RAM Overclocking on Clevo

Hello everyone!!

I'm overclocking my RAMs to their maximum 24/7 stable. I'm using 4x16GB Ballistix Sport LT 2666 MHz CL16

For some reason the maximum overclock that I can get is 3200 MHz CL18 (12 x 133 @1.3V), pretty stable so far, but I can't boot at anything higher than 3200 MHz, is that a Clevo P870TM1 limitation or my RAM can't go any higher?

This dude over here is able to get 4000 MHz CL20 on a 4x8GB config on an ITX MB. What am I missing?:

https://www.overclockers.com/forums...1TXx6z91JZ-QDMD_ldhv5rrwJAHVEij7FeWtzZirBYagQ



PS: RAM overcloking is by far the most complicated and tedious hardware overclocking

 

Each laptop has a limit based on CPU imc and system board trace quality because of design constraints.

There’s also the fact your memory maybe hitting a wall considering you’re using 2666 and have pushed it to 3200 with loose timings already.

Only real, valid way to see what is giving up the ghost is to purchase some quality better tier sticks snd work your way up with stock timings. (IE 3800 CL18 sticks running at 2666 snd keep pushing up till the system can’t handle it, drop back down snd then tighten up your timings).

 

I just want to add that Skylake-X is designed for Quadchannel use.

And this motherboard is specially designed to hit DDR 4000 speeds. At least that's what is stated on the Asrock product page.

 

This moterboard max support 3200Mhz I Have Crucial Ballistix DDR4 SODIMM 32GB 3200MHz CL16 (BL32G32C16S4B). I try to set 3400/3600 at timmings CL18/19 never run this laptop.

 

@jc_denton has his 870TM1 running @ 3333

@Johnksss has his 870TM1 running at 3467 stable and able to test at 3600

So that can't be true.

Unless the sticks are certified at higher speeds, there is no guarantee they will run past their certain speed on platforms.

There's a reason top tier memory is more expensive once you start moving past 3200.

Johnksss sig:



JC's sig:

 

on P775DM3-G @3200

 

So what you mean is that there are more factors to consider when overclocking the memories more than just the RAM specs. I had to loose the timings because it won't be stable at stock timings and at higher clock. Also I had to rise the voltage from 1.20V to 1.30V.

Maybe I need to drop this question on the P870TM1 sub-forum and see what are people getting when overclocking, because 3800MHz SODIMM memories are a bit expensive to buy and maybe hit the same 3200 MHz wall that I'm facing now.

Yeah I noticed that the dude from the post is doing quad channel. but I thought the RAM speed was more in to the RAM than in to the motherboard. Do you know if newer laptops are getting higer clocks? I remember the Area 51M R1 was stuck at 2400 MHz which is ridiculous.

Did you try rising the voltage? I wasn't able to get stable 3200MHz on my modules unless I raised the voltage at 1.30V.

Interesting. I wonder how jc got that exotic clock speed. 3333MHz is not even possible with 2 x 133 x N or 2 x 100 x N which are the configs allowed on my system BIOS. Also it seems that Johnksss has 3800MHz modules, but he is not able to run those memories at XMP profile, which means that the P870TM1 has some sort of limitation on RAM clocks

 

If everyone could OC 2666->3200 modules to greater speeds, they would I think and there would be no market for top tier memory. There's a reason upper tier modules are more expensive. They are rated, guaranteed and give you much more room to play with and rule out your sticks being an issue.

The trade off for let's say 3800 modules is you can really tighten up the timings (which can be equally as important) at lower clocks frequencies. Corsair kits are pretty beastly but you get what you pay for. I picked up some Ballistix CL16 3200 sticks (2x8GB) and they were dirt cheap brand new (~$70) XMP PnP and working fine in my 870TM1. They flat out will not run above 3200 no matter how much I loosen up the timings or increase voltage while I know my system can do at least ~3466 or higher (tested and verified with a Corsair 3600 kit), but if they could do more, I suspect they would have been sold as such by Crucial.

Even all desktops aren't guaranteed to run faster sticks due to QVL, poor quality or sometimes the CPU has a poor IMC for high frequencies (but still can easily hit paper specs). Sometimes it's the memory; sometimes it's the host environment; sometimes it's both.

I had some desktop corsair vengeance 3200 sticks that were CL16 B-die and they crapped out at a bin above even with extremely loose timings. This was on my EVGA FTW Z390 which then took some G.Skill 4000 B-Die sticks and run them like a champ. I tested those same G.Skill 4000 sticks which run like a bat out of hell on my EVGA FTW Z390 in an Asrock Z390 board and they flat out refuse to run at 4000 even with the same 9900k in it at stock. Even pumping up the voltage does nothing. Dropping down to 2666 and climbing, I found the board craps out at 3600, but that's ok because I only needed it to run those same Vengeance 3200 sticks which in the Asrock Z390 board still won't go above 3200.

