Oh I had wrote out an answer to you for this in another thread and then never posted it hehe.

I would say most likely it is temps. Around 44-45c I have seen it is much more likely to error on b die.

But it depends on many things like clocks, timings etc. I have had mem ocs get 55to 60c+ and no error because they were lower clocks and looser timings.

Watercooling dims will help but just putting a fan on them will keep them under 50c if a strong enough fan.

My setup now when testing mem and it stays around 40c @1.55v during various memtests.

 

Wow, cool...you've got a Torrent! I had a pre-order for one, but ended up going with a Lian-Li o11 Air Mini...

...and yeah, I've got a pretty powerful fan on the RAM, but it's at its lowest setting; I think I need to ramp it up a bit to see if it makes a difference to my RAM temps.

They're at about 43 to 44c during TestMem5 stability testing, so I'm probably going to pick up a watercooling kit for them. I can't help but think they'd be stable at 4266MHz if I could just keep them below 40c (I only got two errors in my last TestMem5 stability test.)...

...also, I don't want to have to reduce my clocks etc. when summer comes around.

 

Highly different results, but you're likely to corrupt the PC rather than crash if it is ONLY heat errors.
However, like a cancer that spreads, if the corruption gets to a critical state, the PC can and will crash from that alone. And it tends to happen quite quickly.

 

Thanks, that's something that worries me, because as the temperature goes up as summer approaches, it might start corrupting Windows without me knowing?

I guess it's just a case of being vigilant and running tests as the weather changes.

Currently I'm running the fan on the RAM at 100% and it's stabilised the RAM temps at 39.3/39.4 after two hours of TestMem5, so I'm just about able to run the thermal tests needed to see if I can get another 100 or so MHZ out of the RAM, if it's (more or less) in its optimal thermal range? But with the amount of voltage the sticks are drawing and the extent to which they're OC'd, I think water-cooling is going to be a must...especially as summer approaches...

...I used to roll my eyes when I saw pictures of water-cooled RAM, but here I am with an actual usage case for it now.

I was thinking I might try and augment the existing heat-spreaders with copper heatsinks, but I don't think it's going to give me the 5 to 10c reduction in temps that I'm probably going to need to maintain my RAM OC through the summer months.

While I'm typing this, it actually looks like the RAM temps are still increasing. Currently they're at 39.5/39.6 from 39.3/39.4.

I'm at about 2 hours and 15 minutes into a TestMem5 stability test, so hopefully the fan will be be able to hold the RAM temps down long enough for the P.C. to actually complete the test?

I need to increase the frequency on the RAM and test again, so I think I'll open the window wide (it's very cold here) and flush the system with air, or maybe just run OCCT instead.

 

B-Die starts getting wonky above 45-50*C depending on how tight the timings are, I haven't seen them crash, just throw random error every now and then when running a memtest. Stock/XMP should run fine at that temperature. Normally it's not that big of a deal unless you hammer the memory to find the best, absolutely stable OC or have a situation where hot air from the GPU heats up your RAM considerably.

The DDR5 Samsungs show similar behaviour albeit at a little bit higher temperature threshold.

 

I have a total of five Samsung B-die kits (32GB SR, 3866 and 4000 CL14, 15) in use in various work + play machines and found that it is best to keep RAM temps at or below 40 C when stressed. Usually, strategically-placed 120mm or even 80mm pwm fans are enough (ie. spoiler below, 4x 120mm down-draft for two systems, plus 2x 80mm from the left side).

Often one dimm - the one closest to the VRM / VRM shroud, gets the hottest and that's the one you should base your tests on, IMO.

Spoiler

 

I have a total of five Samsung B-die kits (32GB SR, 3866 and 4000 CL14) in use in various work + play machines and found that it is best to keep RAM temps at or below 40 C when stressed. Usually, strategically-placed 120mm or even 80mm pwm fans are enough (ie. spoiler below, 4x 120mm down-draft for two systems, plus 2x 80mm from the left side).

Often one dimm - the one closest to the VRM / VRM shroud, gets the hottest and that's the one you should base your tests on, IMO.

Spoiler

View attachment 2543167

Wow! That's an interesting cooling setup...you do without extractors? I guess they're not necessary when water-cooling?

 

are B die kits more or less sensitive to overheating than other kits?

 

anyone?
If I use a ram cooler I get temps down to 45 degrees but I don't like the look. Without the cooler I'm hitting 50 degrees so I'd like to get some ram that might be less sensitive to heat. Is B die more or less sensitive to heat? Or is it more about silicon lottery?

 

x2 on plopping a fat fan down on your GPU backplate for really great RAM cooling. I keep mine attached to a 3 step fan controller and can run it from quiet to full tilt. I have a 16gbh (2x8gb) DDR4 3200mhz 16.18.18.36 tRFC 500 kit of Samsung B-Die running quite happily at 4000mhz 15.15.15.34 tRFC 250 and DIMM temps on both sticks are always about 37C ... shout out to RAM sensei Ichirou for the much needed training and assistance on getting that kit dialed in ...

