Overclock.net banner
21 - 40 of 41 Posts

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #21 · (Edited)
I tried some of the timings from your calculator reference. They helped, thank you.

As for now, I did 30 minute OCCT test with new setting (which are faster).
2516366

2516367

Should I try to lower tRP and tRC even more, -1 or -2?
-1 is booting Windows, did test with AIDA, did not see any difference. Did not tried OCCT burn.

Should I try to lower tWRWRSD to 6, so it is same as tWRWRDD?
(Calcutator suggested to raise tWRWRDD to 7 as tWRWRSD)

Could you check if my timings are well set, alligned with each other (all those complex rules :p )?
I read that sometimes higher timing is better to be alligned well with other timings.

Any more suggestions what to change?

If there won't be any suggestions I will go to benchmarking section to check and compare results (also in some game).


As for 3600/1800 testing - I'll try to do that, but as far as I remember, I can tell now when I tested it with 4000 / 1800 and 4000 / 1866 results were in favor of 1866. Results are in my second post here ( C and D setups).

Edit: BIOS hard crash / hard reset when those tRDRDSD, tRDRDDD, tWRWRSD and tWRWRDD all set to 4.
2516401
 

·
Robotic Chemist
Joined
·
3,886 Posts
Should I try to lower tRP and tRC even more, -1 or -2?
-1 is booting Windows, did test with AIDA, did not see any difference. Did not tried OCCT burn.
You might be able to get away with lower values. 12 and 40 seems possible. Do stability testing, of course. I would give it overnight before deciding it was really stable.

Should I try to lower tWRWRSD to 6, so it is same as tWRWRDD?
(Calcutator suggested to raise tWRWRDD to 7 as tWRWRSD)
I would not mess with tRDRDSC, etc. 1-5-4-1-7-6 looks good. :)

Any more suggestions what to change?
You look pretty close to optimal. None of my suggestions below will make a significant performance improvement. But if you are having fun... ;)

Lower tRRDL? Maybe even down to 4.
Try tRDRDSCL and tWRWRSCL at 3.
tWTRL at 10?
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #23 ·
Benchmark results

I did testing with new settings to compare results. Here they are:

2516620


Best results are in bold.

The most important are 3733MHz OC2 and 3733MHz OC4 as I plan to run OC4 daily.
Strange thing happened during comparing them as OC4 should have more performance, but got worse results in Tomb Raider. I waited some time to cool of my GPU (which is a bit OCed and it's boost is dependent on temperatures) - they and archieved same results as OC2 for 1080p.

The thing bugging me is why read is lower on more OCed OC4 profile than on OC2 (everything else is better) - any ideas?
2516630

What do you think?

Benchmark methodology, timings and details

Profiles tested:

3600MHz CL14, tweaked
3733MHz CL14, tweaked OC2
3733MHz CL14, tweaked OC4
3733MHz CL14, XMP all auto
4000MHz CL14, XMP all auto, FCLK/IF 1867MHz
(order of testing: 1.->5.->4.->2.->1.)

Profiles timings (in order written above):
2516625

2516626

2516627

2516628

2516629


Apps tested:
CPU-Z
AIDA64 memory test
Tomb Raider
(order of testing: 1.->2.->3.)

Tomb Raider was tested at two settings:

1) 1920x1080, default medium, TAA, DLLS off
2516621

2516622

2) 3840x1600, ultra all maxed out, TAA, DLLS off
2516623

2516624


The most pushed overclock was 3733MHz CL14, tweaked OC4 profile, which is stable, confirmed with:

Room temperature: probably ~23'C (rainy summer day, not hot)
Idle temps: 43'C / 41.5'C
30 minute OCCT burn: passed, no errors or WHEAs, temps 61.5'C / 60'C
30 minute COD: Warzone: no errors or WHEAs, temps 56'C / 54.5'C
45 minute Cyberpunk 2077: no errors or WHEAs, temps 55'C / 53'C