After 3200 (or even 3000 some would say), unless you are running known, good, cherry picked 3200+->4000 sticks that are sold as such, nothing is guaranteed.

I, personally, would be happy that I was able to get 3200 out of the 2666 sticks, albeit CL18 would have me increasing to 1.35+ to tighten up those timings if possible. At that point, you have to ask yourself do you want to drop $200-400 on memory sticks to eeek out 2-3 higher bins and/or (more importantly) much tighter timings? I'm sitting on a brand new set of factory sealed Corsair 3800 DDR SODIMMs https://www.corsair.com/us/en/Categ...3800MHz-CL18-Memory-Kit/p/CMSX32GX4M4X3800C18 asking myself the same question for the last month or so while they sit up for sale lol. Eventually I'm going to cave and I know it or a system is going to come in needed memory and I'll move my Ballistix sticks to it and then crack open the good stuff.

 

Oh, ok. I was referencing the 870TM1.

 

lol, If I were you I'd be so tempted to open that brand new memories and install them in to my laptop and see how far can I push them. I wonder if there is a laptop that can actually hit 3800MHz in RAM, anything over 3200MHz seems pretty rare. A shame considering new PC MB can push up to 5000+MHz

I'll try to lower my CAS latency a bit and see if I can manage to get it stable. But overclocking RAMs is so time consuming, a lot of restart, trial and error and a lot of testing to check stability. On a side note, I wasn't able to get 3200MHz CL18 stable with voltage over 1.30V. Is that normal? I mean that config was stable @1.30V but unstable @1.35V+.. that doesn't makes any sense to me, unless that's a board limitation of some kind or a temperature issue, because more voltage equals more stable RAM? isn't it?

 

Slots can also make a difference I have found with 1 + 4 being the most stable for me.

 

Probably a temperature issue. Memory can experience errors if it's running too hot.

 

Main problem is the notebook motherboards. They won't get the love they deserve. They aren't equal good made as the desktop boards. Smaller form factor also doesn't make it easier create proper signaling and paths. ITX board is still better because they are engineered more in same way as the ordinary boards for desktops. The quality of the IMC isn't the big problem anymore. Not the reason memory overclock is screwed up for notebooks (we still talk about lower memory speed vs what to expect today).

Put it this way. The old DM board don't handle ram overclock equal good as the newer board. This means they making better board now due Intel have increased ram specs. But still not good enough.

Ram quality still matters. I tried to overclock 2666 sticks in my P870Dm with no success. Bro Jackson managed 2933 on the newer TM board with same sticks. With better sticks he managed a lot better. My cheapo lower specked OEM 2133 sticks was able to do 2800 with better timings than the 2666 kit I tested. With better sticks I expect the outcome also would be better for me. Even on a 5 years old bord designed around Intels 2133MHz specs The newer has to be better, goes damn slow. At least for notebooks, LOOL

Today's 2933 standard speed, should with the same 31% overclock have reached past 3800 without any problems. But as you can see, we are still not there even with the new and fancy. The desktops have come a much longer way in front now than notebooks. Yeah, notebooks don't get any love. They are treated as bastards.

 

But what is it exactly that stop laptop's manufacturer to create boards that can handle faster sticks? I checked today the latest MSI/Asus/Alienware/Clevo high-end gaming laptops and none offered anything faster than 3200MHz, even Alienware Area51m R2 offers 3200MHz sticks with CL20 or something like that which is unacceptable CAS latency at that speed. Maybe most laptops buyers don't care about faster memories?

 

I will drag it down to the added costs. Not worth it if most customers still will buy it with standard ram specs. Most people look on bigger size over faster sticks. And faster/better sticks in the bigger size will for the most upping the total price point even higher. Many of the customers will say no thanks to that.

 

Ram overclocking also depends on chip model/quality. I overclocked 2 cheap oem hynix cjr and 2 hyperx jjr based ram sticks from 2666/2400 to 3000 16 18 18 36 510 on 1,2V. I know that diffrent sticks have diffrent memory chips. For example i bought new hyperx 2x8gb and it had nanya memory, diffrent hyperx afr, jjr even mjr. Here is a nice guide about ram https://www.reddit.com/r/overclocking/wiki/ram/ddr4

 

With vertical slots and having one dimm per channel (that's a quad channel board) also plays a big part.

