DDR4 RAM overclocking 101 guide

[maxfly]

there ya go Batboy! proof that its doing what you had hoped it would well done man!

 

[Brando]

i found this on google. thanks batboy!

p.s. can anyone confirm or deny whether leaving sub timings on auto is the way to go? there seemed to be an implication in the thread that leaving on auto causes them to loosen too much and flush your gains down the toilet. i'm ready to play with my new b die kit but need to do a little homework.

 

[Alaric]

I leave most secondary timings on Auto. From what I've found (confirmed by Woomack), secondaries seem to be mostly the mobo working out how it can give you the primary speed/clocks you asked There's a JEDEC standard for calculating tREFI if you don't want to play with it. Generally higher is better-until it's not and your OC eats your boot.ini. LOL
From my old thread:

The final settings I have are as follows:

CPU @4600 MHz
Uncore @4200 MHz
vCORE 1.352v-measured
VCCIO-measured 1.194v
vDIMM-measured 1.450v
__________________________________________________ __________________________________________________ __________________________________________________ ________________
Timings Changed By Me

RAM @3733 MHz
Multiplier x37.33
CAS 15
tRCD 16
tRP 16
tRAS 31
tRC 48
tRFC 275
tREFI 14559
__________________________________________________ __________________________________________________ __________________________________________________ ________________
Timings Left On Auto In BIOS

tRRD 7
tWR 24
tRTR 7
tRTW 12
tWTR 9
tWTW 9
tRTP 12
tWTP 42
tFAW 40
tWCL 14
tCKE 8
tRTL 65
tIOL 4
BL 8
__________________________________________________ __________________________________________________ __________________________________________________ ________________
Results

Memory Read 52,490 MB/s
Memory Write 55,265 MB/s
Memory Copy 48,270 MB/s
Memory Latency 39.5 ns

winsat mem 45,038.56 MB/s, up from 35,887. That's a 25.5% improvement. Memory Read speed is up 22.5% and Latency is down 6.8%.

The overall effect on performance may not be Earth shattering, but if I had to buy the performance I now have, from scratch, it would run me $889 for a Z270 direct replacement for my mobo, a 7700k and G.Skill 3866 MHz RAM (because my timings are tighter than their 3733 MHz kits) at newegg prices. So what if my overall performance only went up a couple percentage points? I'll take $900 worth of free stuff any day. And almost everything I did, I learned at OCF.

 

[Brando]

So, what are the cool kids using as the definitive test for bandwidth and performance gains?

edit: i think it turned out ok. they really need to stop cranking the vccsa and vccio so high though. they had me at 1.25 and 1.35v which shows as danger red in the bios but is what they came up with on auto settings at ddr4000. i turned it way down and still need to test but so far so good. any tips? (sorry for the format. images aren't working for me today)

[url=https://postimg.cc/image/n666x4enl/] [/URL]

 

[Alaric]

I don't know about the cool kids, I used the trial version of Aida64. I kept VCCIO and VCCSA both under 1.20v (measured). I forget where exactly I got that spec (I believe Woomack was one source), but that was the max I settled on.

 

[Brando]

I don't know about the cool kids, I used the trial version of Aida64. I kept VCCIO and VCCSA both under 1.20v (measured). I forget where exactly I got that spec (I believe Woomack was one source), but that was the max I settled on.

i figured out aida64 but didn't know i could go that low on vccio and vccsa. ill try lowering it even more. it's hard to believe they would put it so high at auto. it's like they want to destroy your cpu.

 

[wingman99]

i figured out aida64 but didn't know i could go that low on vccio and vccsa. ill try lowering it even more. it's hard to believe they would put it so high at auto. it's like they want to destroy your cpu.

All motherboards set high Vccio and Vccsa when overclocking because some Processors IMC need that much voltage to overclock memory.

 

[Alaric]

I used winsat mem in Windows, too. Open a .cmd window (Run as Administrator) and type winsat mem, hit Enter.

