Intel Skylake Core i7-6700K IPC & Overclocking Review @ [H]

Agreed the GPU going from 16x to 8x is not that much of a performance hit, but the DMI bus will be a bottleneck for SATA, ethernet, USB and NVMe SSDs together. The DMI 3,0 link is equivalent bandwidth to just 4x PCIe lanes. The new Samsung NVMe drives do 5.5GB/sec reads. That pretty much uses up your DMI 3.0 bandwidth.

I've very well aware how much bandwidth it has. I got the numbers straight from Intel themselves. But my point still stands. You won't normally saturate the DMI bus because you won't be hitting the maximum bandwidth of the network controller, NVMe drive and USB ports. You may do one or two of these at a time on occasion, but not all the time.

USB 2.0 devices are basically inconsequential. Even if you were using USB 3.0 or worse yet, 3.1 devices, I doubt you'd be maxing out their transfer rates. If you connected some type of external storage and again were doing large transfers then maybe. But again this is only a temporary issue. If you look at your network traffic, unless your doing large file transfers, you won't be hitting 1Gbps speeds. I've never seen a controller actually hit that outside of very controlled conditions with specific tuning in the drivers.
 
I've very well aware how much bandwidth it has. I got the numbers straight from Intel themselves. But my point still stands. You won't normally saturate the DMI bus because you won't be hitting the maximum bandwidth of the network controller, NVMe drive and USB ports. You may do one or two of these at a time on occasion, but not all the time.

If you were copying a file over the network of from external storage to the SSD, does the data have to go over DMI to the CPU and then back?

Could most IO from the SSD to the network or usb to the SSD just route through the chipset without touching the CPU?
 
I have drives that swap data between the two often, while I never measured the actual bandwidth usage these drives take up there is definitely a bottle neck going on. I'm first upgrading them to SATA SSD's as that's a fairly cheap solution and putting them on a DMI 3.0 supported platform for future purposes.
 
So, in almost 5 years all we got was a 25% increase in performance?
 
Considering the lack of progress for Storage in the last 3 decades I would say 25% improvement in five years is a decent step. More than I am used to thats for sure.
 
Considering the lack of progress for Storage in the last 3 decades I would say 25% improvement in five years is a decent step. More than I am used to thats for sure.

Three decades ago, in 1985, HDDs were 10MB and SSDs were unimaginable.
 
Three decades ago, in 1985, HDDs were 10MB and SSDs were unimaginable.

Seriously... 23 years ago I couldn't imagine filling a 1GB HDD. I think I had a 150MB HDD at the time. I now have over 4TB of space... There has been lots of jumps in storage
 
Seriously... 23 years ago I couldn't imagine filling a 1GB HDD. I think I had a 150MB HDD at the time. I now have over 4TB of space... There has been lots of jumps in storage

I might have been using a 40 MB drive at that point, but I may be thinking of the computer before that. Either way, in 92 I didn't have tons of storage and I didn't need it either.
 
6700k is way faster in games than any other chip ever made.

Incorrect. 6700K has slightly more IPC than it's predecessors, but nothing that would qualify as "way faster" in any context. The only reason it does as well as it does against a 5820k is because 95% of the benchmarks out there are done at stock speeds where the 6700k has a 600-700Mhz clockspeed advantage over the 5820k. If both are overclocked, that difference decreases to ~200Mhz. That is why most say go 5820k if you plan to overclock.

On the flip side, the 5820k has two extra cores that will come in handy for gaming. I'm not talking about DX12 theories, i'm talking about DX11 games TODAY that are already making use of more than 4 cores, games like GTAV. DX12 or not, it's pretty clear which direction gaming development is heading, and that is toward using more cores. In addition, real world usage involves running other stuff in the background while you game, and unfortunately even something as simple as windows checking for updates in the background can peg a core.
 
Incorrect. 6700K has slightly more IPC than it's predecessors, but nothing that would qualify as "way faster" in any context. The only reason it does as well as it does against a 5820k is because 95% of the benchmarks out there are done at stock speeds where the 6700k has a 600-700Mhz clockspeed advantage over the 5820k. If both are overclocked, that difference decreases to ~200Mhz. That is why most say go 5820k if you plan to overclock.

