Corsair Vengeance LPX 3600-CL18 OC for Ryzen 5 3600

[examen1996]

@Chomiq The games in question are AC ODYSSEY and WITCHER 3, if i would to restart the machine, they would run (start, be playable all that , and i can play them without crashes or latency spikes).
However, if i leave my computer idle for some time like 1-2 hours(and even so this is not a 100% reproduceable behaviour) they might not start like at all, not minimised nothing at all.
But the strange part is , anything else would still work/start, i can play forza, or GTA V, paladins , apex legends , you name it it starts.

Initially i thought that maybe the ram oc that i had was to blame, but now I', inclined to say that it's a game related issue , which I'm ok with, i jsut want to be sure my system is stable.

 

[ocbenny]

Freeco not saying it will be better for you. The timings it gave me were closer to what my memory could handle. Seems you were lucky with your memory, I cant get the trcdrd below 21. I run 16-21-21-21-36 now on 1,4 V. Ive also entered all the sub and tertiery timings. Using a ryzen 5 3500x/b450 msi max gaming plus.

Edit.. Just realized there are two 3600 corsair lpx 3600 , they have the same product number which is confusing, one which is optimized for amd, running at 18-19-19-39 vs. 18-22-22-42. I have the later one. And trcdrd doesnt scale with voltage.

 

[examen1996]

@ocbenny Op has the same kit as you(and me ), like you I cant really get the memory stable at such extrem timing (even thou I tried, and it looked stable , passed memtest and 2h of prime95).
The strange part is , my kit of ram is not running ,,stable '' at 3.5 xmp settings, it has a general laggyness about it (im talking pure normal usage, not benchmarks).
I too just switched to 1.4 and to the safe presets of ryzen calculator, it just runs better.
The reason why the kit is behaving badly(at least in my oppinion) is because we(the people which got the 18-22-22-42 ) have an older revision, which were not validated for ryzen anyway sice it was realeased in 2017 ......so yeah, corsair is acting wierd om that one , but I digress.

Now , talking real world(and if I am wrong please correct me) after you get the memory stable at cl16 1.4v , the only question left is should you or should you not get a fan for the ram, that one I'm still debating on

 

[Freeco]

To be clear (but I already linked to it in my OP) this is the SKU of Corsair Vengeance LPX I have: CMK16GX4M2D3600C18

VENGEANCE® LPX 16GB (2 x 8GB) DDR4 DRAM 3600MHz C18 Memory Kit - Black

VENGEANCE LPX memory is designed for high-performance overclocking. The heatspreader is made of pure aluminum for faster heat dissipation, and the eight-layer PCB helps manage heat and provides superior overclocking headroom.

www.corsair.com

Mine has Micron E-dies.

As mentioned I bought the same set for my sister's Ryzen build. I didn't bother tinkering with the timings for her, but I did check which IC's her set had, and saw it was from a different brand (don't recall which, sorry). So apparently Corsair uses different ICs for the same SKU's, which can then explain why my results are better than yours. Anyway, likely not all ICs perform the same, even from the same brand.

Btw, there's another set like it, with a different SKU: CMK16GX4M2B3600C18. But this one runs at 2x8GB DDR4 CL18-19-19-39 @1.35V

 

[moproblems99]

The games in question are AC ODYSSEY and WITCHER 3, if i would to restart the machine, they would run (start, be playable all that , and i can play them without crashes or latency spikes).

Sounds like a component or service is crashing and being left in an unusable state and the reset is reinitting them. Check event viewer for any errors.

 

[examen1996]

@moproblems99 Thank you for helping . I think it was just the overclock not being stable enough , even thou the memtest passed , the strange behaviour kept on happening. Little did I know , using my safe configuration , setting the voltage to 1.4 (used to have 1.39) and switching the procodt to 32 kep the machine stable. Now i guess I could overclock further , but after having the kit running at standard xmp and making my machine work like crap...im just glad i found the sweet spot.


