Overclock.net banner
Cancel
Create thread New Forums

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

81 reading
Jump to Latest
26,501 - 26,520 of 31,499 Posts
#26,502 ·
droric said:
Based on my experience I would suspect the memory controller/training issue. I found that a more reliable way of finding out of the memory controller is stable for a given SoC is to do the following. Start 3dmark Speedway and then start ycruncher Vt3. Let them both run for about 20 min, if no errors then restart the PC and be sure a retraining takes place. Then go ahead and repeat the process 3 times and you should be reasonably sure the PC should boot reliably.

I was able to pass 10k% Karhu with too low of vSOC only to have it error out in 2 minutes with Speedway + VT3 at the same time. This really stresses the IF/Memory controller. The other thing that threw off my testing before was having PBO/CO enabled. I found that with a fast memory OC my 9950x needs positive CO on core 0 and 4 to remain stable. Without the memory OC the processor will run about -10 offsets on those cores.
Click to expand...
you were right. It was vsoc too low. Seems y cruncher VT3 is not a good test, OCCT catch the errors with 15 min, if i set all timings auto it catch error instantly. now with 1.26 vsoc is fine, no errors with OCCT.

worth run daily 6400 with 1.26 vsoc or is it too much? better go with 6200 that i can run with arround 1.2vsoc?
 
#26,504 ·
Noreboots said:
you were right. It was vsoc too low. Seems y cruncher VT3 is not a good test, OCCT catch the errors with 15 min, if i set all timings auto it catch error instantly. now with 1.26 vsoc is fine, no errors with OCCT.

worth run daily 6400 with 1.26 vsoc or is it too much? better go with 6200 that i can run with arround 1.2vsoc?
Click to expand...
Good to know that OCCT caught it faster. I never tried OCCT with that config for vSOC but I will for next time since I am planning on getting a 9950x3d next week and will need to start this process over again.

I'm running 1.28vSOC daily but then again I don't care if any hardware dies since I'll just replace it. Plus I've been overclocking my machines for 25 years and have yet to have a single failure aside from spinning HDDs and a single SSD that was 5+ years old. Are you settings your VDDGs manually? On my board ASUS way overvolts the VDDGs and I have more stability with them at 950/950 then 1100/1100 that ASUS was setting. Also my VDDP only needed to be .95v. I think minimizing both voltages helped my system stabilize a higher FCLK and may have improved the memory overclocking performance.
 
#26,507 ·
peroni said:
New MSI bios released to support 9950X3D and 9900X3D CPUs.

Since bios profiles cannot be used across versions, this is a bit of a nightmare with recreating profiles and retest everything. I got my weekend set out.
Click to expand...
A nice new option appeared, of course not mentioned in the released notes.
I took an nvram export and compared to the previous one, there a few more changes but not visible in the UI

Only posting in this thread since it is memory related

Text Screenshot Font Number Software
 
#26,508 ·
Noreboots said:
guys i need some help, @Veii @anta777

ive tested this profile yesterday night and it passes ryzen3d anta777 TM5 for 1:30hr. Today morning after cold boot i decided to test again and it is unstable. Errors arround 30min sometimes earlier.

ive tested looser trtp 15 twr 60, scls and trdwr/twrrd auto but no success.

ive tested vddp variations, increasing vdd misc thinking it is some fclk instability but nothing get rid of the errors.

HDV is borderline mobo, sometimes it works and after some reboot ram is unstable, ***. Asrock really need to improve bios.

increasing vdd/vddio/vddq to 1.35v doesnt help either, increasing vsoc to 1.26v doesnt help

also ive tested IMC with y cruncher VT3 and its fine with 1.24v, passed for many hours without error. But every memory test is leading to error TM5/Karhu

another weird thing is that GDM enable dont run, 6400 GDM ON no matter what i do always error with 2 seconds. GDM disabled otherwise runs the tests much longer.

View attachment 2700286
Click to expand...
Many IODs can’t do 3200 UCLK. Try 1.28 vSOC and for the FCLK, if you haven’t already done so, increase both VDDG to 0.93V.

Just for now, increase VDD to 1.35. You can try lowering it again later.
 
