Overclock.net banner
5,861 - 5,880 of 7,835 Posts
Are your RTLs training the same when temps are different? I know I have had that in the past when RTLs trained different it would cause issues.

Need to look at them in memtweakit to see all 3 if using DR. Should all be the same on each channel. like 71/71/71 and 73/73/73 as an example.

If they change from boot to boot it usually is a sign of being on the edge of stability.
I didn't think to look. They have never been equal though. It's always been on the edge. Passed yCruncher and Karhu and no crashes, but 4266 was very unstable, so 4200 was right at the limit.

The bad thing is I have to drop all the way to 3800 to get any significant reduction in VDDQ and that needs to be 1.5v. I just got done setting up a 3800 tune which seems stable. Just run that for now while I try to decide what I want to replace it with.

In the end it doesn't really matter what the cause is, it just doesn't work anymore. I tried everything I know to do. Increased SA to 1.35v, loose timings at 4200, even tried 1.6v VDDQ. Reflashed the bios even. Nadda. Insta fails in yCruncher FFT.
 
I didn't think to look. They have never been equal though. It's always been on the edge. Passed yCruncher and Karhu and no crashes, but 4266 was very unstable, so 4200 was right at the limit.

The bad thing is I have to drop all the way to 3800 to get any significant reduction in VDDQ and that needs to be 1.5v. I just got done setting up a 3800 tune which seems stable. Just run that for now while I try to decide what I want to replace it with.

In the end it doesn't really matter what the cause is, it just doesn't work anymore. I tried everything I know to do. Increased SA to 1.35v, loose timings at 4200, even tried 1.6v VDDQ. Reflashed the bios even. Nadda. Insta fails in yCruncher FFT.
Yep, sounds like what I experienced. And then I resorted to physical changes as well, like constant reseating of the CPU/RAM and retightening of screws, etc. Nothing seemed to help. If anything, reseating the CPU too much even caused it to start BSODing instead.

Eventually, I took the idea of trying out a new board from someone else who mentioned experiencing something similar in a past generation, and lo and behold, I managed to achieve stability again.

In addition to the CPU and RAM, I think it's important to cool the VRMs on these boards as well, since they are what stabilize the voltage distribution. Alas, not a single DDR4 board has a monoblock option, which bothered me a lot. So I just bought my own VRM waterblock off of eBay instead.
 
Just tried it again just for the heck of it.

Trained at 23C water.
Passed FFT (didn't go further).
Fired up chiller.
Tried FFT again with 17C water. Still passed.
Re-trained. Couldn't make it to the desktop to even look at the RTL's! Sigh...

Anyway this is kinda interesting. Here are the 2 tunes. Now failed one on the right.

Font Screenshot Parallel Rectangle Number
Font Screenshot Parallel Rectangle Number


But of more interest, here is the difference in SotTR 1080p lowest. Just 5 fps average and just 3 fps on the min.

3800CL14
Font Technology Engineering Software Slope


4200CL15
Font Technology Engineering Screenshot Software
 
Just tried it again just for the heck of it.

Trained at 23C water.
Passed FFT (didn't go further).
Fired up chiller.
Tried FFT again with 17C water. Still passed.
Re-trained. Couldn't make it to the desktop to even look at the RTL's! Sigh...

Anyway this is kinda interesting. Here are the 2 tunes. Now failed one on the right.

View attachment 2603444 View attachment 2603445

But of more interest, here is the difference in SotTR 1080p lowest. Just 5 fps average and just 3 fps on the min.

3800CL14
View attachment 2603447

4200CL15
View attachment 2603448
You do have RTL Training enabled or set to Dynamic, right?
 
  • Rep+
Reactions: YoungChris
You do have RTL Training enabled or set to Dynamic, right?
This is what I have enabled,

Font Screenshot Software Technology Electronic device


And one more, 3600CL13, for anyone contemplating how hard to push speeds and voltages. A little more of a hit. Total of 13 fps average and 10 fps on the min from 3600 to 4200. But more than half of that can be made up by just running 3800.

Font Screenshot Parallel Number Rectangle


Font Technology Engineering Screenshot Software
 
I know on my board I have RTL training enabled and then the mode set to auto, if I try to use dynamic it will fail to post.
 
I know on my board I have RTL training enabled and then the mode set to auto, if I try to use dynamic it will fail to post.
I previously tried switching that setting and the round trip latency setting to auto, but no luck.

---------------------------------------

On a more general note, I still wonder if I have had this problem all along since normally I don't train the memory with the chiller running. And the fact that I've had trouble switching back to 4200 when I had set it up to bench at 4266 temporarily. It was the same problem then with FFT insta failing or not even loading windows. I just kept fiddling with it until it stabilized, but maybe the water warmed up since I would have turned off the chiller at the end of the bench session.

Once it's trained the low temps don't seem to bother it. Sub 10C water and FFT still didn't fail, but trained at 20C+ water.

Rectangle Font Software Screenshot Technology
 
This is what I have enabled,

View attachment 2603453

And one more, 3600CL13, for anyone contemplating how hard to push speeds and voltages. A little more of a hit. Total of 13 fps average and 10 fps on the min from 3600 to 4200. But more than half of that can be made up by just running 3800.