Again , thank you very much for your input, your are right !

 

[JasonToddRH]

@Freeco I Just want to thank you for existing and granting me a cheater's guide to ram overclocking, otherwise I would have probably stuck to xmp like the procrastinating peasant that I am. Also, if you would have the time to answer a few of my further questions on the matter, I would be forever grateful and even willing to pay for some insight in the mistery that is ram OC on ryzen 3rd gen.

 

[Freeco]

sure, shoot me some questions.
If I'll be able to answer them is another thing. I also just read some guides from a couple of other guys, and the rest was trial and error.

 

[JasonToddRH]

sure, shoot me some questions.
If I'll be able to answer them is another thing. I also just read some guides from a couple of other guys, and the rest was trial and error.

Well, first off, I'll list key specs that may be relevant: I have the exact same cpu and RAM sticks you do, but I'm rocking the x570 aorus ultra, so I've thought that might give me some more headroom into Overclocking (RAM at least). I was not interested in raising the clock any higher than 3600, because that would, in turn, get me out of the 1:1 ratio everyone talks about(although it seems like no big deal) and I'll probably end up messing with my infinity fabric to remain into that 1:1... Buuut that's a subject for another time. I'm only interested in tightening up timing for the time being.

So... Off the top of my head:
Has the "fast" profile in the DRAM Calculator worked for you?
Have you managed to get trfc under 560 whilst still having good primary and secundary timings?
And if I get some good G.skill 3600 16 16 16 32, should I push the timings further or work on the clock and then the infinity fabric? Because AMD said that the IF clock set to a stable 1900 isn't something most of their cpus can do.
About those G.skill sticks again: Would they be a good upgrade for 70$? (after selling mine).

 

[Freeco]

no, the fast preset didn't work for me. I didn't keep that excel sheet with all my notes, but I remember tCL 14 didn't run stable, I think I didn't even boot with a voltage bump up to 1.5V.
tRFC at 560 was the lowest I could go.
Overall, I used the presets in the DRAM Calc as rough guidelines. Some of the advised timings didn't work for me, while others I could even go beyond.

Likely those GSkill sticks will perform better. The advertised specs are what they've been tested with, but chances you can squeeze a CL15 or even lower out of them are realistic. But... it depends. That's the silicon lottery.
Is it worth the extra 70$? Imo not... I just wanted decent DDR4 3600 CL18 sticks instead of going for a cheapo 3200 CL16 set, and tried to squeeze a little extra out of them just for fun. But if I see the performance increase between the stock timings and my current tweaked timings, don't expect huge improvements from going to even stricter timings. I'd spend that money on something else. But that's just me

btw, you could also try bumping your infinity fabric to 1866MHz and your RAM speed to twice that. I did a few experiments with that, but I think I was a bit restricted by my older X470 motherboard, which was only certified up to 3600MT. Yours might do better.

 

[JasonToddRH]

no, the fast preset didn't work for me. I didn't keep that excel sheet with all my notes, but I remember tCL 14 didn't run stable, I think I didn't even boot with a voltage bump up to 1.5V.
tRFC at 560 was the lowest I could go.
Overall, I used the presets in the DRAM Calc as rough guidelines. Some of the advised timings didn't work for me, while others I could even go beyond.

Likely those GSkill sticks will perform better. The advertised specs are what they've been tested with, but chances you can squeeze a CL15 or even lower out of them are realistic. But... it depends. That's the silicon lottery.
Is it worth the extra 70$? Imo not... I just wanted decent DDR4 3600 CL18 sticks instead of going for a cheapo 3200 CL16 set, and tried to squeeze a little extra out of them just for fun. But if I see the performance increase between the stock timings and my current tweaked timings, don't expect huge improvements from going to even stricter timings. I'd spend that money on something else. But that's just me

btw, you could also try bumping your infinity fabric to 1866MHz and your RAM speed to twice that. I did a few experiments with that, but I think I was a bit restricted by my older X470 motherboard, which was only certified up to 3600MT. Yours might do better.

