DDR4: using 40x10mm fans to cool DIMM's

[magellan]

I have 2x16GiB of F4-4266C17D-32GVKB G-skill Ripjaws V installed on my daisy-chain topology Asrock z390 Taichi along with a 9700k. Getting anywhere near 4100Mhz. has been problematic so I was thinking maybe if I could provide additional cooling with say, six CUI or Delta 40x10mm fans attached to both DIMM's or 3 in-between both DIMM's (I have 4 DIMM slots, even though only 2 are being used ATM) I could add more V-DIMM and maybe get to the rated XMP speed of the RAM (4266 MT/s) or higher by adding more V-DIMM (I'm at 1.43V V-DIMM set in BIOS right now and 0.72V DRAM Vtt). If I can't get a higher operating frequency I'm hoping higher V-DIMM might enable me to tighten timings.

Right now, if I run Prime95 large FFT's with AVX the hottest DIMM can get up to 41°C.

 

[Blaylock]

Memory isn't my forte' but I'll take a crack at this.

Per the G.Skill website those sticks are rated for up to 1.50v so you should have no problems increasing the voltage to try and hit the rated 4266 speeds. There is no guarantee that you will as the IMC is dependent on the CPU and the 9700K has a base ram speed of 2666. Anything over this is overclocking. 41°C on the ram doesn't seem excessive. I'm not sure why to feel you need to add delta fans to cool them. Obviously, anytime you can add airflow I would recommend it but just not sure it's necessary. What chassis is this build in? Is there already airflow over the ram sticks? What CPU cooler are you using?

 

[EarthDog]

41C is nothing. You don't need deltas on RAM. You don't need any cooling on them, really. It's not going to help here.

But high density sticks (2x16 as opposed to 2x8) may play a role. You can add some system agent voltage, +0.2V ,and see if that helps. But if the sticks are supposed to be at 1.5V, they need to be set there in the first place...not 1.43V. Also, what slots are the sticks in? Should be 2/4 moving away from the socket. Is your bios updated to the latest version for the board?

....memory isn't my thing either. Outside of competitive benchmarking, the real world gains between a sweetspot of say 3600 versus 4266 just isn't worth the additional expense (or time trying to get there).

Edit: please update your signature so we know what hardware you're working with exactly.

 

[magellan]

Earthdog, I'm pretty sure I have to sticks in the right slots: slots 2 and 4 when counting the first slot as the one closest to the CPU socket. The board is a Asrock z390 Taichi. I have the latest version of the BIOS. The cooler is a NH-D15s. I had read that Samsung B-die can start to be unstable at as low as 42°C.

 

[EarthDog]

I don't know if that is true or not (unstable at 42C). I don't believe it though. Doesn't make much sense as it can sit at temperature at JEDEC speeds in a case that doesn't have a lot of airflow.

So again... the sticks are supposed to run at 1.5V at 4266. That is the first thing I would try. If that doesn't work, add some System Agent voltage.

Update your signature with your hardware, please.

 

[Brutal-Force]

I have these exact sticks. I can barely boot into 4000Mhz at XMP settings. I can run them at 3800 16-16-16-38-1T @1.5V stable. My DIMMS run stable at 45.6 degrees, so I don't think its your temps.

 

[Mr.Scott]

I have Z490 Taichi and I'll tip you off right now. These boards don't rev the high speed memory. 4000 is almost a pipe dream with any sort of reasonable CAS.
I have good B-die and can't run XMP at any voltage. Best I can do is in the 3800-3900 range at CAS14. Can't do 4000 or better even at CAS19.
SA voltage as high as 1.35v also.
It's totally the board.

 

[Voodoo Rufus]

DDR4 runs cooler than DDR3 and DDR2 before that. 40C is nothing. Unless you're going to pump crazy volts or clocks and deliberately roast it, don't bother with actively cooling it. The thermal sensors tell you what you need to know, but without that, if it passes the touch test (warm and not hot to the touch), just leave them be.

 

[magellan]

I have these exact sticks. I can barely boot into 4000Mhz at XMP settings. I can run them at 3800 16-16-16-38-1T @1.5V stable. My DIMMS run stable at 45.6 degrees, so I don't think its your temps.