 

[Brando]

idk it seems like most people lower it. they should be conservative and if people need more voltage they can find out when they call tech during troubleshooting. obscure voltages at super high levels seem like a bigger danger to me since they could stay that way indefinitely until the imc dies and nobody would be the wiser. just my opinion.

 

[Alaric]

idk it seems like most people lower it. they should be conservative and if people need more voltage they can find out when they call tech during troubleshooting. obscure voltages at super high levels seem like a bigger danger to me since they could stay that way indefinitely until the imc dies and nobody would be the wiser. just my opinion.

I agree, and that's why I set mine manually. All my voltages were set manually at the max I wanted to allow before I turned all the power saving stuff back on. It's a pain, but motherboard manufacturers now seem to all err on the side of stability instead of caution. I guess it makes it easier for everyone to be an OC Superman, but it sure is hard on parts. LOL

 

[wingman99]

idk it seems like most people lower it. they should be conservative and if people need more voltage they can find out when they call tech during troubleshooting. obscure voltages at super high levels seem like a bigger danger to me since they could stay that way indefinitely until the imc dies and nobody would be the wiser. just my opinion.

Well yes folks lower it when they are in the form like you and that is very few and I'm on a lot of overclocking forums. What I see is when folks can't post or find stability they raise it. With stock XMP 4000 17-17-17-47 memory the manufactures set Vccio and Vccsa higher so when you select stock XMP 4000 17-17-17 -47 and the PC is not stable or won't boot you can call tech support and the only thing they will tell you to try to make it easy is lower the memory speed. I call all the time and ask what the most calls for memory are and it's XMP not running the rated speed stable or no post.

It's really is not that dangerous more so than what we do over clocking the processor cores. My i5 8600k runns stock 1.080v load. At 5.0GHz it runs 1.312v load and if I had a delid and better cooling I could run 1.370v load like most folks.

Alaric that Vcore looks safe LOL @1.411v

 

[Alaric]

I thought I read 1.20v was the max for my chip, but the datasheet is 260+ pages for Volume 1. Haven't gone back through it lately. LOL

 

[wingman99]

I thought I read 1.20v was the max for my chip, but the datasheet is 260+ pages for Volume 1. Haven't gone back through it lately. LOL

Data sheets are for stock specification of tolerance for Voltage and AMP's not intended for overclocking the IMC or cores.

 

[Alaric]

That's where I got the max voltage of 1.52v from, which seems to generally be accepted as the limit for my Skylake.

 

[wingman99]

That's where I got the max voltage of 1.52v from, which seems to generally be accepted as the limit for my Skylake.

It is 1.52v at 100AMPs = 152 watts maximum. Intel also states on that page the Voltage and current specification are only valid meeting specifications for clock frequency.

 

[Alaric]

It is 1.52v at 100AMPs = 152 watts maximum. Intel also states on that page the Voltage and current specification are only valid meeting specifications for clock frequency.

Not sure what they mean by that, but thanks for dredging through it all. I took it to mean "STOP HERE". I went through so much data for my overclocks I got a little burned out on it. LOL

 

[Woomack]

1.52V CPU is max in theory what wasn't even checked by Intel. Some engineers said that no one tested it during longer work. The same max was in a couple of generations where minimum voltage requirements and manufacture processes were changing. The same as we could see that the absolute max for DDR3 was 1.55V ... somehow most memory manufacturers were selling memory at 1.65V and were giving full warranty on that. Intel simply sets theoretical max in their specs what is also base for their warranty.

VCCIO/VCCSA, like wingman said some posts ago, is always too high because of luck to CPU/IMC. Think that my MSI Z370 is setting these voltages at about 1.45V when I set the memory at 4200+ but won't set any higher voltage. I guess it's somehow tested max safe voltage but hard to say for how long. On the other hand, I don't need any higher voltages up to 4500+. I still have no idea why 4500 runs 100% stable and 4533 has problems with booting but that's another topic.

