Does the trend towards AIO cooling have negative effects

[TheoneandonlyMrK]

Hi all, I thought I would start a conversation about the above.

Why because I have been custom water cooling for years and initially I even watercooled my ram.

That was a real pain in the ass to be honest.

Now recently I have. Had a Ryzen rig and the move to ddr4 did away with watercooled ram, I can't be bothered.

But and here's my main point ,back in the day ram cooler's were a big thing ,why not now.

If you fit an AIO or custom the amount of air passing over your ram is way way lower than with an air cooler.

Now take the above and add to it the fact EVERYONE want's to run their Ram at the highest speed possible.

Then you load the system, the ram heats up quite slowly compared to everything else in the system and is very rarely Temperature monitored.

On my system the memory going beyond 50°c usually causes a lockup.

Are we advising right, as a community?.

 

[tabascosauz]

Yeah, that's a really good point. Usually, benchmarks won't show "hot box" instability on the RAM, as CPU benchmarks don't stress the RAM (save for P95 Large) and GPU benchmarks don't run long enough to heat up the RAM. Which is why for crashing during intensive games, I look to RAM temperatures as a major possibility; counterintuitively, CSGO is the single worst offender on my entire PC with DIMM temperatures over 50C in 30-60min of gaming.

A lot of people think that you'll only get to those borderline temperatures through 1.5V Vdram, but that's not the only way. BZ did a video a little while ago about how the GPU dumps its heat onto the DIMMs during gaming, which explains a lot of this.

It's much more an issue in small form factor applications where you don't have the luxury of even case airflow. You also have to remember that on ITX boards, you lose one whole side of each DIMM for heat dissipation because the sticks are next to each other. This is very different than most people running two sticks in A2 and B2 with a slot in between.

That said, different ICs react differently to heat. B-die seems much more prone to instability at high temperatures, while CJR and DJR seem more tolerant. Not sure about Rev.E.

When pushing very high MT/s on Intel, it's pretty smart to have a small fan laid over the RAM. In this respect, having a stronger topflow cooler like the C14S is invaluable in providing extra airflow.

 

[FreedomEclipse]

IMO. Ram coolers were never really a big thing. They were just there for cosmetics and obviously people are gonna buy into it otherwise why would RGB exist? people bought ram coolers to show off. Some had fancy LEDs, some didnt. Ram didnt really get hot enough to warrant its own dedicated cooling unless people were going after extreme ram speeds with LN2.

as for your ram locking up at certain temperatures. idk. maybe faulty ram or not enough voltage??

I think one of the biggest negative effects so far is the Asetek are choking innovation with their patents. They arent really putting in the money for R&D to build better products, they are just sitting there, selling licenses & suing companies that dont that make similar products.

 

[sneekypeet]

To me it depends on what you do to OC the RAM. In my testing (DDR4), I stick to 1.5V or less, and I see many of my kits running in the 40-45C range. On the flip side, I have never had a lockup due to heat.

I guess it comes down to what you feel is safe. For me, if you are going to surpass "safe voltages," you need to think about cooling. AIO or air cooling!

@FreedomEclipse I would say differently about DDR2, when fans were all the rage from manufacturers. We were pushing stupid amounts of volts, and coolers then would drop RAM temps to the tune of 5-7 degrees at that time. Just one mans take, of course.

 

[thesmokingman]

But and here's my main point ,back in the day ram cooler's were a big thing ,why not now.

If you fit an AIO or custom the amount of air passing over your ram is way way lower than with an air cooler.

Now take the above and add to it the fact EVERYONE want's to run their Ram at the highest speed possible.

Then you load the system, the ram heats up quite slowly compared to everything else in the system and is very rarely Temperature monitored.

On my system the memory going beyond 50°c usually causes a lockup.

Are we advising right, as a community?.

Ram speed is not that important today. And before you even watercooled your ram, no one else did it either or needed to. Ram cooling is so uber fringe, it's even more fringe than MB vrm waterblocks. And no one in their pragmatic minds would be wasting money on that today. If your ram is overheating, you're case airflow is bad.

 

[NoJuan999]

I use a Scythe Ninja 5 (air cooler) and I don't have any issues with my OC'd RAM.
But I do have 3 120mm front fans that push air across my RAM and the CPU cooler pulls it across the RAM as well.
So I have Excellent airflow across my RAM.
My current RAM kit doesn't have a temp sensor though.
I did test some Patriot 2666 RAM that I OC'd to 2933 on this system for a while and it had Samsung ICs with temp sensor and it never got above ~37c.

As far as AIOs go, I'm not really a fan, they don't cool any better than my air cooler does and they tend to die after 3-5 years.
My air cooler will last as long as I have a CPU socket that it will fit.
I still have an old i7-870 system from 2009 and the Cooler Master cooler that I bought for it still works fine.

