Question Ram timings help.

[bahamut204]

I got twrrd to 4 and lowered trfc a tiny bit to 480. trfc booted at 400 but only on the second training pass so I didn't test it. Think that might be it for tweaking for me. Temps in the stress tests still hitting max about 51c. For day use and gaming. If the tests pass at 51c then daily/gaming should be ok if the temps stay at that or lower. Correct? Temps are a fair bit lower in just daily use and in games they are still a few degrees cooler than the stress tests. Maybe I got the wrong games testing? Is there a program that stresses the cpu, gpu, and ram at the same time? I've tried realbench using 50% ram. Even with the cpu and gpu cooking with that. The ram still reports max 51c. hwi reports that the ram voltage drops to 1.43 while everything is stressed that way. (About 45c-47c gaming and 37c-40c regular use.) If I ever get a place that I can control ambient temps I'll be lowering those by around 6c. Am getting errors in event viewer while gaming and a few while not. But don't think it's ram stability as they seem to be the regular ones I've been getting before the new ram. Any way to lower temps outside of lowering ambient? A ram fan is not possible in this system.

Even trying what you said. Still can't go above 3600. Even 3666 doesn't boot. That's going as far as increasing the primaries to 20-20-20-40-60 and putting everything else back on auto.

 

[Darkbreeze]

Any way to lower temps outside of lowering ambient?

So, if you are getting temps above 45°C and getting errors, but the memory passes Memtest86 under normal conditions, then it is likely your errors are due to the temperature. While memory might be "safe" up to 75-80°C, that JEDEC spec does not mean that it will be "stable" at that temperatures, especially if higher frequencies and tighter timings are what's causing the temperature to climb. Honestly, it's generally not worth the tradeoff in heat and/or stability just to go beyond the XMP speed in a lot of cases.

And again, much of your problems probably come down to some combination of three things.

One, you have mixed sets in play.

Two, you have four DIMMs in play, which means there is very little room in between them and they are going to be holding in heat a lot more than with only two DIMMs installed.

Three, if you have a higher than desirable ambient, then even with good airflow you may be temperature limited from the start.

Another thing you might try if you haven't yet is the Ryzen calculator. It can help with finding favorable combinations of speed, voltage and timings on Ryzen platforms.

 

[Koekieezz]

So, if you are getting temps above 45°C and getting errors, but the memory passes Memtest86 under normal conditions, then it is likely your errors are due to the temperature. While memory might be "safe" up to 75-80°C, that JEDEC spec does not mean that it will be "stable" at that temperatures, especially if higher frequencies and tighter timings are what's causing the temperature to climb. Honestly, it's generally not worth the tradeoff in heat and/or stability just to go beyond the XMP speed in a lot of cases.

And again, much of your problems probably come down to some combination of three things.

One, you have mixed sets in play.

Two, you have four DIMMs in play, which means there is very little room in between them and they are going to be holding in heat a lot more than with only two DIMMs installed.

Three, if you have a higher than desirable ambient, then even with good airflow you may be temperature limited from the start.

Another thing you might try if you haven't yet is the Ryzen calculator. It can help with finding favorable combinations of speed, voltage and timings on Ryzen platforms.

true, up to 50c++ if JEDEC, but wont gurrantee xmp. my Hynix DJR kit (V-Color Prism Pro 4133 3600 C18) would get errors above 47c at 4133 C18 tight subtimings at 1.41v, so yeah its a matter of binning

also op did use 4x16gb which would even make the ram OC hard with tight subtimings and timings

 

[bahamut204]

As you said the 2 internal dimms are the hottest by about 4-5 degrees as all 4 are so close they are basically touching each other. With the new timings that you all helped me get to. As long as memtest86, prime95, tm5 extreme, and real bench all pass at 51c (the 2 inner most dimms) after many hours/passes. Then as long as the temps don't got above that it should be ok for daily use and gaming. Correct?

The xmp speed is 3600 and it was difficult to get that to be stable and the tighter timings. (I can't get the trfc anywhere as low as b die is supposed to be. Probably for temp reasons.) But the other timings are a lot better than what the auto training picked. With 2x16 I could push it to 3800+ easily. But I didn't want to go below 64gb.

At least I could do something with these dimms. Trying 4x32gb at 3200 before and couldn't change any of the timings from auto. (Like this is was different kits. 2 sticks were samsung ? and 2 were spektec ?. Thaiphoon burner had a ? where the die/revision was supposed to be on both of them.)

 

[zx128k]

I have a 120mm fan over my four b-die dimms and temps drop from 50c to 20-25c. I got no errors at ~50c. The main issues when tightening timings was the correct value for the VCCIO on my 10900k. I had the same issue with the 3800x. Faster RAM timings affect the cpu's overclock. I wouldguess the IF on the AMD cpu is hit harder as you tighten timings. Thus you can pass memtests but fail prime 95. This is related to samsung b-die. Other kits if the voltage is too high they become unstable.