I will definetly try cranking up my IF and let you know how that goes. For the time being, tho, I feel as if I've tinkered with enough of my components for a while and I plan on enjoying some good old games. I will come back to tinkering and let you know how the IF clocks respond to my gentle touch. Thank you again for the insight, I wouldn't have touched ram timings this much without it.

 

[examen1996]

Hello @Freeco , @JasonToddRH

The ram is not running well at the fast preset, for me neither does the safe, the one that runs best is at 3600 is 16-19-19-19-39-64-561
However I have tried and tried all the combinations in the world ) , and i can tell you for a fast, it's a rabbit hole, you can run the ram with gdm off(shaves 0.5 from that pesky latency), that need 16-20-19-19-39-64-561 at 1.37 V.
The one that i find to be the most ,,bang for the buck'' is 3733 16-20-20-20-40-64-570 at 1.4V , gdm off. It's not that much voltage and the system is running fine.
The most extreme however(subjective , but whatever) is running 3800 at 16-21-19-19-39-62-580 at 1.44V GDM OFF , there are a few other settings that need to be done for this to be stable.
3600:

http://imgur.com/a/6UXBBCe

3733:

http://imgur.com/a/TXiQhPN

3800:

http://imgur.com/a/J81KoXo

 

[mxking035]

Hello @Freeco , @JasonToddRH

The ram is not running well at the fast preset, for me neither does the safe, the one that runs best is at 3600 is 16-19-19-19-39-64-561
However I have tried and tried all the combinations in the world ) , and i can tell you for a fast, it's a rabbit hole, you can run the ram with gdm off(shaves 0.5 from that pesky latency), that need 16-20-19-19-39-64-561 at 1.37 V.
The one that i find to be the most ,,bang for the buck'' is 3733 16-20-20-20-40-64-570 at 1.4V , gdm off. It's not that much voltage and the system is running fine.
The most extreme however(subjective , but whatever) is running 3800 at 16-21-19-19-39-62-580 at 1.44V GDM OFF , there are a few other settings that need to be done for this to be stable.
3600:

http://imgur.com/a/6UXBBCe

3733:

http://imgur.com/a/TXiQhPN

3800:

http://imgur.com/a/J81KoXo

Hi, can you share your subtimings?

 

[examen1996]

Hi, can you share your subtimings?

Well you can find most of them in the photo albums, currently I'm running at 3733 because of summer + no ac . But yeah sure you will have them on monday

 

[mxking035]

Well you can find most of them in the photo albums, currently I'm running at 3733 because of summer + no ac . But yeah sure you will have them on monday

I'm using 4 sticks ram: 2x8Gb E-Die and 2x8Gb B-Die all same as part numbers as yours. I cannot lower my latency any further. Which one is more important is it latency or the read-write-copy?

 

[Waldorf]

Having build a few R5/7 rigs in past 12-14 month, some info:

Having ram on the motherboard list is just showing you they ran it, not that its guaranteed to work.
All Gskills i tried (even had different chips on em) were listed and none of the Gb X570 would even POST,
but perform at XMP level on other brands (x570).
the two Corsair kits (different chips) not listed for the GB that virtually everyone recommended NOT to get,
worked flawlessly.
All kits were XMP (Not A-XMP), so the oc isnt even verified for amd boards.

Same part number doesnt mean jack for ram, i can have 2 different chips from different brands, so mixing is not a good idea,
nor does it mean all will perf identical, and with AMD having a little more sensitivity on the mem controller,
i dont recommend running sticks with different dies , even if its the same brand.

