Overclock.net banner
Cancel
Create thread New Forums

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

93 reading
Jump to Latest
14,341 - 14,360 of 31,499 Posts
#14,341 ·
Taraquin said:
Asrock B650m hdv so no eclk :/ My old cpu had an avg of CO -41 stable in occt avx2 large extreme, worst core -32. That is a very good bin of cores. New cpu has 1 core failing -25, but passing -20, all other cores stable at -30. Will test further. In my opinion avg of -10 or worse=bad, up to -30 avg, over -30 good. Too bad I cant test eclk, the guys I swapped with dont have eclk either.
Click to expand...
Yeah, would have been super interesting seeing how this would translate to eclk potential
 
#14,342 ·
RosaPanteren said:
Yeah, would have been super interesting seeing how this would translate to eclk potential
Click to expand...
Should work very well in theory, too bad I dont have any friends close that can test :/
 
Save
Reply Quote Rep+ Rep+
#14,343 ·
Melan said:
Latency is kinda wack but it can be typical aida things. Try PyPrime 2.2 instead.

Make sure you have Memory Context Restore, Powerdown Mode and Fast Boot disabled while you're doing stuff to memory for the sake of proper memory training. Later you can reset them to default/turn on for faster boot times.
Also you can disable Geardown mode. Don't forget to run a memory stress test to see if it's actually stable (should be, though)

Voltages could also come down. There's no need to run them high, which might help with stabilising/higher FCLK which also might be why you have higher latency right now.
These are mine just for reference
VSOC 1.1v
CLDO VDDP 1v
MEM VDD 1.4v
MEM VDDQ 1.3v
CPU VDDIO 1.2v
Click to expand...
how you could lower the VSOC? I tried to go below 1.2V in the BIOS, but the motherboard sets it to 1.2. My board is a Gigabyte Aurus Elite AX, BIOS version F8d.
 
#14,345 ·
mrcardio33x@gmail.com said:
Yeah I'm not gonna take out/ breakdown my DIMMS.. I'm not that curious..

I doubt thaiphoon burner is reading it wrong. There's clearly a difference ( tRDRDSD / tWRWRSD as noted ) here among 2X16GB DIMMS..
I suspect most are probably Single Rank but not all.

If anybody else has (2) Ranks in Thaiphoon Burner 2X16GB willing to verify the physical presence of the chips on PCB, knock yourself out lol.
I'll just trust Thaiphoon Burner here.
Click to expand...
@mrcardio33x@gmail.com

I’ve had the same thoughts about these sticks

since I’ve changed now I cant get the same write and copy, even with tighter timings

These are pics of mine, same sticks I believe
 
#14,346 ·
mrcardio33x@gmail.com said:
Yeah I'm not gonna take out/ breakdown my DIMMS.. I'm not that curious..

I doubt thaiphoon burner is reading it wrong. There's clearly a difference ( tRDRDSD / tWRWRSD as noted ) here among 2X16GB DIMMS..
I suspect most are probably Single Rank but not all.

If anybody else has (2) Ranks in Thaiphoon Burner 2X16GB willing to verify the physical presence of the chips on PCB, knock yourself out lol.
I'll just trust Thaiphoon Burner here.
Click to expand...
OK, how tight tRFC can you get. Low density chip tends to have shorter refresh time.
 
#14,347 ·
Nyemasanya said:
how you could lower the VSOC? I tried to go below 1.2V in the BIOS, but the motherboard sets it to 1.2. My board is a Gigabyte Aurus Elite AX, BIOS version F8d.
Click to expand...
I just set it to 1.1v.
There is a description in bios for VDDIO which mentions VSOC though:
VDDIO > VDDP + 0.1V
VDDIO + 0.1V > VDD_SOC
So VSOC has to be somewhere in the middle between VDDIO and VDDP. I don't know why your board autocorrects it.
 