This is why you need to put enough load on the IF as well to make sure its stable. memtest86 is not the best way to test memory. tm5 is better, each test hits a different timing. After a while if you change one timing, you can then tell its wrong by the test that fails in tm5. More often I could pass small ffts, tm5 and then fail large ffts in prime 95. This was vccio voltage for the 10900k and the IF on the 3800x. Overclocking the 3800x I had to set the IF manually to 1800 and the RAM to 3600 for the system to post.

 

[bahamut204]

I have a 120mm fan over my four b-die dimms and temps drop from 50c to 20-25c. I got no errors at ~50c. The main issues when tightening timings was the correct value for the VCCIO on my 10900k. I had the same issue with the 3800x. Faster RAM timings affect the cpu's overclock. I wouldguess the IF on the AMD cpu is hit harder as you tighten timings. Thus you can pass memtests but fail prime 95. This is related to samsung b-die. Other kits if the voltage is too high they become unstable.

This is why you need to put enough load on the IF as well to make sure its stable. memtest86 is not the best way to test memory. tm5 is better, each test hits a different timing. After a while if you change one timing, you can then tell its wrong by the test that fails in tm5. More often I could pass small ffts, tm5 and then fail large ffts in prime 95. This was vccio voltage for the 10900k and the IF on the 3800x. Overclocking the 3800x I had to set the IF manually to 1800 and the RAM to 3600 for the system to post.

How were you able to get a 120mm fan on the dimms? I don't think it's possible in my rig. I'll include a pic at the bottom to see the situation. The only thing keeping them from getting hotter than they are is the amount of raw airflow in the case. With the SoC voltage, what did you have to do vs the auto settings? With the old 128gb @ 3200 it was 1.0v. The new ram it sets to 1.087v. But fluctuates with the load on the system a bit. Like the ram voltage. It can drop by 0.08-0.16v or a bit more with the full system stressed.

I'm still working on them a bit now. One thing I still can't figure out is how to get absolute in tm5 to run without insta crashing. So I'm still using the extreme config. I thought that if it was stable that lowering the voltage on the ram a bit would help with temps. It did not change them at all and almost corrupted my OS. Even though the lower voltage passes all the various tests. memtest86, tm5, prime95, realbench without errors. Put the voltage back up now.

After about 2-3 weeks up time I did get 2 whea errors (first ever). They were both cpu bus errors on cpu14. This happened a few hours after changing my page file size and not restarting the computer. There are also about 20 or so errors in a row from the main m.2 drive a few days earlier about bad blocks. A few netBT errors listed, which is odd because I don't have blue tooth enabled. Some error like this started with the last update too. MDM Declared Configuration: Function (checkNewInstanceData) operation (Read isNewInstanceData) failed with (The parameter is incorrect.) DeviceManagement-Enterprise-Diagnostics-Provider which I don't know what it is and didn't install and the usual error that has been happening for years "Session "Microsoft.Windows.Remediation" failed to start".

I've let each of the tests run all night multiple nights with the same settings and they haven't given any errors. Is there any more ways to test the ram or check for other causes of the errors?

I think I wanted to ask something else. But I kind of forgot after all that.

Here's the pic. You can barely see part of one dimm in it:

http://imgur.com/ITw946o

View: https://imgur.com/ITw946o

 

[Darkbreeze]

Yeah, you're not getting a fan on top of them with that cooler in there. Besides which, getting the cables managed better and blowing the dust out seems to be a much bigger priority right now. The way you have the cables almost basically piled on top of the DIMMs and just the way the cables are, is not great. Need to take some time to re-arrange the cabling and get them out of the way of the air path and if you don't have front intake fans blowing directly towards the front of the CPU cooler on the case, you should get one/some installed as that would help somewhat.

I'd get the dust blown out as well. If I can see dust on things like I can, it means that likely there is far more in the fins of the CPU and graphics card coolers, and probably on or around the DIMMs as well. Likely it's accumulated a bit on everything by now.

 

[zx128k]

memtest5 is the best I found for memory testing. I use Extreme1@anta777.cfg. GSAT (Google stress app test) is the best test I have found. This is not just for testing overclocks but is an excellent way to test system stability, especially if your PC does something odd occasionally. GSAT will stress all the cpu and the ram.

 

[Darkbreeze]

You seriously look at that picture and Memtest5 is the first thing that comes to mind? LOL.

Yes, I'm just giving you hell man.

 

[zx128k]

You seriously look at that picture and Memtest5 is the first thing that comes to mind? LOL.