Ghetto but works like a magic trick for RAM cooling...

...

 

or do particular brands have better/more efficient heat spreaders??

 

...yes, sometimes even the same vendor, ie. GSkill, uses different heat spreaders for different models that have the same IC. Temp will comes down to your 'heat management' (ie. passive airflow and/or active RAM fans) and how much juice the RAM gets - more = hotter.

As already stated, I would try to keep my RAM temps at well below 45 C (40 C is my magic number, per above). I do not own anything but Samsung-B so I can't give advice how non B-die's heat profile looks like.

 

I thought it had been mentioned already but I have read in numerous places out on the intrawebz that Samsung B-Die should be kept below 40C if possible ... I've found mine to work the best at sub 40C for sure ... below 45C would most likely work too but I keep both DIMMS in the upper 30's. Till I had sufficient cooling, I was constantly dealing with corrupted windows startup files LOL ... especially when spending about a month working out my "daily driver" overclock ...

 

snip (sorry) posted by accident. (it's late, I need to sleep). 🤦‍♂️ I wish there was a delete option. 😊

 

Sleep is good !

As to your previous post, I also wish HWInfo would move the WHEA errors further up in the table, though that is probably a custom option anyway in HWInfo. When I'm taking screenies of the 5950X system, it is on a 48 inch LG/4K, but when I do OCN & visual prep, it is on a 27 inch Samsung / 1080p (for another week, then it is moving-files time ) so my other gripe with HWInfo is that it can have pages upon pages of data that are hard to display or share all at once, depending on the usual settings chosen, and important stuff on the bottom rows can be missed.

WHEA errors are important, not least as they show the whole system response, including even various non-RAM specific errors which nevertheless may have been indirectly triggered by a RAM issue. Somewhat related, very tight tFAW and also tRAS can pass various RAM-specific tests but can sometimes lead to a bit of stuttering in certain games ('low 1% and 10% frame times'), even if overall / average frame rate is great.

Your earlier system pic showed some decent cooling arrangement and DIMM temps, far better than what I've seen elsewhere. Still, why not try to see if 1.55v 'works' at the same speeds. I think you were at about 40 C DIMM temp and that's 'just slightly yellow, not deep red' but lowering RAM voltage just a bit is worth trying.

Also, I'm not sure if you have ever seen / used > this tool (there's also a whole thread on it at OCN). While it ostensibly is only for AMD Ryzen RAM tuning, it will get you very far down the road on B-die, including secondaries etc (just choose the fastest Ryzen option as a stand-in for your Intel, then select your Samsung B-die and check 'safe', 'fast' etc). I've used it on both Intel and AMD (same Samsung B-die GTZR family in use) and the author is a very well known specialist. It makes diving down the rabbit hole a lot easier.

When I build a Threadripper system about two years ago, I only had 8 sticks of GSkill GTZR on hand which were actually marketed for Intel Z270/370 etc and not even on the TR mobo's QVL list. I got it fairly tight, but then found the Ryzen Dram calculator...it helped to confirm those secondary timing parameters I had been guessing at. It may not go all the way up to the speeds you want, but just chose the highest one as a good foundation from which you can experiment further.

Finally, if you want to see how deep the rabbit hole goes, check out HWBot / Memory and benches such as > this one . DDR4 4400 CL12 or DDR5 7600 CL 30 , both on LN2 coz cool RAM is happy RAM...the DDR4 (for now) still trumps the DDR5 because of the tighter latencies, given the always present trade-off between bandwidth and latency, but soon, DDR5 will take it, IMO.

 

As a quick addendum... you may want to test your CPU's IMC with really high latencies on the RAM to see whereabouts your IMC will start to get cranky (RAM test errors), then tighten latencies after you know that point. If your CPU also has cache / ring adjustment options, test your known stable RAM speed with as high a stable cache speed as possible. Higher cache speed (and of course higher voltage) will increase temps, though, but are a big performance booster, including for games. If cache/ring isn't adjustable, disregard.

As suggested before, there's a fairly strong relationship between degradation, voltages and temps. I have degraded (only) one CPU (IMC) before via inadequate cooling for high RAM speeds, an Intel HEDT 3970X, but it was covered by extended warranty. Since then, I've become a cooling freak. Anyhow, here's a Buildzoid 'ramble' vid on the subject

Edit: And here is an interesting / fun one where Buildzoid overclock DDR4 to 4800 on Alder Lake. It's Micron (at 1.55v btw), not Samsung-B...but still fun to watch and also helpful re. various test apps to use.

 

I didn't get a chance to check out the stuff you linked to yesterday as I've been working around the clock... I basically pulled an all-nighter and I can barely concentrate tonight; I'm absolutely frazzled. ...and yes: sleep is good; I need some. 😊 I was up all night refurbishing a computer case I sold and which was collected today, so...another step closer to getting an open-loop installed. 😊

As just a quick update, I actually dropped my RAM frequency a step to 4133MHz, and was subsequently able to drop my DRAM voltage to below the XMP voltage. I found there's a 10 millivolt difference between stable and running with errors in TestMem5, and a twenty millivolt difference between stability and loosing workers in Prime95.