One Corsair kit (3600-18-22-22-42) with Samsung die would only do XMP (1.35),
when the next kit (same model) with Micron chips can do 3600-16-19-19-36/1T@1.33v on the same board/bios/cpu.

and none of the "3 digit" settings from the calculator i tried (no matter if oc or safe), so they just staid on xmp level.

i do recommend trying 1.35v first, and this allows you to stabilize oc/timings later on with a bit added voltage.
do have good airflow, especially on Bdies. they clock good but hate higher temps and usually get unstable pretty quickly
when you put load on the rig and it warms up.

 

[examen1996]

There are a few things that impact your settings , first and most important is that you have 4 dims , which are far harder to drive. You will need to change cad bus block settings (use the alt1 from my screenshot) and a few voltages(first pic was what i used to achieve stable 3800 oc.
I helped another guys recently , same ram(2 dims only) a 3600 on a b450 mortar, he also couldn't go lower than 16-21-21... .

 

[Waldorf]

Going past 3600 wont make sense for most ppl just looking to get what they paid for (for different reasons).

I dont remember where i saw it, but some engineering info from a manufacturer (ram?) was talking about mem controller/IF,
of 100% (zen2) cpus sampled, + 60% were able to reach 3600, about 20ish for 3733, und around 5-10% able to hit 38xx).
as well as multiple test/reviews i saw (on on guru3d) showing the sweet spot is 3600 and lowering timings as much as possible.

not talking about someone dropping +300$ on some 64gb Bdie dominator kit...

 

[examen1996]

Going past 3600 wont make sense for most ppl just looking to get what they paid for (for different reasons).

I dont remember where i saw it, but some engineering info from a manufacturer (ram?) was talking about mem controller/IF,
of 100% (zen2) cpus sampled, + 60% were able to reach 3600, about 20ish for 3733, und around 5-10% able to hit 38xx).
as well as multiple test/reviews i saw (on on guru3d) showing the sweet spot is 3600 and lowering timings as much as possible.

not talking about someone dropping +300$ on some 64gb Bdie dominator kit...

I agree with you but, we are talking about an overclock , on a techsite. The fact that after 3600 de gains are minimal is well known but I can say/show that you do gain some fps by going all in with the ram. I stopped experimenting after a while because it was getting annoying . But then again, if your desktop is a gaming machine only and you dont do work stuff with it, a little bit of tunning can get you a some more point in 3d mark and 4-5 fps . If that's worth the huge amount of time needed for stability testing ....well that is subjective.

 

[Waldorf]

oh sure, but even if your on a site about tweaking/tuning, doesnt mean everyone knows the same amount.
lots of ppl come here/read post because of a google search/recommendation etc, and not because they wanna post about their latest LN2 session

just have seen too many times where less experienced users (i shall say) are glancing over posts and dont realise, YMMV..

 

[Taraquin]

Going past 3600 wont make sense for most ppl just looking to get what they paid for (for different reasons).

I dont remember where i saw it, but some engineering info from a manufacturer (ram?) was talking about mem controller/IF,
of 100% (zen2) cpus sampled, + 60% were able to reach 3600, about 20ish for 3733, und around 5-10% able to hit 38xx).
as well as multiple test/reviews i saw (on on guru3d) showing the sweet spot is 3600 and lowering timings as much as possible.

not talking about someone dropping +300$ on some 64gb Bdie dominator kit...

Actually with the latest Agesa 3733 seems possible with Zen 2 for most people with 2 dimms, 4 dimms is harder unless you have t-top. 3800 is also possible for a lot more people now than at release. they gains from 3600 to 3733 aren`t great, but after some testing I see basically the same performance in games with 3600cl14 as 3733cl16, both with tuned subs + GDM off, 1t, cl14 is much easier to achieve on 3600 than 3733 (on my rev E), but it seems like some games and apps prefers the extra BW over slightly lower latency.