On the flip side, the 5820k has two extra cores that will come in handy for gaming. I'm not talking about DX12 theories, i'm talking about DX11 games TODAY that are already making use of more than 4 cores, games like GTAV. DX12 or not, it's pretty clear which direction gaming development is heading, and that is toward using more cores. In addition, real world usage involves running other stuff in the background while you game, and unfortunately even something as simple as windows checking for updates in the background can peg a core.

If you look at the DigitalFoundry videos showing 6700k vs previous 3 generations, the skylakes have a comfortable lead. It is around comfy 20 - 25 fps over prev generations while earlier they only had a delta of around 5 fps. They have a lot of games benchmarked.
So 20 fps is a HUGE LEAD in CPU terms imo
 
Incorrect. 6700K has slightly more IPC than it's predecessors, but nothing that would qualify as "way faster" in any context. The only reason it does as well as it does against a 5820k is because 95% of the benchmarks out there are done at stock speeds where the 6700k has a 600-700Mhz clockspeed advantage over the 5820k. If both are overclocked, that difference decreases to ~200Mhz. That is why most say go 5820k if you plan to overclock.

On the flip side, the 5820k has two extra cores that will come in handy for gaming. I'm not talking about DX12 theories, i'm talking about DX11 games TODAY that are already making use of more than 4 cores, games like GTAV. DX12 or not, it's pretty clear which direction gaming development is heading, and that is toward using more cores. In addition, real world usage involves running other stuff in the background while you game, and unfortunately even something as simple as windows checking for updates in the background can peg a core.

do not forget the 6700K have 8 threads. games can use more than 4 threads but no 8 threads actually.. some games require indeed more than 4 threads as you said GTA V or the venerable crysis 3 will be a massive bottleneck with any other i5 chip, but he is partially right, even under DX12 games will not scale beyond 6 real cores which can be suffice to meet by the 8 threads i7, yes they aren't truly 8 cores, but they do the job with good efficiency.

What is it with you and incorrect exaggerated blanket statements?

actually, with actual games, out of the box he is right... way faster no, but considerably faster out of the box than any other chip in the market, now as GotNoRice said once overclocked to similar speed (which its actually hard for the 5820K go passs 4.5ghz without forget the high cooling ability it require to achieve those clocks) they can be kinda tied..
 
If you look at the DigitalFoundry videos showing 6700k vs previous 3 generations, the skylakes have a comfortable lead. It is around comfy 20 - 25 fps over prev generations while earlier they only had a delta of around 5 fps. They have a lot of games benchmarked.
So 20 fps is a HUGE LEAD in CPU terms imo

1. The "previous 3 generations" that you refer to are all quad core chips on DDR3 motherboards. 5820k is hex-core (6 cores), along with having nearly twice as much L3 cache available, quad-channel DDR4, etc. It doesn't appear that "DigitalFoundry" has ever touched a 5820k.

2. Looking at the videos, even though the processors they are comparing the 6700k against are slower than the 5820k, the difference is still ~5 FPS most of the time.
 
even under DX12 games will not scale beyond 6 real cores

This sounds like a wild guess on your part. Do you have anything to substantiate this? If this is based on anything real, I am very interested.

which can be suffice to meet by the 8 threads i7, yes they aren't truly 8 cores, but they do the job with good efficiency.

I think you are over-estimating the impact of HT. Certainly, I would much rather have HT than not, but it's not going to hold a candle to having more actual cores available. 6700k is obviously the way to go if efficiency if your primary concern.

actually, with actual games, out of the box he is right... way faster no, but considerably faster out of the box than any other chip in the market

Even at stock, the 5775c was shown to be pretty much on-par with the 6700k in most benchmarks, if not faster. 4770k and 4790k both did very well also. 5820k has a slower stock clock than either the 4770k or 4790k, for other reasons, so of course the 6700k looks good against the 5820k stock. Anyone who buys a 5820k for gaming and runs it stock is "doing it wrong", to say the least.
 
This sounds like a wild guess on your part. Do you have anything to substantiate this? If this is based on anything real, I am very interested.



