% increase in performance although the increase for each effect ranged from just .8% (Smart Blur) to 16.7% (Motion Blur). Even simple things like image resizing and rotation saw over a 7% increase in performance.

All-in-all, if you are a heavy Photoshop user Skylake is a very attractive platform. Not only does the i7 6700K have significant performance improvements over the i7 4790K, but the move to DDR4 RAM means that you do not need to use the slower (for Photoshop at least) socket 2011 CPUs if you need more than 32GB of system RAM.

CPU Performance - Premiere Pro CC

Unlike Lightroom and Photoshop where we developed our own benchmark scripts, for Premiere we used the popular PPBM9 benchmark. The only change we made was to run a modified version of the MPEG2 Hardware timeline where we copied and pasted the test clip three times to make the render three times as long.

Overall, the only two places we saw a significant increase in performance was with MPEG2 software encoding and H.264 encoding. Honestly, you should always be using MPEG2 hardware encoding (which uses the GPU) so the 7% improvement to MPEG2 software encoding doesn't mean much. The 6% increase in H.264 encoding, however, is very important as H.264 is one of the most popular codecs used today.

Most professional Premiere Pro uses will likely still want to use a socket 2011 CPU due to the higher core counts, but if you are only using Premiere Pro as a hobby or only use it intermittently, the 6% improvement in H.264 encoding time with the i7 6700K is very significant.

Conclusion

To summarize this article, here is what we found in terms of temperature, power draw, and performance of the i7 6700K compared to the i7 4790K:

  • Power Draw - in most situations, the i7 6700K uses about 4-10 less watts than the i7 4790K. The main exception is if you are using the onboard graphics - in which case expect a 10-12W increase in power draw
  • Temperature - Depending on what you are doing, the i7 6700K should run about 4-7 ºC cooler than the i7-4790K. Even when using onboard graphics (where we saw a 10W higher system power draw) the i7 6700K was still about 3 ºC cooler than the i7 4790K.
  • Performance - In most cases, the performance of the i7 6700K compared to the i7 4790K was either identical or significantly better. We did have a few times that we saw a drop in performance (most noteably a 20% decrease in Linpack) but overall Skylake-S is better than Haswell.
i7 6700K vs i7 4790K Performance Summary
Unigine Heaven Pro 4.0 No change with discrete GPU, >25% faster with onboard graphics.
PCMark 8 Pro .5-1% increase in performance.
Geekbench 3 5% increase in multi-core performance.
Cinebench R15 1% increase in CPU performance, 1.5% decrease in GPU performance.
POV-Ray 7.7% increase in performance.
Linpack 12.5% increase in performance. *Updated using Linpack 11.3
Lightroom CC 2015 1.5-2% increase in most image handling tasks, but a huge 15% increase in image export performance. ~2-3% increase in HDR/Panorama image creation performance.
Photoshop CC 2015 8.5% increase in overall performance. Varies anywhere from .8% to 16.7% depending on the effect.
Premiere Pro CC 2015 6% increase in H.264 encoding performance, marginal increase in MPEG2 encoding performance.

In most situations, the i7 6700K should be anywhere from 1% to 10% faster than the i7 4790K.  We saw some of the most significant performance improvements in Lightroom and Photoshop. Most of the Lightroom actions we tested saw a small 1.5-3% increase in performance, but the 15% increase in performance when exporting images is very significant. In Photoshop, some of the effects we tested saw even greater performance improvements - up to 16.7 percent! Even better, the move to DDR4 RAM means that you do not need to use the slower (for Photoshop at least) socket 2011 CPUs if you need more than 32GB of system RAM since the Skylake-S platform supports up to 64GB of RAM.

Overall, Skylake-S as a platform is very, very good. Most of the specification changes are pretty subtle (with the exception of the move to DDR4) but we were impressed with the lower power draw and temperature of the i7 6700K compared to the i7 4790K. Unless you are a professional Photoshop or Lightroom user, the performance improvements likely won't merit upgrading from Haswell to Skylake-S, but if you are in the market for a new system we see little reason to not use Skylake-S.

Probably the worst thing about Skylake-S is that Intel only has two CPUs (the high-end unlocked i5 6600K and i7 6700K) and one chipset (Z170) available at launch. That is expected to change in the near future but if you want to be an early adopter your choices will be limited to only the high-end options.

Tags: Skylake-S, Haswell, i7 4790K, i7 6700K
DisqusMan

What are your thoughts on Skylake vs Haswell-E. There isn't much of a price difference, both would likely require an upgrade to DDR4. Don't really need integrated gfx and the x99 chips have more cores. You've also got more lanes on x99 for m.2 etc. While I'd certainly take z170 over z97 it is the x99 platform that's the alternative option which it would be useful to have a comparison against.

Posted on 2015-08-05 17:15:26

It is basically still the same situation as Z97 vs X99 was. If your application doesn't need / use more than four cores and you don't need >32GB of RAM (soon to be 64GB) then Skylake / Z170 will save money and give options for slightly higher clock speeds. If you have any use for more cores or higher amounts of RAM, though, then Haswell-E / X99 will let you meet those needs while Z170 cannot.