I've also been able to drop my IO and SA voltages considerably, to within their normal operating ranges. It's improved my temps significantly. I'm probably still going to need to water-cool my RAM to stop it crashing or erroring out, and to keep fan noise down, but it's a considerable improvement.

I paid $210 for my cheapie brand oloy 32gb b die's from newegg and have been super happy with them. They stay cool with their heat spreaders ~42'c max under load/during use while @1.5v, have RGB (if you like that, mine are disabled like all the other LED's since it lights my living room up like a Christmas tree) , and temp sensors. I can boot up to 4100mhz in cl15 but it crashes when I start things. Same thing for cl 14 4000 but only during really intense stressful uses. I saw the same results with higher timings @ 4100. I am guessing it is my 12700k's limited IMC. 3866 cl 14 seems to be the sweet spot with none of the synthetic tests having any issues as well as actual gaming.
View attachment 2543680


View attachment 2543679

That's a nice kit for that money.

I picked up 2x 8GB 3600MHz, CL14-15-15-35, 1.45V G.Skill today, for GBP171.19, that I can only imagine has to be B-die? It was the last kit left; I couldn't believe my luck...although, having said that, it was quite expensive, and I was browsing memory because one of my daily's motherboard channels is failing, and corrupted my OS, causing me to loose all of my bookmarks. ...but it led me to find that RAM kit, which, I guess is the silver lining to this particular cloud. It's a bit expensive but if I can get it properly dialled in it should hopefully be another really fast kit.

My daily's a low end Asus board with fourth gen i3 and mismatched, generic, green-PCB, DDR3, so...once I swap the new RAM, and an i3-10100 into the (edit: sleep deprived) z490 AORUS Elite I was previously gaming on, it should be quite the uplift for my daily rig. I can't wait to get off of this thing and onto something faster; especially now the memory keeps intermittently failing and stopping me from booting into Windows.

 

I paid $210 for my cheapie brand oloy 32gb b die's from newegg and have been super happy with them. They stay cool with their heat spreaders ~42'c max under load/during use while @1.5v, have RGB (if you like that, mine are disabled like all the other LED's since it lights my living room up like a Christmas tree) , and temp sensors. I can boot up to 4100mhz in cl15 but it crashes when I start things. Same thing for cl 14 4000 but only during really intense stressful uses. I saw the same results with higher timings @ 4100. I am guessing it is my 12700k's limited IMC. 3866 cl 14 seems to be the sweet spot with none of the synthetic tests having any issues as well as actual gaming.

 

so are those temp recommendations for OC'ing beyond XMP? ie do you think 50 degrees for ram at XMP settings is ok?
i purchased some 32 GB 3600 CL14 ram sticks and installed them last night and temps max out at 50 degrees during gaming but all appears stable. No WHEA errors recorded and when I run SFC /scannow not seeing any integrity violations.

 

If your sticks are running at 50 degrees C with no errors on XMP then, that temp is probably fine for your sticks.

As far as I know 40 degrees is the thermal limit for B-die, regardless of how it gets to that temperature...?...regardless of whether it's manually OC'd, XMP'd or just left stock?

The BIOS settings the XMP-profile chip on the RAM loads into BIOS, shouldn't push the RAM beyond safe limits. They're generally biased towards stability on a variety (the majority?) of motherboards with a variety of CPU configurations, which is why further tweaking is possible, i.e. because there should still be some overhead depending on your motherboard/CPU combination and the silicone lottery with regards how good the specific motherboard and CPU are at supporting a stable RAM O.C.

I'm surprised your kit runs stable at that temperature with those settings...?...if I didn't know better, I'd say it might not be B-die? But the timings definitely seem to be B-die settings? Although, I don't really have enough experience with this stuff to be able to say that for certain. But 50 degrees is about 20 degrees above B-die's upper thermal limit for stability. Maybe it's Micron Rev.B?

Also, the kit I referred to above, that I just bought, is also 3600 CL14. I fitted it yesterday, and on the same timings/settings as my 32GB kit I'm getting a 10 fps reduction in the Assassin's Creed Odyssey benchmark, so...maybe 3600 CL14 isn't necessarily a B-die bin?

I don't understand why I've got such a drop in FPS going from a 2x16 kit to a 2x8 kit, on the same timings/settings (also this new 2x8 kit won't clock any higher, without extra voltage). Maybe it's because my new kit might be single rank...?...and so I don't have dual-rank interleaving...?...but I didn't think dual-rank interleaving would make a 10FPS difference? It could be that it's just not B-die, and so it's just inherently less efficient...?

...I don't know. I'm going to download Typhoon Burner to see if it can tell me what IC's I have, and I'm going to post a new thread on the topic to see what help I can get from the forum.