Can't increase tREFI

[Od1sseas]

My default tREFI is at ~7000.
100K/80K/65K all unstable in OCCT (Just a few minutes into the test). What's crazy is that even 30K and even 15K!! is unstable!
Only the default tREFI can pass OCCT 30 minutes! What's the issue! Please help me

System:
i5 13600KF
Z790 PG Lightning
2x16GB 72000Mhz C34
RX 6700 XT

 

[ir_cow]

What is your ram temp at? The higher the tREFI value, the more likely to get errors. After all this is the refresh time (in cycles) for the cells. Temperatures play a vital role here. I wouldn't suggest going over 50K. To be honest, outside of benchmarks this value alone isn't going to make one bit of difference in games and probably not at all for the RX 6700 XT.

 

[Od1sseas]

What is your ram temp at? The higher the tREFI value, the more likely to get errors. After all this is the refresh time (in cycles) for the cells. Temperatures play a vital role here. I wouldn't suggest going over 50K. To be honest, outside of benchmarks this value alone isn't going to make one bit of difference in games and probably not at all for the RX 6700 XT.

How do i check my ram temps

 

[ir_cow]

HWInfo (LINK). Like I said, not worth the effort. You are 100% GPU bound in pretty much everything unless your playing on low graphical settings. Not going to see a difference in real world applications. Waste of time to play around with a OC.

 

[Od1sseas]

HWInfo (LINK). Like I said, not worth the effort. You are 100% GPU bound in pretty much everything unless your playing on low graphical settings. Not going to see a difference in real world applications. Waste of time to play around with a OC.

True. But i want that AIDA64 latency number

HWInfo (LINK). Like I said, not worth the effort. You are 100% GPU bound in pretty much everything unless your playing on low graphical settings. Not going to see a difference in real world applications. Waste of time to play around with a OC.

Btw what temperature is good

 

[ir_cow]

Do you know if the memory is stable in the first place before you adjusted the values?

 

[Od1sseas]

Do you know if the memory is stable in the first place before you adjusted the values?

Yes. 30 Minutes Anta777 Extreme, 30 Minutes OCCT and 30 Minutes with TPU's MemTest64

 

[ir_cow]

True. But i want that AIDA64 latency number

Now you sound like me. I OC just because, but I also don't leave it enabled once I'm done submitting to HWBOT for my personal computer unless it servers a purpose.

Btw what temperature is good

Your target is 45c. The higher the tREFI, the lower temps need to be. XMP will error out around 60-65c btw.

 

[phanbuey]

Current Raptor lake memory controller kind of sucks with trefi. even if you pass 24 hours above the 'stock' settings -- give it some days and it will eventually crash.

 

[Od1sseas]

Now you sound like me. I OC just because, but I also don't leave it enabled once I'm done submitting to HWBOT for my personal computer unless it servers a purpose.


Your target is 45c. The higher the tREFI, the lower temps need to be. XMP will error out around 60-65c btw.

Bruh, i'm already at 55C only 5 minutes into Last Of Us.
It all makes sense now...

 

[ir_cow]

Not to discourage you, but you will have to play around as no "one value" fits everyone. Each motherboard (SKU) is different in terms of how well it handles memory adjustments. Some don't even run 7200 at all. CPU IMC is also factor. Now add in memory too, and its just a guessing game really. I normally will bump the memory voltage 0.05V and see if that helps in my quest. Really it can be combination of all of the above.

 

[Od1sseas]

Current Raptor lake memory controller kind of sucks with trefi. even if you pass 24 hours above the 'stock' settings -- give it some days and it will eventually crash.

Should have stayed with team AMD

 

[ir_cow]

Bruh, i'm already at 55C only 5 minutes into Last Of Us.
It all makes sense now...

Well that is probably the reason.

 

[Od1sseas]

Well that is probably the reason.

I wouldn't say "probably"

 

[phanbuey]

Should have stayed with team AMD

they purposely don't even expose TREFI as an option for the same reason.

 

[nguyen]

on my Samsung 6000MT CAS36 kit I reduce the RAM voltage by 0.05V to make them run cooler, so I can increase the Trefi to 111111

 

[Od1sseas]

on my Samsung 6000MT CAS36 kit I reduce the RAM voltage by 0.05V to make them run cooler, so I can increase the Trefi to 111111

0.05V makes that much of a difference?

 

[basco]

i don´t know if all ddr5 ram has a temp sensor but on ddr4 ya can see that in aida64 engineer version if ya have one:


2nd pic should be from my mainboard sensors but don´t know how accurate

 

[nguyen]

0.05V makes that much of a difference?

Yup, easily 5-7C temp reduction

 

[Wirko]

If left at default, does the IMC automatically adjust refresh interval depending on temperature?

 

[Od1sseas]

If left at default, does the IMC automatically adjust refresh interval depending on temperature?

No, it's static

Yup, easily 5-7C temp reduction

Are u sure? Stock voltage for VDD/VDDQ was 1.41V. Dropped it down to 1.38V and it still hits 63C three minutes into OCCT

 

[nguyen]

No, it's static


Are u sure? Stock voltage for VDD/VDDQ was 1.41V. Dropped it down to 1.38V and it still hits 63C three minutes into OCCT
View attachment 305842

You probably need active cooling then

 

[freeagent]

Move some air through your case, that should help quite a bit.

 

[chrcoluk]

Personally as tempting as it is I wouldnt boost trefi, certainly not to max.

The problem with trefi is its temperature sensitive and can bite you if things get hotter (summer, gaming etc.) and can also do so via silent data corruption.

If left at default, does the IMC automatically adjust refresh interval depending on temperature?

Bios will based on memory clock. It will increase the auto trefi value on higher clock speeds.

 

[iameatingjam]

Idk. 100,000 didn't work for me but 60,000 did. I'm only at 6400 mts though. Didn't I hear something about high mts + high trefi being a bad combo?

Also unrelated but I just updated my bios (want a 14700k) and put in a new ssd ( my fifth... yeah I know... I went overboard) and all of a sudden my 30-40-40 config doesn't work anymore. And I tested it extensively before. Now it wont even pass memtest64, the easiest test.

Would a new m.2 on chipset lanes affect the cpu memory controller like this? Or maybe it was the bios update? Anyway 32-38-38 seems to be working for now and is only slightly slower but yet to do more than one hour of anta777 extreme.