Posted on 2015-08-05 17:29:34
therock

I agree. I run my (idle) compiler suite (RAD Studio XE8), work VM w/ office apps idling, and games with a normal background load (I.E. Chrome with a couple dozen tabs, Outlook, et al.) while gaming and I am usually about 17GB into my 32GB. Probably 1% of workstations have that much crap running. Depending on when you buy your components, you're better off with higher-clocked RAM at a lower CAS timing... but the boost isn't significant and usually not worth worrying about.

Posted on 2015-09-19 17:32:57
Oracles

Well, at least they are shipping the 6700k without a cooler. I have a box in the junk room full of noisy, ineffective Intel coolers just waiting for someone to recycle the stupid things.

Posted on 2015-08-28 19:22:52
David Dewis

Guess im sticking with my Overclocked FX8350 for the time being. It might be old, but in real world performance, its not falling behind enough to warrant the cost of upgrading just yet. When i start to see big jumps in import/export times of 5-10 mins, i'll consider it. but at the moment, all the research ive seen suggests i'll be spending nearly £500-£600 for a few seconds less waiting time. I miss the days where new gen CPUs had massive, real world, jumps in performance.

Posted on 2015-09-07 03:21:01
Hunter Dominic

Don't expect massive increases anytime in the near future. Until some space age, cheap technology is found, the performance increases will be minuscule.

Posted on 2015-10-10 11:07:41
Michael F

If TSMC, Glo Fo, or Micron would start licensing some of IBM's new 9nm tech, we could see some major improvements from AMD. If you can just throw an ungodly amount of transistors at the problem, you can make a faster chip than Intel. Or in the automotive world, a 2.6 liter Mazda GT3 racing engine is never going to make more power than a 12 liter Cummins turbo diesel no matter how optimized the design is.

Posted on 2015-11-03 05:48:22
Ry

Quantum tech apparently isn't that far off. Could be seeing this stuff in home PCs in twenty years?

Posted on 2015-12-12 05:55:46
Gots

Interesting . I'm probably going to build a Haswell machine , as I don't need much more than what the 4790K will do .

Posted on 2015-09-11 05:14:55
Jeansen

Do you use statistical algorithms to calculate the in/significant results or do you just read these numbers and conclude(may be ANOVA?)? Cuz even if there is a 15%increase in Photoshop, I think unless someone uses his computer only for Photoshop, this is a good choice. But for everyone else the skylake don't seem appealing for haswell users I would say.

Posted on 2015-11-09 07:38:54
Pete Farrow

Quote "the i7 6700K compared to the i7 4790K was either identical or significantly better", this is a strange conclusion given the benchmarks above that have just been published. The differences are, at best, marginal. The only offering the Skylake has is 64G RAM over 32G RAM. I see no reason to upgrade. "signifcant improvement" in my book means > 30% faster across the board. Not +10% in some cases and -10% in others. I can't really imagine anyone rushing out to buy a skylake system based on the rather meagre performance improvements demonstrated here. It is such a small difference it is hardly worth reviewing. The internal graphics might be considerably better, but does anyone actually use onboard graphics in a high end system likely to be hosting one of these CPUs? The onboard graphics are a bit like having a really fast super car that has a spare wheel with a better tread pattern on it than the previous models, allowing you to drive a bit faster on your space-saver spare when you have a puncture.

Posted on 2016-01-08 02:16:44
Cat Tilley

Why not have a level playing field & use the DDR3L RAM modules that the 6700K will support, and some OEM's will do just this to purge old stock.

Make the RAM equal, then let's see the outcome of the benchmarks. I'm not impressed, the i7-4790K hold a slight advantage over the i7-6700K on the PassMark site, and nearly a year after release, it's not in the top 10 common CPU's yet.

Plus, the other issue going against it, come 18 months from now, tests running Windows 7 or 8.1 will be irrelevant, as one will need Windows 10 to 'access' all of the Broadwell's potential. This was not just a Intel issue, rather one from (strangely) also AMD & Qualcomm, they've joined forces with Microsoft to ensure that their newest products won't be fully utilized with OS's less than Windows 10. So the playing field will never be level again.

Posted on 2016-02-06 19:57:48
mwc1

by now there are many more motherboard options for the 1511 socket, I am looking at Skylake not as much for performance boost over Haswell but wanting to bypass that dreadful Haswell design hoping better thermal and with the regulator on the mother board to help with CPU temps.
Also looking to go with Xeon and forget the i core crap

Posted on 2016-05-19 00:08:33
Anonymous

Thanks a ton for this article. Definetely going for a 6700K, this article changed my mind. Planning to get a GTX 1080, and 32 gigs ram at 2133 aswell.

Posted on 2016-05-29 14:53:52

Why broadwell (5th Gen) Intel core series processors did not make it to desktop PCs?

Posted on 2016-08-21 12:56:26