 

[mxking035]

There are a few things that impact your settings , first and most important is that you have 4 dims , which are far harder to drive. You will need to change cad bus block settings (use the alt1 from my screenshot) and a few voltages(first pic was what i used to achieve stable 3800 oc.
I helped another guys recently , same ram(2 dims only) a 3600 on a b450 mortar, he also couldn't go lower than 16-21-21... .

Thank you for sharing. With this 3733 does your CPU temp increase for daily use?

Update: I just tried 2 sticks only and tRCDRD got below 21 like 16-19-19-36 but found out no benefits lowering it too much and just increasing Write but sacrifice Read, stability and voltage requirement. I tried this with my self limit of max vDram=1.37v and vSoC=1.1v ProcODT=53.3Ω and recommended tRP not lower than tRCDRD (I read somewhere):

Read:Write:Copy:Latency - All tested in TM5, MemTest64 (10 Loops) and my Game - no error, no game crash
16-21-21-21-42 - 1.36v soc=1.050v : 54,939 — 53,123 — 55,626 — 68.4 — same setting as 4 sticks, little loose secondary & tertiary timing
16-21-21-21-42 - 1.36v soc=1.050v : 55,400 — 53,300 — 55,800 — 68.2 — approx only, no screenshot - optimize secondary and tertiary
16-20-20-20-40 - 1.36v soc=1.075v : 55,499 — 53,394 — 56,564 — 67.8 — same 2nd and 3rd timing
16-19-16-19-36 - 1.37v soc=1.100v : 55,186 — 53,697 — 56,531 — 67.8 — same 2nd and 3rd timing

found out 16-20-20-40 is better so i will not get too far ratio from tRC and tRFC as it is limited by ram. Also this tRC and tFAW crashing my mid game (Dota2 - Max Settings) even it pass the TM5 and MemTest64. So I better stay with higher on that as it has small benefits in latency.

For me, better to optimized the secondary and tertiary timings before proceeding to lower primary as it increase more read/write/copy with lower latency and get it stable with ProcODT/RTT with lower voltage of DRAM and SoC requirements.

This is only my experience and this post also is my first search in google overclocking by searching my ram like examen1996 so big help from Freeco on examen1996 also learned here about importing thaiphoon is better and more stable baseline.

3600 16-21-21-21-42 tRFC=575 @1.36v

3600 16-20-20-20-40 tRFC=558 @1.36v

3600 16-19-16-19-36 tRFC=558 @1.37v

 

[examen1996]

Thank you for sharing. With this 3733 does your CPU temp increase for daily use?

Update: I just tried 2 sticks only and tRCDRD got below 21 like 16-19-19-36 but found out no benefits lowering it too much and just increasing Write but sacrifice Read, stability and voltage requirement. I tried this with my self limit of max vDram=1.37v and vSoC=1.1v ProcODT=53.3Ω and recommended tRP not lower than tRCDRD (I read somewhere):

Read:Write:Copy:Latency - All tested in TM5, MemTest64 (10 Loops) and my Game - no error, no game crash
16-21-21-21-42 - 1.36v soc=1.050v : 54,939 — 53,123 — 55,626 — 68.4 — same setting as 4 sticks, little loose secondary & tertiary timing
16-21-21-21-42 - 1.36v soc=1.050v : 55,400 — 53,300 — 55,800 — 68.2 — approx only, no screenshot - optimize secondary and tertiary
16-20-20-20-40 - 1.36v soc=1.075v : 55,499 — 53,394 — 56,564 — 67.8 — same 2nd and 3rd timing
16-19-16-19-36 - 1.37v soc=1.100v : 55,186 — 53,697 — 56,531 — 67.8 — same 2nd and 3rd timing

found out 16-20-20-40 is better so i will not get too far ratio from tRC and tRFC as it is limited by ram. Also this tRC and tFAW crashing my mid game (Dota2 - Max Settings) even it pass the TM5 and MemTest64. So I better stay with higher on that as it has small benefits in latency.