 

[EarthDog]

But and here's my main point ,back in the day ram cooler's were a big thing ,why not now

They were a big marketing thing. That's it. 95% of people overclocking ram wouldn't benefit from them being cooler anyway. A few C isnt the difference in stability or not when overclocking.

RAM never really needed to be cooled. The gaudy heatsinks or included fans are form over function/need. Watercooling them is novel, and as you said, a PITA anyway.

So.. wasnt needed then, not needed now. RGB takes its novel place.

 

[delshay]

My last understanding of memory modules is they are tested in a chamber around 65c. This is not to say that's the maximum operating temperature, you have to go the manufacture site to find that information. If you can't find that information then you have to look up the part number data sheet & get the info from there.

If you are encountering errors before the max temperature, then you have faulty chip(s).

 

[TheoneandonlyMrK]

My last understanding of memory modules is they are tested in a chamber around 65c. This is not to say that's the maximum operating temperature, you have to go the manufacture site to find that information. If you can't find that information then you have to look up the part number data sheet & get the info from there.

If you are in counting errors before the max temperature, then you have faulty chip(s).

We are not talking about jedec speeds, memory is tested to jedec standards, beyond that isn't necessary or compulsory, jedec is.
Xmp speeds are binned sure but I doubt they get environmental testing again.

And my memory is stable, let's not get personal.

This is a general discussion not relative to my system, just my opinion.

Also it's widely accepted that with additional heat ,over clocking margins narrow, and XMP is overclocking.

 

[mtcn77]

As since big towers got updated with mid tower coolers, ram clearance stopped causing an issue. The main problem with air cooling today is the transmittance is already fast enough, they have figured out the lower portion, as for the fin array you are at inverse correlation with how much surface area you have - the bigger the tower the slower you can run its fans for similar noise output. That is the trick, having high pressure fans, yet running them at low rpm over the fins. There are not that many suitable for this purpose, there are more axial fans at +12cm frames.

 

[thesmokingman]

We are not talking about jedec speeds, memory is tested to jedec standards, beyond that isn't necessary or compulsory, jedec is.
Xmp speeds are binned sure but I doubt they get environmental testing again.

And my memory is stable, let's not get personal.

This is a general discussion not relative to my system, just my opinion.

Also it's widely accepted that with additional heat ,over clocking margins narrow, and XMP is overclocking.

You wrote that your ram locks up at 50c...

 

[silentbogo]

But and here's my main point ,back in the day ram cooler's were a big thing ,why not now.

Because of lower voltages and lower power consumption. Any type of dedicated active RAM cooling was relevant up to the end of DDR2 era, or still kinda is in enterprise sector with high RAM densities.
DDR4 in most cases won't even need a heatspreader. The thing is, even most XMP profiles won't push it past 1.35V.
1.5V is the ABSOLUTE maximum for XMP certification, but these days you'll only see these voltages on some super-expensive 4600+MHz kits, or some outdated 3600-4000MHz

models.
The earliest DDR4 had a typical power consumption of 3W per 8GB stick @2133MHz. Newer generations of DRAM are probably a lot lower(~2W or less).
With all of the above taken into account, you are looking at no more than 5W per stick, even if we are talking 16GB DDR4-4000.

Now take the above and add to it the fact EVERYONE want's to run their Ram at the highest speed possible.

I want to run my RAM as stable as possible. But, to the point, that's the reason XMP timings are usually so lax - just to avoid difference in environment temperature and motherboard quality.
If someone wants to push his memory to the absolute max(even if it's just tightening up timings), then there's no reason to put blame for instability on anyone but the user himself.

On my system the memory going beyond 50°c usually causes a lockup.

Then it's probably a good sign that you need to loosen-up those timings. There's a big difference between stock CL18 and your CL14.

 

[TheoneandonlyMrK]

You wrote that your ram locks up at 50c...

I said it has in the passed, not it does.

And the timings were not stock or XMP, soooo.

Again let's not make this personal , I have no issues now.

@silentbogo exactly, Ryzen mem calc is always pushing 2 clicks lower on every setting, I am not using those.


Just so we're clear I'm running Xmp timings on an AMD setup, set of ram sold with an AMD support price hike and sticker with no OC and no issues.

This isn't about me but the design in general.

 

[Sithaer]

Never really cared about ocing my Ram,always run them at the speed it was marketed for.

Atm thats 3200 with the base XMP profile,did not touch any other settings.

They do heat up somewhat,skin touch hot at least but that doesn't concern me as my system is stable and thats all that matters to me.

 

[hat]

IIRC, heatspreaders that started to appear on high-performance DDR sticks actually would increase the temperature of the RAM, not cool it. Something about actually blocking the airflow and reducing the ability of the chips to dissipate heat into free air, as they are stifled by the heatspreader.