What I had read about Samsung B-die is that 42°C is the beginning of where you can experience problems at tight timings. I'm running 15-16-16-31 280 1T and I've had it fail memtest86 at 1.5V when run for 6+ hrs. Since I've dropped the V-DIMM to 1.43V though this hasn't been an issue. But I've also heard that with greater V-DIMM you can get the timings down lower than mine even at higher frequencies. I read about one guy running 15-15-15-32 with the same G-Skill kit at 4256 on a 9900k but that required a V-DIMM of 1.6V, which he claimed was safe as long as you adequately cooled the sticks.

I think the legendary woomack once wrote this:
Just to keep accuracy up , 42c is the entry level when they start to spill out errors as Cell discharge happens faster by increased temps
Which in plain words mean, the tighter tRAS,tRP, tRFC - the higher the chance of choking under increased thermals
Higher tRP would mitigate low vDIMM, where tRP logically increases tRFC.
Longer time to (p)recharge, less affected by faster discharging as wait time is plain longer

 

[magellan]

I have Z490 Taichi and I'll tip you off right now. These boards don't rev the high speed memory. 4000 is almost a pipe dream with any sort of reasonable CAS.
I have good B-die and can't run XMP at any voltage. Best I can do is in the 3800-3900 range at CAS14. Can't do 4000 or better even at CAS19.
SA voltage as high as 1.35v also.
It's totally the board.

I once got my z390 Taichi to run stably at 4100 MT/s and 15-16-16-31 280 2t, but left the MRC Fast Boot disabled when I saved the profile and it would never boot at that speed ever again.

Could it be training at high speeds requires precisely setting the RTL's or IOL's? I know on the high end ASUS boards you can specify RTL's and IOL's not only for individual DIMM's in 4-slot z390 'boards but for individual ranks on the DIMM's.

Is it easier to run higher memory frequencies with 8GiB DIMM's over 16GiB or 32GiB DIMM's?

To get >=4000 MT/s to work with my Asrock z390 Taichi I had to specify the DRAM Init IOL to 7 -- otherwise it would never boot.

 

[Nagorak]

What I had read about Samsung B-die is that 42°C is the beginning of where you can experience problems at tight timings. I'm running 15-16-16-31 280 1T and I've had it fail memtest86 at 1.5V when run for 6+ hrs. Since I've dropped the V-DIMM to 1.43V though this hasn't been an issue. But I've also heard that with greater V-DIMM you can get the timings down lower than mine even at higher frequencies. I read about one guy running 15-15-15-32 with the same G-Skill kit at 4256 on a 9900k but that required a V-DIMM of 1.6V, which he claimed was safe as long as you adequately cooled the sticks.

I think the legendary woomack once wrote this:
Just to keep accuracy up , 42c is the entry level when they start to spill out errors as Cell discharge happens faster by increased temps
Which in plain words mean, the tighter tRAS,tRP, tRFC - the higher the chance of choking under increased thermals
Higher tRP would mitigate low vDIMM, where tRP logically increases tRFC.
Longer time to (p)recharge, less affected by faster discharging as wait time is plain longer

Samsung b-die does become unstable in the low 40s. Under memory stress testing b-die will overheat and become unstable even at lower voltage. They also can overheat when gaming and your GPU is dumping 50-60C air straight on to them.

If you want to run with the tightest timings with b-die direct cooling helps a lot. You don't need anything special though, something like this works fine: https://www.amazon.com/Corsair-Veng...s=corsair+memory+cooler&qid=1623036167&sr=8-1

With other memory ICs cooling is not as important. Also with b-die if you drop back by 1 TCL/TRP/TRC or so you can get it stable even at higher temperatures, but the memory won't be achieving its full potential.

 

[magellan]

@Nagorak
Unfortunately that corsair cooler idea won't work because I have a 120x38mm fan with 65mm of shrouds right above it feeding my noctua NH-D15s, but the 40x10mm fans attached directly to the DIMM's would.

 

[Woomack]

I had no problems with overheating of Samsung B up to ~60°C. They are overclocking better or worse and depend on the bin, IC quality, and used PCB, there are various results but up to about 55-60°C shouldn't be any issue with overclocking at CL15+. Most overclockers who mention tight timings, have CL12-12-12 or something near on their mind.
The typical temp for Samsung B in a PC case with an average airflow and 1.35-1.45V is ~45-50°C during work. The IC is designed to work up to at least 75°C but its overclocking will drop above some temp (depends on many factors).