View attachment 2603454

View attachment 2603455
This is just the one spot..I think you already know…you have to disable
Fast boot or set slow training. I have to train mine every boot …to stay at 4200 …I could try setting the rtls at 71s…but then I’d have to enable retry fail to ensure it’ll boot. One out of 10 boots it’ll miss train and 71/71/73 or some weird combo and I just reboot and retrain 🤷‍♂️
 
This is just the one spot..I think you already know…you have to disable
Fast boot or set slow training. I have to train mine every boot …to stay at 4200 …I could try setting the rtls at 71s…but then I’d have to enable retry fail to ensure it’ll boot. One out of 10 boots it’ll miss train and 71/71/73 or some weird combo and I just reboot and retrain 🤷‍♂️
Well actually once it was trained it just kept working. I even rebooted a couple of times (with fast boot on to avoid training), and then took that screenshot above with it still passing FFT. Once it fails to train properly with sub ambient water temps and becomes unstable, there seems to be no way to recover without warming the water back up.

I tried all 3 training methods it had, fiddling with all relevant voltages (and some that really aren't), enable/disable VDDQ training, etc...

Edit: Nevermind, I found slow training

Edit 2: Nope didn't help
 
Just tried it again just for the heck of it.

Trained at 23C water.
Passed FFT (didn't go further).
Fired up chiller.
Tried FFT again with 17C water. Still passed.
Re-trained. Couldn't make it to the desktop to even look at the RTL's! Sigh...

Anyway this is kinda interesting. Here are the 2 tunes. Now failed one on the right.

View attachment 2603444 View attachment 2603445

But of more interest, here is the difference in SotTR 1080p lowest. Just 5 fps average and just 3 fps on the min.

3800CL14
View attachment 2603447

4200CL15
View attachment 2603448
Whats your voltages in 4200mhz and 3800mhz?
VCCSA, VCCSQ and DRAM?
 
Not to but in on this other guys issue but I did have a question.
So let's say I have RTL training turned on and my board trains the RTL's such as
69/69
25/25
71/71
25/25
Now let's say after using the PC for like a week and having rebooted maybe 5 times and those RTL's have not changed would it be beneficial for me to change the 69/69 values to match the 71/71 values and then turn all training off?
 
Well actually once it was trained it just kept working. I even rebooted a couple of times (with fast boot on to avoid training), and then took that screenshot above with it still passing FFT. Once it fails to train properly with sub ambient water temps and becomes unstable, there seems to be no way to recover without warming the water back up.

I tried all 3 training methods it had, fiddling with all relevant voltages (and some that really aren't), enable/disable VDDQ training, etc...

Edit: Nevermind, I found slow training

Edit 2: Nope didn't help
I don’t know anything about chillers/water. When it first time boots the water is warmer? Maybe have it train cold water somehow?? 🤷‍♂️😂
 
Whats your voltages in 4200mhz and 3800mhz?
VCCSA, VCCSQ and DRAM?
1.25v SA for all of them

3800: 1.55v DRAM, 1.5v VDDQ TX
4200: 1.62v DRAM, 1.58v VDDQ TX
 
Not to but in on this other guys issue but I did have a question.
So let's say I have RTL training turned on and my board trains the RTL's such as
69/69
25/25
71/71
25/25
Now let's say after using the PC for like a week and having rebooted maybe 5 times and those RTL's have not changed would it be beneficial for me to change the 69/69 values to match the 71/71 values and then turn all training off?
Meh you could try, running 4000-4133? May not like, you would need to get it train at those values then turn off…I never have luck setting them manually…
 
I don’t know anything about chillers/water. When it first time boots the water is warmer? Maybe have it train cold water somehow?? 🤷‍♂️😂
When I train it without the chiller, the water will be a couple of degrees above ambient (22-23C) which is enough for training to succeed. The CPU is simply warmer. Why that would make this kind of difference, I have no idea. I've heard of this type of thing being an issue for LN2 guys, but that's much much colder obviously.

I'm pretty sure this has been an issue all along, but possibly the minimum temperature for training to succeed has increased over time.
 
Meh you could try, running 4000-4133? May not like, you would need to get it train at those values then turn off…I never have luck setting them manually…
No, I run 4 SR dimms at 3600 14-14-14-34.
 
1.25v SA for all of them

3800: 1.55v DRAM, 1.5v VDDQ TX
4200: 1.62v DRAM, 1.58v VDDQ TX
Oh you would need more sa I’m sure…up to 1.35. I do 4200 at 1.52vdimm, 1.35 vddq, 1.33 sa (1.35 set)
 
Not to but in on this other guys issue but I did have a question.
So let's say I have RTL training turned on and my board trains the RTL's such as
69/69
25/25
71/71
25/25
Now let's say after using the PC for like a week and having rebooted maybe 5 times and those RTL's have not changed would it be beneficial for me to change the 69/69 values to match the 71/71 values and then turn all training off?
No. The first channel should be less by up to a couple of clocks. I would set the them in BIOS to 69/71 and disable RTL training.
 
Oh you would need more sa I’m sure…up to 1.35. I do 4200 at 1.52vdimm, 1.35 vddq, 1.33 sa (1.35 set)
I did need that much on previous bios releases, but v1.19 was magic and let me only use 1.25v SA even for 4200. I tried increasing it all the way to 1.4v, but this didn't help my cold water training issue. Pushing VDDQ TX right to the limit of 1.61v doesn't help either (anything higher always fails to post), or increasing DRAM voltage. I could drone on and on about all the things I tried today without success. :) even worse than I already have...
 
No. The first channel should be less by up to a couple of clocks. I would set the them in BIOS to 69/71 and disable RTL training.
So basically just like it already is.
I guess those RTL's I posted were the RTL's at 4000 which I cannot get stable in gear 1 with 4 dimms so I just stuck everything back to 3600. At 3600 the board trains
65/67
25/25
67/67
25/25
 
  • Rep+
Reactions: Uncle Dubbs
5,861 - 5,880 of 7,835 Posts