Anyway, unless you have rads all over your case, I wouldn't expect the change in airflow due to adding a single rad (even if 360mm or larger) to cause problems. There's plenty of places to put fans.

 

[FordGT90Concept]

The only RAM that ever needed active cooling was:
GDDR (they have always pushed the limit for maximum bandwidth)
HBM (ridiculously high density)
FBDIMM (each stick effectively has a memory controller that gets stupid hot)

That's it. UDIMMs nor RDIMMs never needed to be actively cooled so AIO not blowing air across them doesn't really matter.

 

[EarthDog]

and XMP is overclocking.

To be clear, XMP is only overclocking the IMC of the CPU. The sticks are running at their rated speeds on the box when using XMP.

 

[Bill_Bright]

But and here's my main point ,back in the day ram cooler's were a big thing ,why not now.

I think the answer is pretty obvious - lower operating voltages - as silentbogo just beat me saying.

But I also question whether or not there really is a "trend" going on here. Yes, among those who choose alternative cooling, AiO's are gaining in popularity. But I think it is the ease in implementing alternative cooling these days that makes it more attractive to many.

But I will also point out computing electronics in general is much more efficient than in years past. The devices either demand less power, or produce more work for the same amount of power. In any case, it produces less heat per watt pulled from the wall.

Then case cooling has improved dramatically too. Cases are coming with lots of large fans and lots of additional fan options.

It is that ease to implement, improved case cooling, and more efficient electronics that drives me to question the need.

For those of us who have been around awhile, it used to be a real challenge to achieve any sort of bragging rights by maintaining a good balance between overclocking performance and proper cooling.

WAY WAY back in the day, if you wanted to overclock the CPU, to change voltages you had to cut circuit traces on the motherboard and solder in jumpers. Schematics? Ha! We don't need no stinking schematics! Motherboard makers didn't provide them anyway. You just had to figure it out yourself - typically by trial and error - lots of errors. Talk about voiding warranties!

Case cooling? Fun when 80mm fans were the biggest available and you were lucky if the case provided the option to add even one additional fan. To add cooling it was common for us to cut "blow-hole" openings in the top of the case with a hole saw.

To add water cooling, you had to visit plumbing supply or fish aquarium supply stores, and perhaps a machine shop too. You had to constantly worry about leaks and mold and fungus!

Today, what's the challenge? Board/chipset makers provide overclocking features in the BIOS Setup Menu. Or the board or CPU makers provide a simple utility to overclock. If you make a mistake, what happens? The CPU simply shuts down. You reset and start over. No big deal. When was the last time you saw a AiO spring a leak?

So yeah, it is still neat to push clocks and voltages to see how far they can go and still be stable. But to answer your question, are there negative affects to AiO cooling? Not really except to say AiO has played a HUGE role in making alterative cooling so simple AND SAFE these days, that implementers don't really have to know what they are doing (or the consequences) like in the old days. And I think that is a shame.

 

[TheoneandonlyMrK]

I think the answer is pretty obvious - lower operating voltages - as silentbogo just beat me saying.

But I also question whether or not there really is a "trend" going on here. Yes, among those who choose alternative cooling, AiO's are gaining in popularity. But I think it is the ease in implementing alternative cooling these days that makes it more attractive to many.

But I will also point out computing electronics in general is much more efficient than in years past. The devices either demand less power, or produce more work for the same amount of power. In any case, it produces less heat per watt pulled from the wall.

Then case cooling has improved dramatically too. Cases are coming with lots of large fans and lots of additional fan options.

It is that ease to implement, improved case cooling, and more efficient electronics that drives me to question the need.

For those of us who have been around awhile, it used to be a real challenge to achieve any sort of bragging rights by maintaining a good balance between overclocking performance and proper cooling.

WAY WAY back in the day, if you wanted to overclock the CPU, to change voltages you had to cut circuit traces on the motherboard and solder in jumpers. Schematics? Ha! We don't need no stinking schematics! Motherboard makers didn't provide them anyway. You just had to figure it out yourself - typically by trial and error - lots of errors. Talk about voiding warranties!

Case cooling? Fun when 80mm fans were the biggest available and you were lucky if the case provided the option to add even one additional fan. To add cooling it was common for us to cut "blow-hole" openings in the top of the case with a hole saw.

To add water cooling, you had to visit plumbing supply or fish aquarium supply stores, and perhaps a machine shop too. You had to constantly worry about leaks and mold and fungus!

Today, what's the challenge? Board/chipset makers provide overclocking features in the BIOS Setup Menu. Or the board or CPU makers provide a simple utility to overclock. If you make a mistake, what happens? The CPU simply shuts down. You reset and start over. No big deal. When was the last time you saw a AiO spring a leak?