In the case of the mentioned 2x16GB kit, it highly depends on the used motherboard and IMC. These kits are not always working well past DDR4-4000 and only on better memory controllers can make 4266+.
Btw. 4266 CL17 kits are designed to work with 1.50V. Actually, they may need 1.55V for that. It's easier to set tighter timings than lower voltage. A typical, above-average kit, will make something like 4266 16-16-16 1.55V but also 17-x-x 1.50-1.55V.

 

[Brutal-Force]

I had no problems with overheating of Samsung B up to ~60°C. They are overclocking better or worse and depend on the bin, IC quality, and used PCB, there are various results but up to about 55-60°C shouldn't be any issue with overclocking at CL15+. Most overclockers who mention tight timings, have CL12-12-12 or something near on their mind.
The typical temp for Samsung B in a PC case with an average airflow and 1.35-1.45V is ~45-50°C during work. The IC is designed to work up to at least 75°C but its overclocking will drop above some temp (depends on many factors).

In the case of the mentioned 2x16GB kit, it highly depends on the used motherboard and IMC. These kits are not always working well past DDR4-4000 and only on better memory controllers can make 4266+.
Btw. 4266 CL17 kits are designed to work with 1.50V. Actually, they may need 1.55V for that. It's easier to set tighter timings than lower voltage. A typical, above-average kit, will make something like 4266 16-16-16 1.55V but also 17-x-x 1.50-1.55V.

With a 4266 CL 17 Kit, the lowest I could get is 16-16-17-44-1T Overvolted to 1.6V or 16-15-18-38-2T at 1.5V.

I am still working on my tweaks and think I can get it a little lower, but honestly I think your numbers are right on the money. My temps under stress are ~47 degrees C.

I have not been able to get higher than 4000 Mhz, or timings lower then CL11@3200Mhz (1.8V)

No matter the speed or timings, I haven't been able to achieve better than 57ns latency without juicing above 1.65V

Disclaimer: this is really my first time digging into RAM overclocking and I think its actually more time consuming than CPU overclocking. I have been tweaking the same kit for a month and a half.

 

[magellan]

What is the maximum recommended voltage for Samsung B-die? I had thought the JEDEC maximum recommended voltage for DDR4 was 1.5V.

 

[Woomack]

Voltages are always specified as default +/- 10%. There is no clear info about how high is max safe/recommended. Every source will point to official documents so you can find info about 1.20V and +/- 10-15% (I don't remember exactly how it was for Samsung). Since Samsung B is on the market in various batches, but pretty much without significant changes for years, then users and brands like G.Skill, already confirmed that anything up to 1.6V is safe. There is no brand that would risk a high RMA rate so if they release memory kits at 1.55-1.65V then it has to be safe. Every popular brand gives a lifetime warranty on its memory products.

New 3600 CL14, 3800 CL14, 4000 CL14-15, 4266 CL16, 4600+ kits have almost only 1.50-1.60V XMP. Most DDR4-5000+ kits have 1.55-1.65V XMP.

On the other hand, Intel or AMD are not providing clear info about how high is max safe voltage for their memory controllers/chipsets. They say that the maximum specified is stock +/- 10% or ranges that always seem quite low. Motherboard manufacturers usually suggest the maximum voltages, marking high values with yellow/red or adding info next to the voltage options. You can find it mainly on higher gaming/overclocking motherboards.

Btw. maybe not Samsung but I was recently checking how high I can set 2x32GB Micron in my ITX box. I made a 45GB RAM disk, put there one game that was running 24/7, and overclocked RAM to 4400 at tighter timings and 1.6V. At about 72-75°C, the RAM was acting fine. At more than 75°C, there were stability issues. One time I got the bluescreen, one time the PC restarted, and one time the game closed.

 

[magellan]

I've been running my G-skill at 4008Mhz. When I was running at >= 1.5V (as indicated in the BIOS or HWiNFO) I would inevitably get errors in memtest86 after 6+ hrs. When I dropped the V-DIMM to 1.43V (1.448V indicated in BIOS readings and HWiNFO64) I was able to run up to 15 hrs. of memtest86 with no problems.

 

[freeagent]

I have the top fan in my Meshify C situated in such a way that it keeps the air from the fronts in, as well as blow on my ram, and half of my heatsink. It is a TY-143 with a custom curve. Running 1.5v in these summer temps I don't usually see over 35c with something like TestMem5. If I do, I have not seen 40c on them yet.. and I have been to 1.65v, but not with TM5. My sticks are usually cool to lukewarm to the touch..