What I couldn't not push more above this (OC4) profile:
tCL 14 -> 13 or 12, not tested
tRCDWR 15 -> 14, errors
tRCDRD 15 -> 14, errors

tRP 13 -> 12 Windows loaded, not tested more
tRAS 29 -> 28 and lower, not tested
tRC 42 -> 41 and lower, not tested
tRRDS 4 -> 3, Motherboard does not show 3
tRRDL 8 -> 4, Windows loaded, one of many settings changed that could have led to error while testing tigher timings
tFAW 16 -> lower, not tested
tWTRS 4 -> 3, not tested
tWTRL 8 -> 4, not tested
tWR 12 -> 10, Windows loaded, one of many settings changed that could have led to error while testing tigher timings
tRFC 256 -> lower, not tested
tRDRDSCL 3 -> 2, not tested
tWRWRSCL 3 -> 2, not tested
tCWL 14 -> lower, not tested
tRTP 8 -> 5 or 6, BIOS automatically changed 5 to 6 and 6 was one of many settings changed that could have led to error while testing tigher timings
tRDWR 8 -> 7, 6, not tested
tWRRD 4 -> 3, not tested
tRDRDSC 1 -> 0, not tested
tRDRDSD 5 -> 4, Windows loaded, one of many settings changed that could have led to error while testing tigher timings
tRDRDDD 4 -> 3, not tested
tWRWRSC 1 -> 0, not tested
tWRWRSD 7 -> 6, Windows loaded, one of many settings changed that could have led to error while testing tigher timings
tWRWRDD 6 -> 5, not tested
tWRWRSD 7 and tWRWRDD 6 -> both to 5, BIOS hard crash and reset
tCKE 1 -> 0, not tested
Any other -> not tested

Lowest stable voltage -> 1.44V
 

·
Robotic Chemist
Joined
·
3,886 Posts
The thing bugging me is why read is lower on more OCed OC4 profile than on OC2 (everything else is better) - any ideas?
I am not sure I trust AIDA64's benchmark enough to focus on it too much, even with repeats, your result ranges do overlap. :(

Very nice data set though, I should do that too. If you compare the standard deviations the difference is more than 3x a standard deviation, so it does look like it could be real. It is possible tWTRS and/or tWTRL is slightly too low, causing more correctable errors during reads (slower, but no errors during testing). It could also be tRFC, that might be a critical ns, but if so any changes in temperature or voltage would likely push it one way or the other.

Edit:
Actually, if I do the math properly, (delta the averages instead of the max values) then the difference is just under 3x the standard deviation of OC4 and under 2x the standard deviation of OC2. Because the slower ones are likely due to Windows doing something in the background the highest values do seem more significant, but it still seems like a somewhat unreliable dataset to get a good understanding of memory timings on this scale. More data needed! :geek:
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #26 · (Edited)
Update -> managed to get 3733MHz CL14-14-14-28, temps need to be lower to be stable

1)
I am running now OC4.1 profile with slightly lower tRFC comprared to OC4
, because I used this calculator:

It seems stable.
Room temperature: probably ~23'C (summer day, not hot)
30 minute OCCT burn: passed, no errors or WHEAs, temps 60.5'C / 59,5'C

OC4.1 ZenTimings table:
2518214

2)
XMP timinings at lower RAM frequency for G.SKILL F4-4000C14D-32GTZN

I have discovered possible XMP timings at lower RAM frequency in OCCT -> System Info -> Memory -> Certified Profile section.


4000MHz - 14-15-15-35 (XMP default)
3866MHz - 14-15-15-34
3733MHz - 14-14-14-33

Here is the photo:
2518215

3)
I managed to run at 3733MHz CL14-14-14-28, same tweaked subtimings with provisory cooler.


Comparing to OC4.1
tRCDWR 15 -> 14
tRCDRD 15 -> 14
tRP 13 -> 14
tRAS 29 -> 28
RAM Voltage 1.44V -> 1.45V (gonna try 1.44V with new cooler, for now I did not test 1.44V)
Rest is the same.

WITHOUT COOLER
Error after few minutes of OCCT burn at arouund 55-58'C.

WITH COOLER (big fan aimed at open case)
RAM average peak temps 50.50'C, max 50.75'C
So cooling got me -10'C average peak temp.
30 minute OCCT burn: no errors or WHEAs.

Conclusion - CL14-14-14-28 possible, errors due to high temp, not IMC or memory sticks.

Benchmarks:
Tomb Raider 1080p medium - same fps, 213
AIDA64 - comparable results

OC5 ZenTimings table and cooler photo:
2518216

2518217

4)
As for now I considering adding RAM cooling fan / cooler:

I am worried it will be ugly.
I am interested only in 3-pin DC / 4-pin PWN option. No molex or sata power supply.
May be RGB with motherboard A-RGB / D-RGB connection, I don't want additional software or USB connections.
Black, grey, silver or blue colour varations.

I also read somewhere that lowering temps on RAM got someone 0,1% lows in game higher.