#14,349 ·
RosaPanteren said:
@mrcardio33x@gmail.com

I’ve had the same thoughts about these sticks

since I’ve changed now I cant get the same write and copy, even with tighter timings

These are pics of mine, same sticks I believe
Click to expand...
Ic’s on both sides

idk if this makes them dual rank though as it has been said these are down binned 32gb sticks

but it’s something odd with these

and their cheap, so I might just order a new pair as I broke one of the them removing the heatspreader
 

Attachments

#14,350 · (Edited)
lostking said:
OK, how tight tRFC can you get. Low density chip tends to have shorter refresh time.
Click to expand...
tRFC may go a bit lower but 120ns is stable and likely not going much lower.
tRDRDSD / tWRWRSD do not post below "7"
Many perhaps Most other 2X16GB Adie can set tRDRDSD / tWRWRSD to (1) as they aren't being used.
There must be a fundamental design difference here, rather its "Single Rank" vs. "Dual Rank" I'm not sure but Thaiphoon Burner would seem to indicate "Dual Rank".
I do know that these are damn good DIMMS / Can run 6200/ CL26 @ 1.55v VDD maybe a bit lower, etc. Only cost ~ $100 bucks to boot :)
Image
Image
 
Save
Reply Quote Rep+ Rep+
#14,351 ·
I'm having a weird stability issue that seems tied to my RAM/RTX 4090 with Oculus Link/Air Link & Virtual Desktop on my Quest 3. I get severe video encoder artifacting with my seemingly tried and true 6000 MT/s RAM settings (passes several tests including Karhu with no errors).

I am using this kit with the attached settings:

Are you a human?

www.newegg.com www.newegg.com
Image


When I swapped out my RTX 4090 for a 4070, the 6000 MT/s settings work - maybe the 4090 was defective? Likely not, as any other encoder task works fine, and the issue goes away with the 4090 when I set the RAM speed to auto settings (5200 MT/s - anything higher or manually tweaked seems to not work). There were also a few other users who had identical artifacting issues that were resolved by lowering their DDR5 RAM speeds. FWIW, my system is not stable at any speed over 6000 MT/s, assuming the CPU is to blame but not sure.

Any suggestions on what to try with either my current kit, or a different kit to rule out some sort of defect with the RAM? Really at a loss here for what could be happening. Not sure if the RAM could pass stress tests but still be defective, but also not sure why my 4090 can't operate this workload when the RAM is set above 5200 MT/s.
 
#14,352 ·
RosaPanteren said:
Ic’s on both sides

idk if this makes them dual rank though as it has been said these are down binned 32gb sticks

but it’s something odd with these

and their cheap, so I might just order a new pair as I broke one of the them removing the heatspreader
Click to expand...
It is pretty sad that these are dual rank dimms. Usually, single rank a-die can go beyond 8000+. For dual rank, you can get 7200 if lucky.
 
#14,353 ·
Nitemare3219 said:
I'm having a weird stability issue that seems tied to my RAM/RTX 4090 with Oculus Link/Air Link & Virtual Desktop on my Quest 3. I get severe video encoder artifacting with my seemingly tried and true 6000 MT/s RAM settings (passes several tests including Karhu with no errors).
View attachment 2647831

I am using this kit with the attached settings:

Are you a human?

www.newegg.com www.newegg.com
View attachment 2647830

When I swapped out my RTX 4090 for a 4070, the 6000 MT/s settings work - maybe the 4090 was defective? Likely not, as any other encoder task works fine, and the issue goes away with the 4090 when I set the RAM speed to auto settings (5200 MT/s - anything higher or manually tweaked seems to not work). There were also a few other users who had identical artifacting issues that were resolved by lowering their DDR5 RAM speeds. FWIW, my system is not stable at any speed over 6000 MT/s, assuming the CPU is to blame but not sure.

Any suggestions on what to try with either my current kit, or a different kit to rule out some sort of defect with the RAM? Really at a loss here for what could be happening. Not sure if the RAM could pass stress tests but still be defective, but also not sure why my 4090 can't operate this workload when the RAM is set above 5200 MT/s.
Click to expand...
Watch for DIMM temperature. Make sure the 4090 is not toasting your ram.
 
#14,354 · (Edited)
Image


Error 3 hours in when I was at work. What's the most likely scenario from above? I guess going back to TM5 might have its benefits with somewhat of an indicator at what might be wrong. Karhu is just usually quicker than TM5 at erroring out.

I know I got some advice to use lower voltages, but I was actually struggling. Not sure if it's the combo of 7950x3D and a-die at this speed/timings.