 

3800 has now been validated as being able to run on P870TM1 thanks to @Prema of course.
@electrosoft

 

Oh man.....

I think I might have to delist and crack those sticks open....damn.

 

I had better luck with my old 9900K "P0", reaching 3600MT/s with 4x8Gb 4000 sticks. The R0 doesn't play as well with them, capping out at around 3460 and only working when the starts align.
The trick is to clear nvram with each attempt at a new speed, or to boot directly into bios after nvram reset and set memory profile to custom then reboot and raise the ratio without touching the timings.
I documented the process in the OC lounge if you can dig it up.

As for 3333MT/s, it's at 133Mhz x 12 ratio + 133 QCLK ODD Ratio - I tuned them as tight as 1.45v would allow me.

 

But how? lol, you need the latest Prema BIOS? the one from 2019? I also see that you use single rank sticks, is that because single rank sticks clock higher?

How you clean the NVRAM? just by shutting down the laptop? Could you please link me to your tutorial? Impressive performance you're getting by the way, 39.1ms latency!

 

Thanks guys!

@Tyranus07
They were able to run 3700 with a 8700K.

As to them running at 3800 Mhz, all I did was set the multiplier and voltage settings and let it boot on it's own. No tuning or adjusting. I was curios as I looked for prior runs and thought Gee, none of my runs were while I had a 9900K.

 

Hmmm, true my previous run with 3600 sticks that capped at 3467 were with my 8086k.

@jc_denton I haven't tried any runs with my 9900k p0 or r0 since I've been coasting by with these inexpensive Ballistix 3200 sticks.

@Tyranus07 You'll notice both Johnksss and JC are running more expensive. top tier memory kits from Corsair. If you really want to push your memory, you might want to think about investing in a set. 4x8 tends to be better on Z3X.

 

Single rank sticks do seem to do better, less capacitive loading.

 

Yeah so it seems is possible to achieve much better RAM speeds than what I'm getting right now, with better sticks. Probably single rank sticks. It is good to know that the P870TM1 has no a hard limitation on RAM speed. At this moment is kinda hard to me to buy new sticks, because a) those sticks faster than 3200MHz are not pretty common, and thus are expensive. b) I'm building a desktop PC so that's where my money is going at the moment.

The good thing is I was able to get stable 3200MHz CL17

 

On my P775DM3-G i have stable 3200MHz cl15 on 1.35V

 

What sticks are you using?

 

You clear nvram by pulling both batteries and holding power button for 30s, this way the machine flushes any old settings and trains anew.

As for the guide or reference, it's around here:
http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1827#post-10974733

you might want to go a few pages back and forward there, but as far as I remember it's around there.

 

Crucial Ballistix DDR4 SODIMM 16GB 3200MHz CL16 (BL2K8G32C16S4B)

 

For a quick bench things sometimes seem stable, but it usually takes a lot of fine-tuning to actually get it stable.

Getting this TeamGroup 2667Mhz (standard RAM without XMP) rock-stable at 3200Mhz CL16 took quite a bit of fine-tuning while keeping the voltage at 1.35v, even though it can be made to boot and pass AIDA64 at 3600Mhz:

Spoiler

[\SPOILER]

Here the TM5 Mod (run as admin) in case you wanna test stability of your own RAM OCs.

https://mega.nz/file/SJMgTAYD#n_8QVY6FS7LzJkBe79k7UB0CUm8LDD_S5PFi241VRNQ

 

I think today it's not rare to see a ram stick sold as 2666 cl16 for exemple but capable of 3200 cl16. Of course there are some brand that want to sell as cheap as possible, but if demand exist, it's possible to sell 3200 stick as 2666 and just program them to work at that speed by default. I remember seeing this on some old Athlon/Phenom CPU
Now, talking about what the main board or CPU is capable, well it's quite impressive!
I think you can have some variation using different stick in different slots as voltage can varry a bit depending on the lenght of the track.
For performance sometime it can be better to go with lowering timing instead of finding highest frequency.
@Prema thank for sharing

 

Sorry, forgot, it's this RAM:

Spoiler

[\SPOILER]

 

Are they silk screened like that?

 

Yes, they are Team Elite labeled.
The manufacturer is SpecTek Inc., which is a specialized subsidiary of Micron.

Here the same 2*16GB 2667Mhz (CL19) sticks stable at 3400Mhz CL20 @ 1.35v:

Spoiler

 

It was funny when people complained that gamers nexus were "only" using 3200Mhz ram in their testing without realising it was at CL14 with optimised sub timings.

 

Yes, timings can be tightened gradually as clocks are dropped:

Spoiler

[\SPOILER]

 

They were basically the same as 3600-3733Mhz results with looser timings.