As far as I could find some coolers:
Corsair Dominator Platinum RGB (CMDAF2) - edit: RGB control works only with Corsair Commander / Link, not with motherboard channels. Default colour: red IIRC.
Corsair Vengeance Airflow (CMYAF) - but it's ugly and red...
G.SKILL Turbulence II / III - blah, Molex connector... And ugly, only white LED (I need blue or RGB)
Some watercooling or heatspreader kits - I don't want, they could void warranty, I also don't want to risk damaging my memory

Maybe buy some of them and swap fans?

Any options, suggestions?
 

·
Registered
Joined
·
309 Posts
Nice work so far :)

I use the g.skill cooler, but swapped in gelid 50mm fans. The gelids are superior to 4 other sets of 50mm fans I tested. No RGB, but the fans impellers are white.
DSCF5926_edited.jpg
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #28 · (Edited)
Nice work so far :)

I use the g.skill cooler, but swapped in gelid 50mm fans. The gelids are superior to 4 other sets of 50mm fans I tested. No RGB, but the fans impellers are white.
View attachment 2518315
What about temperatures and noise with RAM fans off / on low / on high?
Are you happy with noise?
Did you test temp changes when gaming or OCCT burn?

What are those Gelids and does this sort out the problem with molex, only 4-pin PWM?
 

·
Registered
Joined
·
309 Posts
They have 3 pin connectors, and while I don't stress test my components nearly as hard as some do, I did see a minimum of 5C temp reduction in games (likely closer to 10C). I havn't had cooling issues with my bdie kit. So havn't really made good measurements with them. My cjr kit is the reason I have the cooler. They got unstable past 35 degrees. But I figured it won't hurt to use them on my current kit. Especially since I run 1.54V now.

I use manual fan controllers... Sometimes simple is better, for me, anyhow. The gelids both have really great airflow compared to the other 4 sets I have, but are also much quieter. In the winter time I can hear the fans on high, medium speed they are essentially silent in a silent house. But this time of year with ac running in my house, I cannot hear them at all on high. But being that I have a manual controller I don't know what rpm or if I'm even maxing them out. But I can say they flow plenty of air.

For reference, out of 5 sets they are the quietest by far (really happy), and 2nd best flowing. Best flowing being easily 4x louder, and also much thicker (20mm?)

I bought 1 fan set at a time. Not being happy with each set, gelid was the last I got, and they are very nice.
Screenshot_20210722-093054.png
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #30 · (Edited)
Thank you. As I can't get any new or used G.SKILL Turbulence in Poland I will try to buy - Corsair Dominator Platinum RGB.
I managed to talk with one guy who has one used, we will probably make a deal when he comes back from Holidays next week.

I plan to:
-use it probably without RGB as I don't have Corsair controller and default lightning if red IIRC (I need blue); fans are translucent so maybe blue RGB RAM will light them a bit
-try it if it helps with temps (60'C -> 50'C at load) and if it's loud
-it is 4-pin PWM, but 2 pins are used for RGB without controller connection, so it is probably DC... But I think I could just change fan header mode in mobo to DC and then control it, as I do not need 10-30% power, more like 40-60% (don't want it loud also)
-if fans are loud I will try putting your fans, some screwdriver and maybe they will fit.

If I can reach 50'C max temps then I could go with 3733-14-14-14-28 instead of 3733-14-15-13-29, which would make my timings more alligned.
 

·
Registered
Joined
·
309 Posts
Sounds like a plan. Are your operating temps at 60C? Mine are at 33! Thats with 4 sticks at 14-15-14-14-40 3800 1.54v
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #32 ·
Sounds like a plan. Are your operating temps at 60C? Mine are at 33! Thats with 4 sticks at 14-15-14-14-40 3800 1.54v
3733-14-15-13-29-subs tweaked
1.44V
Idle about 42'C.
When gaming about 56-58'C
When OCCT burn stops at max 60,5'C
No errors or WHEAs

3733-14-14-14-28-subs tweaked (same)
1.44V
Errors above 50'C, I need to cool them to run this perfect timing.

My previous XMP default sticks at:
3600-14-15-15-35

1.45V
Ran at similiar temps, idle 42'C, games 56-58'C
No errors or WHEas

Don't know, maybe bad airflow (3x140mm front intake, 1x140mm rear exhaust, 3x120mm+360mm rad top exhaust). Front intake is mesh, not glass. They all run at 20-30% low speeds when idle and about 30-45% when gaming. But still I see this strange. Maybe there is some hot "air pillow" behing and above my PC, since it is under the desk near the back and side wall, the desk is confirmed to wall so no air can escape to the top, see here:
2518353

2518354

2518355
 

·
Robotic Chemist
Joined
·
3,886 Posts
Do you have a measurement for internal air temperature? 3733-14-15-13-29-subs tweaked is pretty good though, nothing to get too worried about. 60°C is not dangerous or anything.