Back to one of the earlier questions. Sub-timings may affect performance more than the main timings.
From sub-timings, some have to be synchronized (or at least can't be higher/lower than some others) and some that you can set how you like and won't affect stability. The list is really long. One example is tCL and tWCL. One is read, one is write. Write can't be higher than read so you set CL16 but WCL15, 14 or 12. It can be the same as CL but at auto, tWCL is always lower. You may need to adjust that at more relaxed timings/high frequency as memory timing tables are not including really relaxed timings.

The easiest way to tweak sub-timings without knowing them (but can take a lot of time) is to try one by one. You can lower a couple of timings at once and check if it works, then run a quick memory test, even AIDA64 as when system boots and runs in some quick tests then in most cases can be stabilized.

New motherboards have memory profiles, usually 3-6. Use them as they are helping to save a lot of time. Once you find something quite stable (not even fully stable) then save it as your next step may end on clearing CMOS.

When you read various threads on various forums then you will see that most users recommend Samsung. Maybe it's the fastest IC but in most cases, people just don't know other IC and they don't even try to tweak them. For some reason, new Microns run higher on some motherboards than Samsung (almost only Ryzen boards). Then is possible to tighten sub-timings to the level of good Samsung and even though the main timings are more relaxed then general performance is about the same.

 

[wingman99]

You can use HWmonitor to observer processor package watts. However, it is not that accurate, I tried two motherboards with the same clock and load temperature and the readings differed. Intel is stating the specifications are only valid stock clock speed.

There is no specifications tolerance for memory, the ICs are rated for 1.20v tested at 1.35-1.5v from G.SKILL http://www.gskill.com/en/product/f4-3200c14d-32gtz So the motherboard manufacture engineers are setting the Vccio and Vccsa for folks just to be able to run XMP memory or manually overclock with ease and successes. The few folks in the forums know less voltage is better when overclocking, even Intel states that.

Intel's rules for overclocking

The Rule: Maximize frequency and minimize voltage while extracting as much heat as possible so the system remains stable. https://www.intel.com/content/www/us/en/gaming/how-to-overclock.html

With memory it is better to get it up and running stable first, then if that goes well lower the voltage for IMC and ram.

 

[Woomack]

The first HyperX Predator (also reviewed on the OCF) had 1.5V in XMP. When I got my kit about 2 months later then it had already 1.35V. Anyway, somehow they tested that it wouldn't kill anything or it wouldn't pass internal tests ... I guess
Some brands have 1.5V kits right now but all of them are 4500+. This also brings some questions like who will actually run that memory at 4500+ when there are no guaranteed platforms for that and only maybe 5 motherboards will run at 4500 or higher (I mean stable without issues).
JEDEC specs is 10% above stock which is 1.20V. We are much above that since the premiere of first DDR4

Intel is pretty much clueless about all these settings. Their engineers are not passing full info to their marketing and marketing is creating BS articles. Go to any Intel conference and you will hear so much BS you will probably laugh for a big part of that. Last time I heard that Intel IGP was designed for gamers and they were trying to give examples ... running videos made on much stronger setups.

No matter what you do, above some point you have to set higher voltages. IO/SA are scaling with frequency. VTT is scaling with VDIMM. Personally I push the limits for competitive benching but for 24/7 work I like low voltages as performance gain because of couple of tighter timings is not helping in anything.

On my MSI mobo hwmonitor or hwinfo have problems to read many sensors. I can't even tell how high are most voltages as it's not listed even in BIOS and is not "greyed out" next to voltage settings. At least in my case, I assume that SA/IO run at 1.40-1.45V just because manual settings up to these values give similar results. For 24/7 work I try not to set more than 1.35V.
Funny mobo I can say, have to guess many things but if you know what and how to set then it's one of the easiest boards to set high memory frequency.

 

[Alaric]

My voltages were set based on my AIO's thermal headroom and what specs I could find, like the max vDIMM G. Skill sells. I got some luck, too, considering my board/chipset.