 

I've been playing around with timings, doing 3050mhz cl14, not had time to finish it just yet. Really does not like 16x though even at looser timings.

 

@Meaker@Sager It's still nice!
Even if i'm not using DDR4 but DDR3, for now i can just run my 4 sticks @1600 CL9 and 1T command rate timing on my P150SM. They're all disigned to run at default 1600 CL11 and whatever with CL12 timings i just can't boot at 1866, even with 2T command rate.
From what i understand, higher frequency -> better bandwidth and lower timings -> lower latency. Am i wrong?

 

Higher speed at the same timings also means lower latencies too as it is measured in click cycles so if each cycle is shorter that counts too.

My ddr3 4gb sticks were 1600c11 and they did that at 2133mhz c11. Good times.

 

Some of ram OC is pure voodoo, for instance swapping the sticks around and suddenly getting 200mhz more at the top end.

 

Yeah odd signal integrity issues as you go beyond spec. Back with my old 980m I lost 100mhz-150mhz of ram speed populating more VRMs.

 

I cracked them open and started working with them and I'm hitting a hard wall @ 3200 on both the extra Ballistix 3200 set and Corsair 3800 with this 9900ks r0. Even super loose timings and going up and down 1.25->1.45 along with covering the VCCIO/SA spectrum (1.10->1.20) won't
get past the wall. My 8086k did 3466 in here @ stock IO/SA with a set of 3600 sticks before. 3200 works fine and passes TM5 no problem on both sets. Anything past 3200 and it won't even boot. This is in pairs. I haven't single stick slot hopped yet. This is on the newest/latest Prema (1.07 R3)

Like you experienced, it might be an r0 issue. I might end up swapping back in my p0 if I can't break past this wall.

 

Or it's just sample to sample variation on some trait the notebook is more sensitive to.

 

The Ballistix sticks are the control in this experiment. They worked fine with the exact same conditions @ a "dirty" 3333 that would error out on memtest but always booted with my now long
gone 8086k. I'm relidding my p0 atm w/ the BartX to give it a whirl tomorrow to see if the same conditions exist with a different (but previously in there) CPU.

But as you say, there can be many other subtle variations (or even changes to the physical traces themselves over time) that could be causing it.

 

I went looking through my old post regarding the correct training, some notes:

Training guide:
http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1815#post-10973151

3600mhz notes:
http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1820#post-10973760
http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1819#post-10973732

Another thing do try is turn fast boot off, that way it may actually boot through a wall. Downside is that it will need to retrain each reboot, which sucks since it has a decent chance of failing (and takes a while)

 

Yes but your sample size of the revisions is not large, there could be chip to chip variations.

 

Hey, I work with the sample size I have....

 

I'll give it whirl next time I have time. Right now, my next project is my troublesome desktop (see: desktop forums). After a full day of testing 2-3 sets of memory and 3 different NVMe drives and switching back and forth between the p0 and r0, I'm spent.

The p0 was WAY worse in the TM than the r0 it balked at 3200 repeatedly throwing up errors in TM5 even with VCC adjustments yet on my desktop it handled 4000 mem at a lower vcc setting than my ks. I had also forgotten how much louder the p0 was in the 870TM relatively speaking to the 9900ks.

I did try stock scaling then stock + VCC scaling when it all came to a halt @ 3200. Maybe my steps were wrong. I was feeding it extra voltage at times though. From your guide, it seems the idea is to stay the course @ stock voltage and SPD timings and just scale frequency and adjusting VCCIO/VCCIA if necessary so the limitation is the memory eventually?

When I am tuning for OCs or mem, I usually remove the battery and cmos battery on laptops. I've done this since back in the good Alienware days.

Then when it fails to boot I just unplug the power and for extra oomph hit the power button for a few seconds. Plugging back in resets and I get the Prema message.

Boot to stock timings which for Corsair varied depending on the number of sticks I had installed. their SPD is 2666, but with all 4 installed it defaulted to 2400.

I never tried training them as a complete set setting at the odd stock 2400 boot. I'll have to do that.

Your post outlined the sequence as:

" nvram reset > set custom profile, keep timings and stock 1.2v > train > full shutdown from windows > go up a multiplier add voltage > train > full shutdown."

So install, reset, set custom but keep timings as presented and stock Vmem. What do you do when it fails to boot? (IE, no full shutdown from Windows available). Eventually climbing will fail before boot. I just want to make sure I'm not missing a step or something.

Earlier today I slot hopped with individual sticks just testing with no major finds except slot one seems the most consistent past 3000mhz both frequency and tighter timings.