For me, better to optimized the secondary and tertiary timings before proceeding to lower primary as it increase more read/write/copy with lower latency and get it stable with ProcODT/RTT with lower voltage of DRAM and SoC requirements.

This is only my experience and this post also is my first search in google overclocking by searching my ram like examen1996 so big help from Freeco on examen1996 also learned here about importing thaiphoon is better and more stable baseline.

3600 16-21-21-21-42 tRFC=575 @1.36v
View attachment 167871View attachment 167872

3600 16-20-20-20-40 tRFC=558 @1.36v
View attachment 167873View attachment 167874

3600 16-19-16-19-36 tRFC=558 @1.37v
View attachment 167875View attachment 167876

View attachment 167877View attachment 167878

Heh, yeah , lets just say i wa sa enthusiast noob and a lot of people had a lot of patience with m ) .
How did you get those increadible read scores my dude , is it because of the cpu ?, i thought it only affected the write performance....now i want new cpu )

Regarding the temps, I think it might got hotter , but like I already had a NH-d14 in a phanteks p400a full of fans ..... my cooling is a bit overkill for this cpu, I prepared for it as it was my last cpu (Fx-8120 ) )

What I can tell you and this is more than bro science is that ram overclocking is a bit of masochist work, sometimes you can get lower numbers (timings i mean) and the results can be worse.
Or, my favorite , lower timings, better results but you get issues like micro-stutters in some games.(looking at you Forza horizon 4).

One great example of my pain is you last set of scores, 3600 16-19-16-19-36 , I bet my cheap ass rx 580 that if I put those numbers in my bios , in about 3-4 hours of memtest, without any errors, some games will simply stop working, AC Odyssey being one.

I took a few photos of what I'm running now, with voltages and all, I have been running this combination for more than 3 months with games, Linux with VMs that had hardware pass-through enable(got my masters degree this year ) and a boatload of 3dmark().

Another config that boosts your performance is GDM, if you manage to disable you GDM, my boy you are going to share some time off that pesky latency, and it's also a indicator that you really do have a stable oc on your ram .
But be aware that when you disable it, you will probably have to change your cad bus drive settings, I always do 60-20-20-40 (as shown here: https://docs.google.com/spreadsheets/d/1c3RTF_ZPjep-Zfimgoca2Ef1gSjZM0rSHVWLkknbfUI/edit#gid=0 ) .

If you want to continue your OC experiment, let me know, I would gladly help you with what I can

 

[mxking035]

Heh, yeah , lets just say i wa sa enthusiast noob and a lot of people had a lot of patience with m ) .
Another config that boosts your performance is GDM, if you manage to disable you GDM, my boy you are going to share some time off that pesky latency, and it's also a indicator that you really do have a stable oc on your ram .
But be aware that when you disable it, you will probably have to change your cad bus drive settings, I always do 60-20-20-40 (as shown here: https://docs.google.com/spreadsheets/d/1c3RTF_ZPjep-Zfimgoca2Ef1gSjZM0rSHVWLkknbfUI/edit#gid=0 ) .
If you want to continue your OC experiment, let me know, I would gladly help you with what I can

Thanks for the link, I'll check this one. I didn't tried yet disabling GDM but i will try It with that Cad Bus.
My Cad Bus in Auto is 60-20-20-20 in 2 sticks, and change it to 24-30-24-24 as recommended by 1usmus and 24-20-24-24 for 4 sticks which is also same as Auto.

Regarding read, before I tried only primary and got around 51Gb/s, and from XMP 72ns to 70.6ns
After changing tCKE from 9 to 1, tRDRD-SCL & tWRWR-SCL from 6,6 to 5,5, tFAW: 38 to 35, Read jump from 51gb.s to 56gb/s in 4 sticks.

I notice that your tCKE is 0, can you try it for 1. I will try your settings and check if the CPU play the role on this.

 

[examen1996]

I forgot to add this to my earlier post :

http://imgur.com/a/b0G9iON