Overclock.net banner
Cancel
Create thread New Forums

*Official* Intel DDR5 OC and 24/7 daily Memory Stability Thread

29 reading
Jump to Latest
4.7M views 34K replies 1.2K participants last post by  remittor  
#1 · (Edited)
Let it begin, boys...

Rectangle Font Material property Parallel Screenshot



@cstkl1 @Nizzen @PhoenixMDA



Font Number Terrestrial plant Slope Screenshot


Musical instrument Musical instrument accessory Electronic instrument Automotive design Office supplies


Font Publication Fashion accessory Gadget Audio equipment



Product Rectangle Font Circuit component Carbon


Circuit component Rectangle Audio equipment Font Electronic instrument


Triangle Asphalt Road surface Rectangle Wood



www.adata.com

ADATA

The future is ours to define. ADATA provide SSD,HDD,USB,DDR5,external ssds and power banks. We believe our attitude and vision can determine how advanced the future will be.
www.adata.com www.adata.com
gskill.com

G.SKILL International Enterprise Co., Ltd.

G.SKILL specializes in high-performance and overclocked memory, and provides PC component and peripheral products designed for overclockers, PC gamers, professionals, and enthusiasts around the world.
gskill.com gskill.com

Arni90 said:
There's this guide: MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper
While it's written for DDR4, and the specific timing recommendations don't really apply, the principle and order of tightening timings remains the same (lower one timing at a time, test stability, rinse and repeat)

Here's what's changed in terms of DDR5 OC on Intel CPUs:
  • tCL needs to be a multiple of the gear ratio, in gear 2 that means it has to be an even number, while gear 4 means it needs to be a multiple of 4
  • tRCD and tRP are now separate timings
  • tREFI can now run as high as 262143, this is the only timing where higher = better.
  • Tertiary timings for Intel on DDR5 are now at minimum 7, rather than 4.
  • Due to the increased number of banks on DDR5, _dg timings have a significant impact on performance, _sg timings doesn't make the same kind of difference.
  • tRFCpb / tRFCsb is a new timing, which refreshes a single bank, this allows the CPU to still access memory while data is being refreshed. Tightening it works similarly to tRFC.
  • Some testmem5 presets work better than others, ironically enough I've found 1usmus_v3 to work quite well.
  • tWR is controlled by tWRPRE and tWRPDEN, if you want a formula: tWR = tWRPDEN - (tCWL + 8)
  • tWTR_S and tWTR_L are controlled by tWRRD_dg and tWRRD_sg respectively. tWRRD_dg and _sg can have a significant impact on performance, and they can often only tighten 2-6 ticks from stock timings.

For the Hynix ICs you're using, here are my suggestions as a starting point:
VDD/VDDQ: 1.35V
tCL 11 ns
tRCD 13 ns
tRP 13 ns
tRAS 13 ns
tRFC 130 ns
tRFCpb 90 ns

Note that I listed these values in terms of absolute time, and you'll have to multiply them by the actual frequency to get a suggested timing.

Example: tCL at DDR5-5600
DDR5-5600 has an actual frequency of 2.8 GHz
tCL is suggested 11 ns

tCL in ticks = 11 ns * 2.8 GHz = 30.8
Rounded up to the nearest multiple of 2, we get tCL 32 as a starting point.

For the rest of the timings, we then get
tRCD 37
tRP 37
tRAS 37
tRFC 364
tRFCpb 252

I use GigaHertz here because Giga (10^9) complements nano (10^-9) so that I don't need to multiply or divide by 1000


Hope this is at least somewhat helpful to you.
Click to expand...
 

Attachments

  • Rep+
  • Helpful
  • Love
Reactions: trailerguyhd, Idol, BoBtimus and 19 others
Save
Reply Quote Rep+ Rep+
#23,901 ·
affxct said:
TX VDDQ above 1.45V is kinda sus, but hardly ever required. MC is VDD2 and we technically do not yet know what the limit is, but I would suggest not going above 1.5V for 24/7 use, not that you'd need more anyway. What your RAM can do at 1.5 VDD/VDDQ is usually the optimal limit, and having VDD2 above VDD isn't necessary.

I have run 1.6 VDD2 before and didn't notice any degradation, but I didn't feel super comfortable with it.
Click to expand...
I can agree, after 1.5V VDD/VDDQ it won’t scale anymore and not only that but it’s getting even more unstable from heat increasement.

I am stuck at the moment, nothing I change will make me more stable and I can’t tighten my timings at all. I really hope that watercooling the dims will make a big difference , it’s my last hope.
 
  • Rep+
Reactions: affxct
Save
Reply Quote Rep+ Rep+
#23,902 ·
Vhypur said:
damn maybe just me but those last couple offsets besides the last look super high, but u also have a leaky cpu right idk if that means anything? for me my offsets on LLC4 are like -30 to -40 max - though i'm sure alot of stuff can change this i've gotten a couple points multiple times to where if I even lower them 0.001 it will fail stuff like cb15 and or gb3 until i raise it back up, i guess the question is what benchmarks do you run for stability? if i were to try some offsets anywhere near those it would instantly crash/error on benchmarks

here is the cpu i'm using atm though which is my bros (old pic without offsets)
View attachment 2646471
Click to expand...
This is far from leaky, it's a nice one!
Last point I need to be high, because the TVB boost is 6.2 GHz. You can test it with Geekbench 6. Just run it several times, if you always get result, the low load boost is good.
 