Yes, I'm just giving you hell man.

I have seen network racks with servers not cleaned for years. Hell my first PC was never cleaned, worked fine full of crap. Now I have water I have to clean the rad once a month. Was thinking after he got round to removing the dust. DIMMs only do 6 watts. Mine are 4xDIMMs with little space. They never overheat. With the fan they are covered in dust. I hover that once per month too. Dust wont be the first thing I think is the problem.

 

[Darkbreeze]

Dust is never on my radar in terms of "being the problem". Unless the problem is a trying to finely tune the configuration of something that depends almost entirely on managing it's thermal envelope in order to be successful. Or if the problem is a noisy bearing, which of course, by then it's probably too late anyhow.

 

[zx128k]

Dust is never on my radar in terms of "being the problem". Unless the problem is a trying to finely tune the configuration of something that depends almost entirely on managing it's thermal envelope in order to be successful. Or if the problem is a noisy bearing, which of course, by then it's probably too late anyhow.

DRAM doesn't overheat for me unless the timings are very tight. Also my PC normally has worse dust issues because of the million fans I have. If there is an overheating issue, it will be seen in temps. Also fan can die when youclean out all the dust. The dust can get inside them and kill them. Even so I have had 8xDIMM finely tuned quad channel. Never cleaned them in 7 years, same with the water rad. Stuff I just wont do today. Massive overclock for the time. 4930k over 4.5GHz DDR3 RAM 2400. Dust is not an issue unless it is one. People wreck components cleaning their PC. Mostly fans. So I only ask them to clean their case if there is a temp issue.

Everyone is ment to monitor temps as they overclock. Hell you are ment to do it every so often at stock and clean the case once per month.

Anyway my case gets that amount of dust per month because its open and it doesn't affect temps.

 

[Darkbreeze]

DRAM doesn't overheat for me unless the timings are very tight.

Seems to me this is exactly what we were talking about, just before we started talking about something else.

 

[zx128k]

Seems to me this is exactly what we were talking about, just before we started talking about something else.

I have yet to get my b-die to overheat at 1.5 volts with fully tightened timings and I mean tightened. They are 4xDIMMs with hardly a space between them. In the most stressful test they pull 6-7watts. There are kits that are fine upto 1.6 volts. There are RAM kits that overheat quickly if the voltage gets too high. The watts of the RAM is very low and it takes hours most of the time for it to reach maximum temps. You can get tons of dust on the DIMMs and not overheat. Game loads are very low. He states he passed prime 95 and tm5. He needs to troubleshoot and of course clean his case at least once per month.

Getting them to overheat, fail takes extreme timings and voltage for me. Even so I like putting a fan over the RAM sticks. They hit 50c without it, measured with a multi-meter.

There is two main timings that are affected by heat with b-die chips. So long as you don't go too high you are fine. Also most people go too high and report they are fine. This is a video of 1.61 volts into b-die, DDR4-6000.

The tRFC timings are very dependent on temperatures, as they are related to capacitor leakage, which is affected by temperature. Therefore, higher temperatures will need higher tRFC values. tRFC2 and tRFC4 are timings that activate when the operating temperature of DRAM hits 85 °C. Below these temperatures, these timings don't do anything. B-Die is temperature sensitive, and its ideal range is ~30-40 °C. However, some may be able to withstand higher temperatures, so YMMV. M8E, on the other hand, doesn't seem to be as strongly temperature sensitive, demonstrated by buildzoid.

I have tRFC at 300 for DDR4-4000, 1.5Volts and its fine over 50c. With B-die and AMD. 1.6Volts is were it gets bad. tRFC timings are important if the RAM heats up a lot. As the guide states best temps between 30-40c I use a fan but I pass tests in the 50c+ range with the same timings. Dust is only a real issue on the DIMMs if you run a Fan. Also some IC's degrade above 1.35 volts.

If you look at the information in the first post you can see his memory temps are lower than mine. Between 40-46c. His DRAM is 1.44 volts and mine is 1.45 volts stock. His tRFC is 500 which is far higher than my 300 (lower is harder on the RAM). This is the main temp affected timing.

You are right bout the dust but I am not sure its his issue at the moment.

 

[Darkbreeze]

I'm sure it's not, but the combination of poor cable management, dust and possibly (Don't really know) lack of any direct or residual airflow in the area of the DIMMs, probably IS at least contributing. No matter what, it can't hurt to correct those things.

 

[zx128k]

I'm sure it's not, but the combination of poor cable management, dust and possibly (Don't really know) lack of any direct or residual airflow in the area of the DIMMs, probably IS at least contributing. No matter what, it can't hurt to correct those things.

The ram temps are fine, they are in the 45c~. Mine are 52c with tighter timing by far. Also my DIMMs are caked in duct because of the fan. It take one pass of the hover every month to clean but they don't overheat on me. Its more vccio that gets me for stablity. B-die is ment to take upto 80c but lower only if the right timings are tighten. Like I stated tFRC is one of then. I am 300 and he is much higher at 500.

IC​	tRFC (ns)​
S8B Samsung B-Die​	120 - 180​
N8B​	150 - 170​
H8D​	240 - 260​
H8A, H8C​	260 - 280​
M8E, M16B​	280 - 310​
S8C​	300 - 340​

Next is tRFC. Default for 8 Gb ICs is 350 ns (note the units).

• Note: Tightening tRFC too much can result in system freezes/lock-ups.
• tRFC is the number of cycles for which the DRAM capacitors are "recharged" or refreshed. Because capacitor charge loss is proportional to temperature, RAM operating at higher temperatures may need substantially higher tRFC values.
• To convert to ns: 2000 * timing / ddr_freq.
  For example, tRFC 250 at DDR4-3200 is 2000 * 250 / 3200 = 156.25 ns.

As you can see a tRFC of 500 is very loose. Mine is 2000300/4000=150ns which is about the middle of the S8B values in ns. His is 2000*500/3600= 277.78ns. S8B Samsung B-Die tighten is 120 - 180ns. Also normal temps for him is below 50c load and below 40c browsing. This is not an airflow issue as far as I can tell. He has my stock vDIMM and timings. DDR4-3600 CL14-15-15-35 1.45V The RAM would have to be faulty. I have 100% the same b-die bin as him.

F4-3600C14D-32GTZN is not in his motherboards QVL but G.SKILL F4-3600C16D-16GVK 2x 8GB SS Samsung 16-16-16-36 1.35 for 1 or 2 DIMMs. My kit is basically Team TF10D48G3600HC14CBK 8GB SS Samsung 14-15-15-35 1.45 up to four DIMMs. His kit is basically Team TDPGD432G3600HC14CDC01 & TF10D432G3600HC14CDC01 2x 16GB DS Samsung 14-15-15-35 1.45 upto 2x DIMMs. His likely issue is he's 4xDIMMs or 64GB which the motherboard doesn't support. If you look at TF10D432G3600HC14CDC01 & TF10D464G3200HC14BQC01 which are 4x16GB DS Samsung 14-15-15-35 1.45 only 2xDIMMs are supported. Thats the only issue I can see.

Basically the IMC has to like this, increasing the frequency will be very hard with 4xDIMMs DS.

Configuration Max Safe Speed (Up to) OC Speed (Up to)
2 x 1 Rank 3600MHz 4400MHz
2 x 2 Rank 3600MHz 4000MHz
4 x 1 Rank 3600MHz 4000MHz
4 x 2 Rank 2933MHz 3600MHz

With 4x1 Rank I get 4000MHz and it was a real pain on a 10900k. He should have issues increasing above DDR4-3600, getting DDR4-3600 DS 4xDIMms is a good overclock in this case. There is an overclocking guide here.

I wouldn't use TM5 Absolute. GSAT+TM5 Extreme is what I used for the RAM side. Then Prime 95 small fts and large ffts.

They really push the overclocks in this thread.

It boots very consistently even without slopes and tested prime95 80k-192k and it passed no issues at 1.34v IO 1.35v SA. The problem seems to be the sticks above 37C on air it starts to be unstable on TM5. My kit seems to like CL17. CL16 requires a lot of vdimm. 4600 17-17-17-32 seems to be the sweetspot for now.

Here temps matter.

 

[bahamut204]

I the last pic given the pc had been dusted about a month prior. Unfortunately with the fans on it it's also as much a vacuum as a heater. But I probably don't have the airflow balanced correctly which might affect temps. It's just raw fan speed right now. Am working on changing the cabling. There's not much room to manage them as some of them can barely reach from the psu as it is.

Right now I have the trfc at 480, I did try 400 but the computer took a few tries to boot with that so I changed it back. Still no luck getting it above 3600. Still have roughly the same issues as before. Computer freezes for a bit when exiting a match in halo, random stuttering/freezing in most games. Wondering if it's something else that's wrong with the computer that's causing it. Like a bad cable, drive, or other hardware/software?

Does ram being unstable cause media and data integrity errors? When I had the 128gb kit with the bad dimm about 1200 of them appeared on my os drive. In the time since and since I started with this new 64gb kit about 140 more appeared on diskinfo. Also a few other weird things happening. Like my gpu is idling around 50% vram usage and task manager is claiming that firefox is using almost 13tb of vram...

Since starting the thread a new bios came out but I haven't tried it yet.

 

[zx128k]

Unstable RAM can cause data corruption.