I think you are over-estimating the impact of HT. Certainly, I would much rather have HT than not, but it's not going to hold a candle to having more actual cores available. 6700k is obviously the way to go if efficiency if your primary concern.



Even at stock, the 5775c was shown to be pretty much on-par with the 6700k in most benchmarks, if not faster. 4770k and 4790k both did very well also. 5820k has a slower stock clock than either the 4770k or 4790k, for other reasons, so of course the 6700k looks good against the 5820k stock. Anyone who buys a 5820k for gaming and runs it stock is "doing it wrong", to say the least.

im on phone right now, but will be able to post more stuff later, for now I think the fastest i can pick for you is directly from AMD blog DX12 Scaling

This is also a good read for early DX12 bench

Another one from Legitreview that show no scaling over 6 cores (strange that they didn't test 4c + HT)
 
Yeah, I see nothing in the ongoing discussion of this that makes sense of the post that started it all. Why would anyone buying a Skylake for gaming not OC it? And it doesn't sound like OCing the Xeon makes it faster than Skylake, just brings it up to parity. Add in that the X99 ecosystem is more expensive than Z170, and I just don't get it.
 
The real winners here would be the ones going from Sandy to Skylake

Only if you want to hug 60fps or higher at 1080p or dont mind reducing settings/going multi card to prevent being GPU occupied at higher res.
Otherwise higher res has less CPU dependence.

You cant help making false blanket statements.
 
Only if you want to hug 60fps or higher at 1080p or dont mind reducing settings/going multi card to prevent being GPU occupied at higher res.
Otherwise higher res has less CPU dependence.

You cant help making false blanket statements.

what? you said something completely unrelated to my statement.
2600k to 6700k, a couple of users reported 20 fps + gain in minimum fps and they can see the difference in gameplay in newer titles like Project Cars.
Why would someone NOT wanna hug 60fps?
 
what? you said something completely unrelated to my statement.
2600k to 6700k, a couple of users reported 20 fps + gain in minimum fps and they can see the difference in gameplay in newer titles like Project Cars.
Why would someone NOT wanna hug 60fps?
You rarely see the wood for the trees which is why you cant see the relevance.

Most people will reduce settings to prevent being CPU occupied, its easy enough to do in Project Cars.
If you want to keep the settings maxed in PCars then a CPU upgrade will help, and ONLY if you cant get a high overclock on your Sandybridge, AND are running 1080p with a very fast card or multi card at higher res. Not exactly all Sandybridge owners.
If you can overclock to 4.7GHz+ on Sandybridge, Project Cars will hardly see a benefit.

If you dont play Project Cars, you arent going to care one bit about it.
 
Last edited:
You rarely see the wood for the trees which is why you cant see the relevance.

Most people will reduce settings to prevent being CPU occupied, its easy enough to do in Project Cars.
If you want to keep the settings maxed in PCars then a CPU upgrade will help, and ONLY if you cant get a high overclock on your Sandybridge, AND are running 1080p with a very fast card or multi card at higher res. Not exactly all Sandybridge owners.
If you can overclock to 4.7GHz+ on Sandybridge, Project Cars will hardly see a benefit.

If you dont play Project Cars, you arent going to care one bit about it.

Really? I've only got my 2500K @ 4.4Ghz and I'm not having performance issues in Project Cars @ 1080p. It still seems to be limited by my 2x Radeon 280xs. I'm thinking of getting a new CPU next year, more to get a mobo with NVMe support, but I'm sad I don't need one yet.
 
Really? I've only got my 2500K @ 4.4Ghz and I'm not having performance issues in Project Cars @ 1080p. It still seems to be limited by my 2x Radeon 280xs. I'm thinking of getting a new CPU next year, more to get a mobo with NVMe support, but I'm sad I don't need one yet.

Yeah but only if you want 60fps "minimum" with "max" settings.
Its quite hard to become CPU limited but when pushing for the most fluent experience and graphically intense settings, it can be done.

My 2500K at 4.3GHz wasnt fast enough to maintain 60fps with all settings maxed on a clocked 980ti (used DSR 1440p for AA).
I worked out that I needed just under 10% extra CPU to prevent bottlenecking.