  • Rep+
Reactions: Vhypur
Save
Reply Quote Rep+ Rep+
#23,903 ·
affxct said:
I have run 1.6 VDD2 before and didn't notice any degradation, but I didn't feel super comfortable with it.
Click to expand...
VDD2 is made by an external Voltage regulator on the board, I don't think it could degrade the CPU, or the RAM at all.
For me it scales well until 1.537V. But that depends on the RAM VDD too.
 
  • Rep+
Reactions: affxct
Save
Reply Quote Rep+ Rep+
#23,904 ·
TeamGroup Xtreem 48GB DDR5 8000 C38 1.4v

Anyone else have these sticks?

These are optimized timings, but definitely reduced or min/maxed. I'm curious what others have gotten with them + 790 Apex Encore.


Image
 
Save
Reply Quote Rep+ Rep+
#23,909 ·
tibcsi0407 said:
VDD2 is made by an external Voltage regulator on the board, I don't think it could degrade the CPU, or the RAM at all.
For me it scales well until 1.537V. But that depends on the RAM VDD too.
Click to expand...
Exactly; however I know there was a limit that individuals always used to quote with DDR4. However, the fact that we had 1.6V XMP kits suggests our CPUs (and boards) should be capable of withstanding at least 1.6V from the DDR rail.
 
N
#23,911 · (Edited)
Clock em up 9000c38 gaming 😎🤟
 
  • Rep+
  • Wow
  • Love
Reactions: Kaliz, Thunderclap, Jor3lBR and 3 others
Save
Reply Quote Rep+ Rep+
#23,913 ·
Finally installed my Iceman Cooler Direct Ram Block today with Arctic TP-3 Thermal Pads. Did some quick testing with some 8000c36 profile and only 1.45V Vdd/Vddq. Will go back to my old 8200c34 profile and test this again but so far the contact and my temps are pretty decent. Using a Mora-420 with Dual D5, Iceman Direct Die and Iceman Direct Ram block. After 1h of y-cruncher the Delta was 2C. Water: 26.6C Sticks: 28.5C
Image

Image
 
#23,914 ·
cLxFPS said:
Finally installed my Iceman Cooler Direct Ram Block today with Arctic TP-3 Thermal Pads. Did some quick testing with some 8000c36 profile and only 1.45V Vdd/Vddq. Will go back to my old 8200c34 profile and test this again but so far the contact and my temps are pretty decent. Using a Mora-420 with Dual D5, Iceman Direct Die and Iceman Direct Ram block. After 1h of y-cruncher the Delta was 2C. Water: 26.6C Sticks: 28.5C
View attachment 2646632
View attachment 2646633
Click to expand...
Nice build! What backplate are you using with those screws there? Looks like a button head with a spacer?
 
  • Rep+
Reactions: cLxFPS
Save
Reply Quote Rep+ Rep+
#23,915 ·
affxct said:
Exactly; however I know there was a limit that individuals always used to quote with DDR4. However, the fact that we had 1.6V XMP kits suggests our CPUs (and boards) should be capable of withstanding at least 1.6V from the DDR rail.
Click to expand...
Yes. Actually that’s where the value gets yellow color in ASUS bios. Or maybe a bit higher.
It’s CPU dependent how good it scales with memory speed .
 
Save
Reply Quote Rep+ Rep+
#23,917 ·
affxct said:
View attachment 2646222 View attachment 2646221

I know it's been a while, but I recently decided to rebuild a modest rig. I picked up some relatively cheap Trident Z5 6400C32 A-die and a relatively cheap Maximus Z690 Formula.

Settings:
7000 32-42-42-42-28-2T (Maximus Tweak Mode 2)
ctlvrefup - 126
RttWr - 34
RttNomRd - 48
RttNomWr - 48
RttPark - 40
RttParkDqs - 34
SA PLL - 1.02V
MC PLL - 1.02V
VCCIN AUX - Auto
VCCSA - 1.1V
TX VDDQ - 1.3V
VDD2 - 1.475V
VDD/VDDQ - 1.5V
VPP - Auto

The 13900K I'm using is an SP 101 (110, 83) and has MC SP 75. I have it under a Thermal Grizzy contact frame. I haven't really attempted 7200 or 7400 because I don't believe the board has the capability for it, but it'll be interesting to see how 24Gbit M-die performs. At 7000, this board doesn't exhibit any re-train weirdness with Hynix 16Gbit A-die.
Click to expand...
henlo:)
 
#23,920 ·
CR1 ? Asus equivalent of 1T ?

If so isn’t DDR5 made to use 2T ?

I was running my old DDR4 with 1T and the difference was pretty big in games vs 2T.
 
Save
Reply Quote Rep+ Rep+
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling