cancel
Showing results for 
Search instead for 
Did you mean: 

Ivy Bridge-E Easy Overclocking Guide – The Listy/Wordy Edition

Raja
Level 13
Standard disclaimer: Overclocking is not guaranteed to work, or guaranteed to be reliable. Do so at your own risk.



Read the guide through in its entirety before attempting any overclocking!







Ivy Bridge-E CPU Core Overclocking Overview



1) In my binning of 45 samples, 2% of CPUs will do 4.8GHz at 1.40V. 20% will do 4.7GHz. Almost 48% will do 4.6GHz at 1.40V. 28% will do 4.5GHz at 1.40V. 2% will do only 4.4GHz at 1.40V. The average frequency is therefore 4.6GHz on a reasonable CPU sample. Not a massive sample size, but I think it gives us a realistic indication of what to expect from retail processors.



2) The power draw of Ivy-E is very low. We measured 120 Watts at full AVX load at 4.6GHz. However, we’re dealing with a smaller substrate/process node which means getting heat away from the die quickly is key. Same trend we are used to by now on the 22nm process.

4.6GHz at 1.40V is doable using water cooling – we get loaded temps of around 80C on a good triple radiator water loop. For users with air coolers, we recommend a maximum of 1.30V for Vcore - of course this is dependent on the type of air cooler and its capabilities. Use a voltage that keeps loaded temps below 75 Celsius or so.

We do recommend placing a fan over the VRM heatsink if the system is going to be overclocked.




Overclocking Ivy-E for the masses is centered on two voltages: Vcore and VCCSA (more on VCCSA later).

(i) For Vcore the highest we’re using here internally is 1.40V (cooling dependent obviously).

(ii) Our auto rules will set Vcore and VSSA as changes are made. We recommend starting on auto and simply setting the multipliers, then tuning voltages lower or higher depending upon the CPU sample.

That’s most of the basics in short-form apart from specifics about Ivy-E and memory overclocking. Before we get to the overclocking section, there’s a memory kit related topic that comes up on these forums (and others) frequently. It seems many users are not aware that combining memory kits – even those of the same model can lead to system instability. Make sure you read the section below before proceeding to overclock the system.




Notes on X79 memory purchasing



Let’s start with what not to do: DO NOT COMBINE MULTIPLE MEMORY KITS, EVEN IF THEY ARE THE SAME MODEL. While it may seem attractive to populate as many memory slots as we can, there is a sensible way of doing it that makes setting the system up and getting it stable easier. Purchase a single kit rated at the desired density and timings you wish to run – we recommend memory kits rated no faster than DDR3-1600 for ease-of-use. Speeds faster than DDR3-1600 require an IMC that is “strong”, and may require lots of manual tuning to obtain stability.



Quick check list of Ivy-E memory related items:


1) DDR3-1866 is supported as a stock speed. However, it is only supported with one DIMM per channel. If all slots are populated, then the maximum supported memory speed is DDR3-1600. Both situations are likely subject to memory timing, too. Intel usually makes this information available in their white papers.

Supported speeds are those the processor can run without needing any kind of voltage adjustment. Anything faster than this is defined as overclocking and quite obviously, may require adjustment of voltages and memory timings to work.

The idea then is to select a memory kit that suits one’s mindset. If plug-and-play operation is desirable, then it would be wise to select memory kits rated no faster than DDR3-1600 or DDR3-1866 (depending upon density).

If one is inclined to tune a system manually, then purchasing a faster memory kit is possible. Bear in mind that performance gains from running faster memory are minuscule in most applications. If budget is constrained, or you’re simply looking for good performance return for expenditure, then spending excessively on fast memory kits is ill-advised.



2) Look for memory kits that are qualified to run on the platform they will be used in. In this case, we are looking for X79 qualified memory kits. Memory kits qualified on other platforms may not reach their specified timings when used on X79 and vice versa, too. That’s because the SPD and XMP is programmed with memory timings for the platform they were binned on.


3) We’ll say it one more time: Don’t combine multiple kits to make up a higher density. Purchase a single memory kit rated at the speed and required density (other purchasing advice withstanding).



Following the advice above should keep most out of trouble. If inclined towards memory related tomfoolery (overclocking or tweaking) then the following section is worth a read.






Notes about Ivy-E's Memory Controller


It’s time to start writing about about Ivy-E’s memory controller. On average, Ivy-E is a little more robust than Sandybridge-E when it comes to memory overclocking. Higher memory ratios are available (over DDR3-2400) and the CPUs seem to have an easier time reaching DDR3-2400.




I like lists, they keep things on point without BS. My (naive?) belief is that people are more likely to follow lists than large paragraphs of text. Yep, it’s time for another list. This one’s about Ivy-E’s memory controller capabilities:


1) Most Ivy-E CPU samples are capable of reaching DDR3-2400 with 32GB of memory.


2) Some samples are capable of speeds beyond DDR3-2666 with 16GB of memory.


3) As CPU core frequency is increased it can become more difficult to reach memory speeds over DDR3-2133. This situation is also affected by the amount of memory used. For example, it may be easier to run speeds over DDR3-2133 if only 16GB of memory is used.


4) VCCSA is the voltage that helps facilitate memory overclocking on Ivy-E. DRAM voltage does too (obviously). Speeds over DDR3-2400 may require high levels of VCCSA if high density memory configurations are used. We have used up to 1.40V with 64GB of memory at speeds over DDR3-2400.


5) Before you are wowed by the voltage or the promise of running DDR3-2400, remember that even if you find a single memory kit rated at these speeds, there is no guarantee that the CPU memory controller will manage the frequency. Each CPU sample varies - some are better and some are worse at overclocking.


6) Bear in mind that memory timings are just as important as memory frequency. Chasing raw frequency at the expense of timings is worthless on a 24/7 system.

Just because a CPU won’t run a given memory frequency in a stable manner, don’t despair! It’s usually "technically" faster to drop the memory ratio one step lower and run a tighter set of timings. Doing so can be complex, but that’s what the forum is for. We’re here to guide you on what to run in these situations. Ask!

7) Memory speeds over DDR3-2400 may be more stable when using the 125 MHz BCLK strap.
453,127 Views
186 REPLIES 186

Hi Raja -

Thanks for the info in the OP ...put your guide to good use today on a new 4960X in RIVE with the latest BIOS (4502)...is capable of running strap125 / 166 with DDR3 2666, even touching 5 GHz at still reasonable v-core voltage

IMPORTANT :confused: QUESTION on VCCSA (and by extension VTT). Your guide states: "Overclocking Ivy-E for the masses is centered on two voltages: Vcore and VCCSA (more on VCCSA later)." ..."Speeds over DDR3-2400 may require high levels of VCCSA if high density memory configurations are used. We have used up to 1.40V with 64GB of memory at speeds over DDR3-2400."

Once past DDR3 2333 and on to 2666, VCCSA will auto-set at about 1.25v to 1.3v on my 4960x / RIVE...with Sandy-W / C2, that was 'no-no' territory, at least according to some posters here and elsewhere...while I may be able to get the VCCSA down a bit manually with further testing at the higher (2333+) memory speeds, is 1.25v to 1.3v 'acceptable' for a system that isn't on 24 / 7 ? Is there a max value beyond which is 'no - go' ?

Thanks for your help ! 🙂

Raja@ASUS wrote:
The DRAM will work, but might not be fully stable without some tweaking, or you might have to run it a bit slower. I would try that first before you purchase new memory.


Hi Rjaj

Have now installed 4960X....

I have a stable OC of 4.6, with 4.7 I'm getting the cold boot BSOD issue 0x124. It usually does it at the log into Widows screen or somtimes make it to desktop, then about 40 seconds latter. So I just power down clear CMOS put 4.7 OC profile back and is good to go. Tried setting Ram to 1600 made no difference. I did try BIOS 4502 but could even post with 4.7.
Is there any setting I can try to get over the hurdle or am I at the limitations of the CPU?