What res are you using?
 
Last edited:
Now intel is focused on gaming ????? since when???? with poor intel graphics? par with great processor?

and i am kinda confused because these are basically 5th gen processors????? if im not wrong?

and over all performances looks very promising.

waiting for 6600K review now
 
Last edited:
Does anyone know what is next in line for Intel releasing a new chip in the future? If so when?>
 
Thanks for another great article. Really helped me decide it was time to upgrade my machine from a 4 year old 2600k to a 6700k and retire the old stuff to the living room for HTPC duties.
 
When will intel release the Skylake 6 cores cpu ?

Presumably Q1 2017. Possibly (rumor) they will skip Broadwell-E, due in Q1 2016. and release Skylake E then. But offically the next 6 core processor to be released is the Broadwell E, in Q1, 2016.
 
Presumably Q1 2017. Possibly (rumor) they will skip Broadwell-E, due in Q1 2016. and release Skylake E then. But offically the next 6 core processor to be released is the Broadwell E, in Q1, 2016.

Thank you. Do you the Broadwell E 6core will have price same as 5820k ?
 
No idea, but i expect it to be around the same.

This is why I went Skylake in my upgrade from Yorkfield. I ran the numbers and, to go Z97/DDR3/Haswell would have saved me at most 25% for equal-quality parts. I only upgrade those base components every 4-5 years. Why not be as up-to-date on my infrastructure as possible if I'm going to keep it that long?

The CPU differential in my i5-K bracket was like $240 vs. $260.
 
This is why I went Skylake in my upgrade from Yorkfield. I ran the numbers and, to go Z97/DDR3/Haswell would have saved me at most 25% for equal-quality parts. I only upgrade those base components every 4-5 years. Why not be as up-to-date on my infrastructure as possible if I'm going to keep it that long?

The CPU differential in my i5-K bracket was like $240 vs. $260.

The X99 is pretty comparable to the Z170 platform actually.

DDR4 check.
PCIe 3.0 check
Bootable NVMe check
USB 3.1 check

The only realy difference is the setup of the PCIe lanes.
A 5830K or better provides the best PCIe setup, 40 PCIe 3.0 lanes straight to the CPU.
The 5820K only has 28 PCIe 3.0 lanes to the CPU and 8 PCIe 2.0 lanes on the DMI 2.0 bus.
The Z170 platform has 16 PCIe 3.0 to the CPUE and then 20 PCIe 3.0 lanes on the DMI 3.0 bus.

So while the 5820K and the Z170 have slightly different setup of PCIe lanes, they are really pretty similar.

So with the platform out of the way it comes down to 4 cores + better IPC v 6 cores.
 
The X99 is pretty comparable to the Z170 platform actually.

DDR4 check.
PCIe 3.0 check
Bootable NVMe check
USB 3.1 check

The only realy difference is the setup of the PCIe lanes.
A 5830K or better provides the best PCIe setup, 40 PCIe 3.0 lanes straight to the CPU.
The 5820K only has 28 PCIe 3.0 lanes to the CPU and 8 PCIe 2.0 lanes on the DMI 2.0 bus.
The Z170 platform has 16 PCIe 3.0 to the CPUE and then 20 PCIe 3.0 lanes on the DMI 3.0 bus.

So while the 5820K and the Z170 have slightly different setup of PCIe lanes, they are really pretty similar.

So with the platform out of the way it comes down to 4 cores + better IPC v 6 cores.

Oh, I was mainly riffing off of how little difference there seems to be between CPU generations in prices these days, not so much Z170 vs. X99. Still, since you brought it up, how much will an OC-friendly X99 board plus a 5830K set you back? My 6600K plus an Asus Z170-K was just under $400.
 
I went 5820K which I managed to get for $280 and I picked up a mobo for right around $200.

Not bad considering I get 2 more cores, 8 more threads and giant L3 cache.
 
I went 5820K which I managed to get for $280 and I picked up a mobo for right around $200.

Not bad considering I get 2 more cores, 8 more threads and giant L3 cache.

True words, Obiwan!

Edit: But where did you get a price like that on a 5820K? NewEgg has them at $389.99
 
Back
Top