I see up to 50°C with no airflow, but that is using an open bench "case".
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #34 ·
I'm not worried, my previous DDR4 3600 CL14 sticks worked for 6 month at same temps without errors :D
In my previous PC I had front 280mm Kraken X52, so it blowed at DDR3 RAMs, but I did not check their temps back then. Worked flawlessly for about 7 years.
I read somewhere that DDR4 max is about 85'C not to damage sticks.

I did not measue air temp, but here are all my temp sensors (10 hours of only idle and Chrome browsing, no gaming):
2518362
 

·
Registered
Joined
·
309 Posts
Yeah, 85 is pretty high. I figured it'd be up there, but I think you want to keep it as cool as can be with over-volting. Especially seems high when my cjr kit wouldn't run over 35 lol

I have reverse flow on mine, front 280mm rad as exhaust, rear intake, 2 top intakes, 1 top front 120mm rad as exhaust, and another bottom front 120mm rad as exhaust... Even under max load my vrm, ram, chipset temps never see more than 45C
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #36 ·
Temps most of the time of 1.5h playing COD: Warzone:
CPU 73'C
GPU 72'C
RAM 57'C

Here are all the max temps:
2518379

I won't reserse my cooling mostly due to "I don't want to touch it again" and LED fans could look worse.
I plan to try Curve Optimiser to try this undervolting.
But firstly I want to try cooling RAM to see if it could help with lowering it's temps.
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #37 · (Edited)
1)
OC4.1 - first crash

Today I was playting Warhammer 40,000 Battlesector (v 1.0).
First mission - no problems
Second mission - crash (game closed) in first turn, then PC reset in second turn, then PC reset in first turn (I played 3 times)

Then I switched to 3733 14-15-15-35 (all auto), 1.45V - finished second mission without problems.

It is strange since I played much more e.g. COD: Warzone which seems to be more demanding game and no errors, game or PC crashes.

So it is worth noting that WH40K: Battlesector could be either bugged game or good for testing memory.

2)
WHEA category has "errors" and "operational".
I don't have any errors, but I have some operational.
Are WHEA Operational something to look up or I don't have to worry at all, just ignore them?
2518440
 

·
Registered
Joined
·
309 Posts
As far as pbo curve goes... I've found the gains to be super minimal in contrast to the huge pain it's been for me to work with. Stable setting would result in back to back crashes roughly 2 days later?? I simply resulted to under-volting to the most stable voltage for a given cpu load (40-50%) performance does drop off at higher loads, but I really don't care. Its still very good, and I just don't ever load my cpu more than %50. But I do see performance lift at medium loads.

For whea, I'd at least look and see what is being logged.

For me, ghost recon is the most problematic title. Everything else will run great except for it, lol. I had to increase my trfcs a hair and go to 1.54v from 1.53, havn't had issues with anything in quite some time.

Have you found 1.44v to be most stable or is it an issue of temperature? I would be running close to 1.5v I would think? Maybe more?
 

·
Registered
5950X
Joined
·
32 Posts
Discussion Starter · #39 ·
Did the PBO Curve put temps down?
I plan to try Curve Optimiser to lower my temps, not sure about using PBO and changing it's limits.

As for WHEA Operational - I only see:
ID 42 (keywords: WHEA Event Log Enties)
ID 5 (keywords: none)

I lowered from 1.55V XMP default to lowest stable which was 1.44V. Maybe adding 0.01V add stability with this game or further overclock.
The issue for further overclocking is for sure temperature - I checked this with open case and big fan aimed at case. Max temps dropped from 60'C to 50'C max and I did not get errors in OCCT burn at 14-14-14-28 (which I do not use now, errors in Warhammer game was with 14-15-13-29, which is also rock stable in Warzone etc.). I am waiting at the moment for RAM cooling fans.
 

·
Registered
Joined
·
309 Posts
Huh, that's odd with the whea. Lowest stable I saw for me was 1.53, not sure if due to low temps or due to using 4 sticks instead of 2. 1.54 required by that 1 game.

I recon the fans should help
 
21 - 40 of 41 Posts
Top