#26,509 ·
I think I've minimized my SoC voltage as low as it'll go for 6400/2133 on this CPU. Started with 1.3v SoC, Karhu was stable at 1.26v over night but testing with y-cruncher VT3 failed within the first 2 iterations consistently. Working back up to 1.275v SoC I've passed 20 iterations of VT3 and OCCT CPU + RAM (Large, Extreme, Variable, AVX2) for 1 hour. Will do more extensive testing later.

Next attempt will be to get 6400/2133 GDM Off working and stable and if all that works, eventually try to stabilize 2200 FCLK since I haven't tested increased CCD or IOD voltages, only auto which currently sets to 0.903v for my board.

Technology Number Software Screenshot Video Game Software
 
  • Rep+
Reactions: UTVOL06
Save
Reply Quote Rep+ Rep+
#26,510 ·
anta777 said:
1. IF=2133*32 read (=4266 DRAM)
2. Vsoc, vddp, vddio closer stock voltage
Click to expand...
Я сделал некоторые измерения. Так вот, на 6000 производительность ПК лучше в тех задачах, которые мне нужны. Какой совет вы мне дадите? Поставьте на 6000 и не насилуйте память? Память с заводским XMR 6400.
 

Attachments

#26,513 ·
an.perov1970 said:
Я сделал некоторые измерения. Так вот, на 6000 производительность ПК лучше в тех задачах, которые мне нужны. Какой совет вы мне дадите? Поставьте на 6000 и не насилуйте память? Память с заводским XMR 6400.
Click to expand...
6400 should be the best (if it's stable). I'd keep fine tuning the 6400 setup.

Corona benchmark might stress CPU more than RAM, possibly more limited by a power limit when running 3200 UCLK, limiting CPU power. Check CPU PPT limit % in HWInfo64. Increase in PBO settings if needed.

I'd run these very safe (but better than what you have) values:
tRP 36
tRAS 59
tRC 95
tWTRS 6
tWTRL 24
tRDRDSCL 7
tWRWRSCL 17
tRDWR 17
tWRRD 5
tRFC 448

VDDQ 1.35V
VDDIO 1.35V

Work on getting the above 100% stable before worrying about benchmarks.
 
#26,514 · (Edited)
bradsm87 said:
6400 should be the best (if it's stable). I'd keep fine tuning the 6400 setup.

Corona benchmark might stress CPU more than RAM, possibly more limited by a power limit when running 3200 UCLK, limiting CPU power. Check CPU PPT limit % in HWInfo64. Increase in PBO settings if needed.

I'd run these very safe (but better than what you have) values:
tRP 36
tRAS 59
tRC 95
tWTRS 6
tWTRL 24
tRDRDSCL 7
tWRWRSCL 17
tRDWR 17
tWRRD 5
tRFC 448

VDDQ 1.35V
VDDIO 1.35V

Work on getting the above 100% stable before worrying about benchmarks.
Click to expand...
It's not about the tests. The PC is loaded with Corona from morning until late evening, that's why the tests are with it. We need the maximum result from the PC, especially for this task. But without excessive claims on the components. So: squeeze the maximum out of the possible, without crossing the line.
 
#26,516 ·
an.perov1970 said:
Thank you. I will definitely try it.

PPT - unlimited. 230 W
Click to expand...
I'd just work on optimising stability and voltages. Should make for a good, high performing and efficient setup.

Another good option would be to run DDR5-6200 and try and get 2167 FCLK stable. That's what I run myself. Latency is almost as good as 2067 but much better than 2133 for some reason.
 
#26,517 ·
bradsm87 said:
I'd just work on optimising stability and voltages. Should make for a good, high performing and efficient setup.

Another good option would be to run DDR5-6200 and try and get 2167 FCLK stable. That's what I run myself. Latency is almost as good as 2067 but much better than 2133 for some reason.
Click to expand...
Yes, I am calm and confident about the 7900x. But the second PC with 9900x does not want to work 6400 1:1. It works on 6200 for me.
 

Attachments

#26,519 ·
an.perov1970 said:
Yes, I am calm and confident about the 7900x. But the second PC with 9900x does not want to work 6400 1:1. It works on 6200 for me.
Click to expand...
If you can do 6200 1:1 with 1.2vsoc, chances are high you should be able to do 6400 1:1.. you already tried 1.3vsoc and it wasn't stable?
 
26,501 - 26,520 of 31,499 Posts
About this Discussion
31K Replies
1.1K Participants
Last post:
gupsterg
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