Would the next BIOS update when it comes out be more stable?

Or another option, would the Black Edition be more stable?
System specs
Win 11 Pro 21H2
Rampage VI Extreme
BIOS 3801
I9 10920 @5.0GHz all cores
EK-KIT G360 (CPU only)
Samsung 960 PRO M.2 NVMe 1TB
Samsung 970 EVO M.2 PLUS NVMe 2TB
Kingston Fury Renegade 4TB PCIe 4.0 NVMe M.2 SSD with Heatsink
1 Samsung SSD 4TB EVO
1 X Samsung SSD 860 EVO 4TB
2 x 10TB drives
1 x 2TB drive
ASUS ROG Strix GeForce RTX 4090 OC 24GB @3015 MHz
32GB G Skill Trident Z 3466MHz 15-17-17-37
Corsair AX1200W PSU
Sound Denon AVC-A1
LGC1 77

Raja
Level 13
After a certain point, adding more voltage makes no difference and can make things worse. You just need to find the most effective point when tuning the voltage for the return you're getting in performance (assuming the tuning you're doing is performance related and not just chasing frequency).

Raja@ASUS wrote:
After a certain point, adding more voltage makes no difference and can make things worse. You just need to find the most effective point when tuning the voltage for the return you're getting in performance (assuming the tuning you're doing is performance related and not just chasing frequency).


Thanks Raja - 🙂 ...I have now run DDR3 2666 w/VCCSA and VTT at 1.175v settings, including memory stress testing, CineR15 etc. ...when using 'auto settings', there was just this big jump from 1.105 / 1.1 at DDR3 2333 to 1.25 / 1.3v...I realize why it has to do that on 'auto settings' which have to satisfy the largest possible number of clients.

A very much related follow-up question. On more than one occasion, I have read that VCCSA (and possibly VTT) should "Always be around 0.3v to 0.4v from vCore'...is that true ? I ask because I'm currently testing the 4960X from 4.833GHz to 5GHz...it seems that my vCore is 'too low' > NOT re crashing etc, but re this '0.3 to 0.4v' rule, if it is indeed that. So if I drop VCCSA / VTT to accommodate working (and lowish) vCore, VCCSA / VTT end up in the 1.12 - 1.14v + - range which is too low for DDR3 2666 (though fine for lower DDR speeds). Ergo: Does this 0.3 to 0.4v rule exist ? Does it mean 'no more' or 'at least' ?

Once again, THANKS for your knowledge on this...and btw, I did try one of the Rampage IV / BIOS 4502 'Rampage preset memory' settings form Schamino - wow, not only works great but on some things is initially counter-intuitive, yet leads to a 10 point gain on CineR15 at a given CPU / Memory speed...please pas on my 'thanks' :cool:

Raja
Level 13
The 0.3v rule was centred on SNB-E. Ivy-E may not have the same requirements, or the delta may be different (if it exists). Such things are usually governed by how internal domains are coupled to one another (DC). Experimentation is required to work out if any such voltage delta relationship exists. A crude guideline to work such things out is to see how Intel set the rails up in relationship to one another at stock and work from there - using common sense as voltages are increased.

-Raja

Raja@ASUS wrote:
The 0.3v rule was centred on SNB-E. Ivy-E may not have the same requirements, or the delta may be different (if it exists). Such things are usually governed by how internal domains are coupled to one another (DC). Experimentation is required to work out if any such voltage delta relationship exists. A crude guideline to work such things out is to see how Intel set the rails up in relationship to one another at stock and work from there - using common sense as voltages are increased.

-Raja


Thanks Raja - ...Intel's 'stock' parameters and their relationships to each other (vCore, VCCSA, VTT etc) are certainly a good starting point...though I don't think that I actually ever ran any chip at stock 😄

Wondering if s.th. like the embedded pic (source HWBot) re SB-E exists for Ivy-E...have you seen such a data sheet ?
29395

snowking7 wrote:
Thanks Raja - ...Intel's 'stock' parameters and their relationships to each other (vCore, VCCSA, VTT etc) are certainly a good starting point...though I don't think that I actually ever ran any chip at stock 😄

Wondering if s.th. like the embedded pic (source HWBot) re SB-E exists for Ivy-E...have you seen such a data sheet ?
29395



That's taken from Intel white papers - you'll need to surf the Intel site and check. Bear in mind any shown maximums relate to stock operating frequency only - as the max voltage Intel stipulates is based on power consumption. A lot of people don't realize that - the white papers are written for the technically qualified, and are often misconstrued by the unwitting.

-Raja

Raja@ASUS wrote:
That's taken from Intel white papers - you'll need to surf the Intel site and check. Bear in mind any shown maximums relate to stock operating frequency only - as the max voltage Intel stipulates is based on power consumption. A lot of people don't realize that - the white papers are written for the technically qualified, and are often misconstrued by the unwitting.

-Raja


Tx Raja - ...I have all the relevant white papers now (close to 1000 pages combined / pdf 😄 ), and I get your point about stock voltages for white papers (am in the business for decades)....for me, VCCSA is one of the most critical parameters and in the white papers, I found two 'max VCCSA'...1.25v and 1.4v...which is why I asked in this thread.

Obviously, it also comes down to the amount and speed of RAM, 64GB of 2666 RAM is a different animal than 16 GB of DDR3 1866. Still, I generally get nervous with VCCSA over 1.2v...finally, what I cannot (not yet?) find in the white papers was that aforementioned .3v to .4v 'delta' to vCore (and related, .6v delta to DRAM)...that apparently came from an Intel engineer when discussing SB-E at a conference...I'll keep my eyes / ears open for similar info @ Ivy-E, and if you find a reliable (Intel) source, please post - Thanks again for your responses / help

xarot
Level 11
I have 4x8 GB Corsair 1866 RAM and when overclocking I cannot even boot with stock 0.95 V VCCSA, for Prime95 stability I had to raise VCCSA to 1.1 V. I have 3-way SLI running @ PCIE 3.0 so this may be the cause? 🙂 Previously I could run VCCSA as low as 0.95 when I had the same setup but G.Skill 16 GB 1600 RAM. Even then when I tried for 4.5 GHz I got some really weird things going on, like the monitor signal on/off constantly in BIOS screen. Could that also be because lack of VCCSA? When I am running XMP only so I am at 4 GHz on all cores, it works at stock VCCSA (AUTO).

Currently running 4.5 GHz and aiming for Prime95 27.9 stability, voltages are now 1.375 Vcore (+0.06 V / high LLC), 1.1 VCCSA, 1.125 VTT, 1.825 PLL, internal PLL overvoltage disabled, CPU current capability AUTO. So far 7 hours in and looks good..
Main: i9-10980XE - Rampage VI Extreme Encore - 64 GB G.Skill Trident Z Royal 3600 CL16 - Strix RTX 3090 - Phanteks Enthoo Primo - Corsair AX1500i - Samsung 960 PRO 1 TB + Intel 600P 1TB - Water cooling
HTPC: i7-6950X - X99-M WS - 32 GB G.Skill RipjawsV DDR4-2400 - GTX1050TI - Bitfenix Pandora - Corsair AX860 - Intel 750 400 GB + Samsung 1 TB 850 EVO
All around: i9-7980XE - Rampage VI Extreme - 64 GB G.Skill 4000 CL18-19-19-39 - Strix RTX3090 - Phanteks P500A - Samsung 960 EVO 512 GB - Water cooling

Raja
Level 13
That kind of info will never be in the white papers as the white papers do not stipulate anything in regard to overclocking (other than it not being covered). If any DC deltas exist, they would be part of stock spec and margins - so you can see why Intel does not need to put them into a white paper that deals only with stock parameters.