Is anyone else with a 7950x3D running a-die at 8000? What's your timings/experience? Especially with voltages. Nitro enabled or disabled?
 
#14,355 ·
zgigi666999 said:
I have 2x24 m die hynix gskill memory 6400 wondering who can help me tight my timings i got a 7800x3d chip Zen-Timings-Screenshot hosted at ImgBB

i posted my link of currently zentimings if anyone could help me tight the ram i would appreciate that very much :) thank you in advance
Click to expand...
Image
Image


if you set XMP, reboot and overclock the secundariew and tertiary timings + memclk/uclk in 1:1 its looks something like this:
 
#14,356 ·
lostking said:
It is pretty sad that these are dual rank dimms. Usually, single rank a-die can go beyond 8000+. For dual rank, you can get 7200 if lucky.
Click to expand...
No, for 2dcp boards these sticks are really good

I had them at 6400cl26 with 54.4ns latency

Probably could have gotten them to a decent 6600 1:1 if I had more time

so for 1:1, if these are dual rank, these are probably the best of what you could get

+8000 on 2dcp is still really hard for 2dcp, and really not that much of a deal for 1 ccd chips

so imho, if 2dcp and single ccd, these sticks rock
 
#14,358 ·
Audioboxer said:
View attachment 2647832

Error 3 hours in when I was at work. What's the most likely scenario from above? I guess going back to TM5 might have its benefits with somewhat of an indicator at what might be wrong. Karhu is just usually quicker than TM5 at erroring out.

I know I got some advice to use lower voltages, but I was actually struggling. Not sure if it's the combo of 7950x3D and a-die at this speed/timings.

Is anyone else with a 7950x3D running a-die at 8000? What's your timings/experience? Especially with voltages. Nitro enabled or disabled?
Click to expand...
Just a sanity check at this point, do you have and PBO (especially CO) on that you can temporarily disable to isolate it down to just being memory at this point? Those mem temps reported if the chips are anywhere close to the temp sensor look nice and low, one blip after 13k% with nice low temps would have me looking at the CPU instead unless it can be eliminated from any doubt.
 
#14,359 · (Edited)
metalshark said:
Just a sanity check at this point, do you have and PBO (especially CO) on that you can temporarily disable to isolate it down to just being memory at this point? Those mem temps reported if the chips are anywhere close to the temp sensor look nice and low, one blip after 13k% with nice low temps would have me looking at the CPU instead unless it can be eliminated from any doubt.
Click to expand...
CPU on default for now!

It's likely either voltage or resistance this far in that has a minor issue. Or it's a single timing that needs a small change.

Balancing voltage on a 7950x3D seems to be a wee game in of itself when it comes to things that will stress the IMC. Especially on a-die where you can be trying rather tight timings and a tRFC at 120ns.

I'm running something else with the VDDQ a bit closer to VDD and some minor changes to resistances.

Unlike my DDR4 days where I pretty much slowly learned what all the resistances did, there's some new ones on DDR5 I haven't done much reading on yet. Like DramDqDs which I feel just from playing with two kits of memory for a limited time has a decent impact on SNR (signal noise ratio).

As always other people's timings aren't just going to plug and play at times and there seems to be an even smaller pool of people running 8000 a-die with a 7950x3D compared to 6400 1:1.

And finally, your usual reminder that even the Karhi 10k% club might not be enough :p Then again most serious people OCing use multiple stability tests. I just find starting with Karhu usually finds instability quickest. Then I'll do TM5 and y-cruncher.
 
#14,360 ·
lostking said:
It is pretty sad that these are dual rank dimms. Usually, single rank a-die can go beyond 8000+. For dual rank, you can get 7200 if lucky.
Click to expand...
I suppose it would be "sad" if for example you were running a dual CCD CPU like a 7950X3D but seeing as the 7800X3D doesn't have enough cores to take advantage of 2:1 mode it therefore makes more sense to run 1:1.

For the 7800X3D I'd much rather have a better performing kit at 1:1.
You may get a bit better bandwidth at 8000MT/s but unless you have a GENE and or get lucky with the CPU that's likely not happening.
 
Save
Reply Quote Rep+ Rep+
14,341 - 14,360 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