So yeah, it is still neat to push clocks and voltages to see how far they can go and still be stable. But to answer your question, are there negative affects to AiO cooling? Not really except to say AiO has played a HUGE role in making alterative cooling so simple AND SAFE these days, that implementers don't really have to know what they are doing (or the consequences) like in the old days. And I think that is a shame.

Your right, but do consider the many cases that are sold that lack reasonable airflow to begin with.

@EarthDog Xmp is still an OC just done on mass by a company who will guarantee to some degree that function but I get your point.

 

[sneekypeet]

Something to consider too, is that each IC maker has a different threshold for thermals. I was doing a bit of digging, and Micron shows in multiple PDFs that there DDR4 is tested functional up to 85C and they even ran testing at 105C.

 

[TheoneandonlyMrK]

Something to consider too, is that each IC maker has a different threshold for thermals. I was doing a bit of digging, and Micron shows in multiple PDFs that there DDR4 is tested functional up to 85C and they even ran testing at 105C.

True but as I said before and you are probably aware they test at jedec standard speeds across various environments but to defined standards, to pass certification tests, Xmp speeds are beyond those and are effectively OEM overclocked.
Taken in isolation each point is true and correct and fine.
I would however argue that way within those same thermal parameters you can very easily setup a kit to fail way sooner by running higher speeds or tighter, they will have configured the memory so as to achieve that stability.

 

[sneekypeet]

True but as I said before and you are probably aware they test at jedec standard speeds across various environments but to defined standards, to pass certification tests, Xmp speeds are beyond those and are effectively OEM overclocked.
Taken in isolation each point is true and correct and fine.
I would however argue that way within those same thermal parameters you can very easily setup a kit to fail way sooner by running higher speeds or tighter, they will have configured the memory so as to achieve that stability.

Lets take your theory into account. Standard JDEC speeds @ 1.2V will likely never break 30C operational temperature in a ventilated environment. So if you were using Micron RAM, that is a metric shit-ton of overhead before you need to worry about anything detrimental. In the PDF, the temperature I was referring to was the point to where damages could occur. Not that all of them will, but it is the baseline result to ensure they as a maker have room to bin higher kits, and still have room for variables like low air flow, or no airflow at all. All I did was google DDR4 Thermal threshold/maximum, and PDFs popped up for me to scan through....have a look.

EDIT: maybe I'm wrong, another set of eyes on some of the stuff I saw couldn't hurt. I am writing while following, so maybe I saw it in the wrong light.

@FordGT90Concept IIRC on RDIMM way back when heat spreaders first hit the scene, I am near certain it was an interference layer to keep from issues with early cordless devices.

 

[TheoneandonlyMrK]

Lets take your theory into account. Standard JDEC speeds @ 1.2V will likely never break 30C operational temperature in a ventilated environment. So if you were using Micron RAM, that is a metric shit-ton of overhead before you need to worry about anything detrimental. In the PDF, the temperature I was referring to was the point to where damages could occur. Not that all of them will, but it is the baseline result to ensure they as a maker have room to bin higher kits, and still have room for variables like low air flow, or no airflow at all. All I did was google DDR4 Thermal threshold/maximum, and PDFs popped up for me to scan through....have a look.

EDIT: maybe I'm wrong, another set of eyes on some of the stuff I saw couldn't hurt. I am writing while following, so maybe I saw it in the wrong light.

@FordGT90Concept IIRC on RDIMM way back when heat spreaders first hit the scene, I am near certain it was an interference layer to keep from issues with early cordless devices.

No I don't doubt your right to be honest I did say as much I just wanted to state that those were controlled settings and environment tests and that use cases after that could cause stability issues within temperature ranges that are acceptable but at timings that are tighter and higher frequencies then jedec stock ,for which speeds(jedec) more than adequate testing is done.

Not at you.


This was a request for comment not me stating a factual argument against AIOs

I was interested in your opinions regarding it, still am.

 

[nguyen]

I have only tested this recently but my Samsung B-die kit does pop up some error in Memtest64 when the temp is >50C, which doesn't happen in game because all my case fans would spun up and cool the RAM to <50C. Just to be safe I relaxed the the tRFC and tREFI and Memtest64 doesn't show any error when Temp >50.

Did some googling and there is a college lecture slide about memory scheduling which incidentally mention temperature and tREFI

Basically if your memory show error with high temp, just lower the tREFI, I had 32000 before and lowering it down to 16000 eliminate any error when the memory temp is >50
Note: Increasing tREFI will improve the timespy CPU score by 300-400 points (learned this from Jaytwocents). Though with X570 board I haven't seen tREFI timing so this is Intel only or it goes by different name on AMD ?
Attached is the lecture slides.

 

[sneekypeet]

Reading some of the PDFs, I did run across explanations of Thermal Refresh, where the temperature activates a command to do so. It might be possible that something like that is what you are seeing ~ the 50C mark.