Overclocking Intel’s Core i7-7700K: Kaby Lake Hits The Desktop!

[ledhead11]

The chipset hasn't been involved in memory support since the first Core i-series. If you look at chipset specs since Intel integrated the memory controller in their CPUs, chipsets support no memory whatsoever.

Sow how does that fit in answering my question here?

Intel has very tight restrictions regarding approved parts, that are meant to protect itself from liability. Anything you do outside of that restriction is technically considered overclocking. For example, I have a four-DIMM set of DDR4-3866, I put it in the motherboard and set XMP, and the motherboard doesn't boot. The motherboard manufacturer did everything they could to make sure it worked, and so did the memory manufacturer. The CPU might work with it, or it might not.

Obviously this doesn't apply to DDR4-2666 in most cases, yet like I said, Intel is very strict with its approvals for a reason.

That actually explains a lot of what I saw when trying to research 2011 & 2011 v3 MOBO's. I came across numerous manufacturer specs on what each board supported but then also conflicting stories of what of users actually used. In the end I just played it safe for my X79 and stuck with 2133mhz and no problems other than setting BIOS to the right XMP. It even slightly OC'd the RAM to 2172.

 

[Crashman]

The chipset hasn't been involved in memory support since the first Core i-series. If you look at chipset specs since Intel integrated the memory controller in their CPUs, chipsets support no memory whatsoever.

Sow how does that fit in answering my question here?

Intel has very tight restrictions regarding approved parts, that are meant to protect itself from liability. Anything you do outside of that restriction is technically considered overclocking. For example, I have a four-DIMM set of DDR4-3866, I put it in the motherboard and set XMP, and the motherboard doesn't boot. The motherboard manufacturer did everything they could to make sure it worked, and so did the memory manufacturer. The CPU might work with it, or it might not.

Obviously this doesn't apply to DDR4-2666 in most cases, yet like I said, Intel is very strict with its approvals for a reason.

That actually explains a lot of what I saw when trying to research 2011 & 2011 v3 MOBO's. I came across numerous manufacturer specs on what each board supported but then also conflicting stories of what of users actually used. In the end I just played it safe for my X79 and stuck with 2133mhz and no problems other than setting BIOS to the right XMP. It even slightly OC'd the RAM to 2172.

You can check in here, most DDR4-2400 is safe on most X99 boards, JEDEC even approved it but the foot dragging that always happens put X99 on an earlier approval revision. Still, most DDR4-2400 is XMP (or XMP plus standard JEDEC settings).

And most Z170s are safe with recent DDR4-3200 samples. The early ones were a little more hit-or-miss. You could certainly run DDR4-2400 on those as well without taking that small risk, or ask someone about their experience.

 

[Ewitte]

I think I'm done for now I recently replaced a 4790K with a 6700k. It gave me 10% performance (at may 24/7 overclock), native m.2 and better temperatures (much my 4790k was hot). Not sure where to go from here because there I'm just fine sticking with PCIe 3x4 SSD drives and just normal SATA drives for bulk storage. Optane even if it is PCIe 3x8 its shining performance will come from 4k QD1 performance which is currently only 50ish MB/s.

I think the only thing that would push me to upgrade is reasonably priced 6-8 core setups.

 

[anbello262]

I think I'm done for now I recently replaced a 4790K with a 6700k. It gave me 10% performance (at may 24/7 overclock), native m.2 and better temperatures (much my 4790k was hot). Not sure where to go from here because there I'm just fine sticking with PCIe 3x4 SSD drives and just normal SATA drives for bulk storage. Optane even if it is PCIe 3x8 its shining performance will come from 4k QD1 performance which is currently only 50ish MB/s.

I think the only thing that would push me to upgrade is reasonably priced 6-8 core setups.

I don't really understand why you would be looking for an upgrade right now.
Even more, may I ask why you upgraded on the first place?

 

[littleleo]

I think I'm done for now I recently replaced a 4790K with a 6700k. It gave me 10% performance (at may 24/7 overclock), native m.2 and better temperatures (much my 4790k was hot). Not sure where to go from here because there I'm just fine sticking with PCIe 3x4 SSD drives and just normal SATA drives for bulk storage. Optane even if it is PCIe 3x8 its shining performance will come from 4k QD1 performance which is currently only 50ish MB/s.

I think the only thing that would push me to upgrade is reasonably priced 6-8 core setups.

I don't really understand why you would be looking for an upgrade right now.
Even more, may I ask why you upgraded on the first place?

Oh stop that we need more if not all to upgrade except for me, lol.

Ewitte I got to admit I was surprised you got a 10% boost upgrading from the Haswell to Skylake. I would expect a single digit improvement, was that because you got a higher O/C on the Skylake than the Haswell?

 

[Olle P]

It's come to my attention that the key difference in performance between Skylake and Kaby Lake wasn't even tested here. It's the iGPU!
For a fair comparison the interesting benchmarks are those done without a discrete graphics card.

Now that the test results are in I found this to be false.
The only difference lies in support for additional features.

Sow how does that fit in answering my question here?

Intel has very tight restrictions regarding approved parts, that are meant to protect itself from liability. Anything you do outside of that restriction is technically considered overclocking.

My question in the other thread is all about the technical aspects.
* What clock controllers/where are involved?
* What does it take to make it possible to set them at speeds above what's officially "supported"?
* At which interface(s) can one expect to run into trouble while trying to maximize the RAM speed?

Like:
- Do you need a "K" version CPU to "overclock" the RAM?
- Can you do it with a "B" or "H" type chipset?

 

[Crashman]

My question in the other thread is all about the technical aspects.
* What clock controllers/where are involved?
* What does it take to make it possible to set them at speeds above what's officially "supported"?
* At which interface(s) can one expect to run into trouble while trying to maximize the RAM speed?

Like:
- Do you need a "K" version CPU to "overclock" the RAM?
- Can you do it with a "B" or "H" type chipset?

Those are great questions! Since I only have K CPUs, I don't have complete answers. The ratios built into the memory controller go up to 15.5x. The memory controller can run at 5/4 the CPU's BCLK. Double that to get the answer, and the CPUs limit is DDR4-4133.
Someone with a non-K CPU can tell you what's available with those, but I believe the nature of the lock is in the core firmware. I think H-series chipsets instruct a limit of DDR4-2400 and below, again via core firmware, but maybe Joe Trott can tell you more?

 

[Onus]

It seems to be worse; on H170 and H110 boards I've tested (and a B150) I've set RAM higher, e.g. using XMP settings, and it gets set to DDR4-2133.

 

[Crashman]

It seems to be worse; on H170 and H110 boards I've tested (and a B150) I've set RAM higher, e.g. using XMP settings, and it gets set to DDR4-2133.

Were you trying to set up DD4-2400 by XMP? Because if the platform limit were 2400 and the RAM were something higher, it would jump back to default

 

[Onus]

Hmmm, it was the Crucial I was sent; might be higher. I think XMP was the only place to adjust on those boards though; you couldn't set a data rate. You might ask Eric what he sees for options on the Z-boards he's testing.

 

[Olle P]

I've got a Core i5-6600K and MSI Z170A-Pro.
* The range of possible RAM settings available seems limited by the UEFI.
* Default settings (as well as XMP settings) are defined by the RAM stick in use.
* The UEFI allows me to try other settings than those "approved" by the RAM.
* From within the UEFI I see absolutely no connection between the CPU's controller and the RAM settings.

I'll have to take a look at my daughter's computer to see what difference it makes to use a Core i3 with an H-type chipset.

 

[Crashman]

I've got a Core i5-6600K and MSI Z170A-Pro.
* The range of possible RAM settings available seems limited by the UEFI.
* Default settings (as well as XMP settings) are defined by the RAM stick in use.
* The UEFI allows me to try other settings than those "approved" by the RAM.
* From within the UEFI I see absolutely no connection between the CPU's controller and the RAM settings.

I'll have to take a look at my daughter's computer to see what difference it makes to use a Core i3 with an H-type chipset.

What do you mean by "* The range of possible RAM settings available seems limited by the UEFI."? It seems to contradict your statement "* The UEFI allows me to try other settings than those "approved" by the RAM."

 

[InvalidError]

* The UEFI allows me to try other settings than those "approved" by the RAM.

Nearly all retail DIY motherboards from the past ~30 years allow manual overriding of DIMMs' SPD/XMP default profiles, nothing new there.

 

[Olle P]

What do you mean by "* The range of possible RAM settings available seems limited by the UEFI."?

Comparing different motherboards with the same chipset they list different (theoretical) "maximum" RAM speeds.
Since they all use the same controller hardware (CPU and chipset) the differences must(?) be defined by those programming the UEFI. Right?

* The UEFI allows me to try other settings than those "approved" by the RAM.

... nothing new there.

Not new, but still a function that do exist and seems relevant to the context.

 

[Crashman]

What do you mean by "* The range of possible RAM settings available seems limited by the UEFI."?

Comparing different motherboards with the same chipset they list different (theoretical) "maximum" RAM speeds.
Since they all use the same controller hardware (CPU and chipset) the differences must(?) be defined by those programming the UEFI. Right?

Oh no, not at all. You'll see two boards, one advertised as having DDR4-3866 capability and the other as having DDR4-4000 capability, but when you go into UEFI you find they both have multipliers up to 4133. That's because 4133 is the highest data rate the CPU's memory controller can set.

Half of this is marketing: Asus is going to want people to think they're giving them something for all the extra money they spent on a higher-model board.

The other half is an overclocking test. Brand A may have had a set of DDR4-3866 and a set of DDR4-4000 on hand and determined one was stable at the DDR4-3866 XMP settings, the other was stable at the DDR4-4000 XMP settings. I don't even trust these types of ratings because my own tests show things like cheating on XMP voltage (1.37 to 1.38V instead of the rated 1.35V), that a board that works with one set of DDR4-4000 might not work with another set of DDR4-4000, and that there can even be 100MHz or so difference between two boards produced on the same production run.

So the best thing to do is ignore whatever a manufacturer says about overclocked speeds and read a couple reviews instead. The actual limit is signal integrity, the settings are available whether the board is stable or not.

And I've actually tested a few cheaper boards that overclocked DRAM better than "higher rated" expensive boards from the same brand, so jump back to what I said about marketing.

 

[Onus]

Marketing droids are the bane of everyone's existence, which is probably one reason why the Moderators so enjoy squishing spammers, and as quickly as possible.