I watercooled an NVME SSD... These results were unexpected! 

Just watched this video. I'm a former SSD FW engineer and currently do lots of storage device testing. What you're seeing with the low writes is because you filled up the SLC write cache. You should look at graphs of the bandwidth over time, you can add HWInfo and HWMonitor parameters into MSI Afterburner or pickup a testing utility. And the DRAM in an SSD is predominantly used for the FTL mapping table, there's very little space allocated for caching.

Anvil's Storage Utility has an endurance test mode that runs indefinitely. He created it back in the earlier days to see how many writes different SSDs could take before failure.

What I got from this video is that Jay should make a "water cooled everything" system since he doesn't recommend it and none of us ever will

To saturate and stress the NVME you could use a program called HDTune (The PRO version) in it, if a drive is completely wiped, (can use windows tool diskpart to do it) you can do a write test (on the benchmark tab) which is already very demanding, but if you go to file -> options -> benchmark and select full test (something along those lines I'm writing from memory) you will be doing a even more demanding test, this is the way I have found to stress test nvme drives as much as possible although when the test ends you still have to start it again but from my experience it takes a lot longer then crystaldisk, I hope this helps you guys! Edit: Don't forget to select the right drive you wanna test, on the drop down menu

The reason for the varying of write speed is the SLC cache filling out. 980 pro is a TLC drive, which means there three bits stored per memory cell. With SLC cache, data is written only 1 bit per cell until all the cells run out. During this the speed is extremely fast. Once you start writing multiple bits per cell the write speeds are slowed down significantly.

For testing try out HardDisk Sentinel PRO. You can do various surface tests ("write only" too), with user defined data, patterns, block-range (or whole disk), and times (up to 9999). Also great for Smart reading and logging.

Fun fact, if the nand gets cool enough it's speed goes down a tad, though longevity does go up. There actually is an optimal temp for speed in the Nand. The controller you want as cool as possible like most other IC's.

You also have to be careful about write cache on these drives distorting your readings. For instance, if the test is writing less than the size of the units write cache, you may get numbers that are not indicative of normal long-term operation for the drive. So if you want to test how temps affect the read / write speed of the drives, just make sure you remove all other possible variables first. Thanks for your videos! Always look forward to them.

Ive always appreciated Jay taking us along for the ride as he learns. Some of the other channels spoil the fun knowing everything about it.

I've only been using nvme 2 years. I think the highest I've seen is 70s in a Coffee Lake i9 laptop, loves to run hot. My other Samsung 970 evo, and now 980 Pro have never got beyond the 60 C range.

I would like to see the "all watercooled" build but without monoblock. Separat blocks for SSD, RAM, CPU, power supply, GPU, VRMs.....

My go to for filesystem benchmarking/stressing are IOR and MDTEST. Both are MPI based tests. IOR specifically drills on read and write performance. MDTEST is designed to hammer on the metadata performance. Usually these are done against large networked filesystems, but they can be run against a local filesystem as well. Just be careful not to fill your OS drive by accident.

OG Title, "I watercooled and NVME SSD... These results were unexpected!"

Hello Jay, You are the main culprit in building a water-cooled PC. And my best teacher regarding water-cooling until today. I am an usual old (54) Romanian guy but I still love to upgrade and build PC"s. In my water-cooled system I have a Samsung NvMe SSD, and it never worked according to specs. until I cover it with a EK heat-sink. Let me share my two cents: SSD to cool - bad, SSD to hot - bad. There is a middle spot temperature and they will work fine.

In Windows, "cipher /w:c:\" on the command prompt will overwrite/encrypt all unused disk space on the C drive. It's for rendering data on free space irrecoverable, and also causes 100% disk usage.

Jay - a couple of things, 1 about this vid & 1 about an older 1. This vid - for stressing an NVME, if HWInfo can monitor throughput/performance, & not just temps, then you could use it for that, & make some sort of batch script to repeatedly write, delete, write again to the drive as many times as needed. Older vid - so I happened to watch that vid where you were modding a plastic model of an engine into a case, & saw the way you went about bending the acrylic (or whatever type of plastic it was, I forget exactly). If you wanted to do stuff like that more often, you could get (or maybe even make your own) acrylic bending machine. If you're not familiar, they do them on Amazon.

It's good to see that even the big RU-vid channels occasionally make an editing error, not just me. 12:53, 13:00. BTW, any chance of testing some of the fan-equipped NVMe cooler, like the "Titanium Micro TMHSFM3 Nitro M.2 NVME 2280 Heatsink Dual Cooler with 30mm PWM Fan"?

The Varying Write Speed is because of 4k file writing, it does normal, 2k and 4k in sequence.

Fun fact: NAND flash has it's operating temperature little bit higher than anyone would expect. But I heard that Intel and Micron's 3DXPoint, as in older Intel's 900 and 905 SSDs has to heat up every single cell to 400°C for write. I purposefully mention cell, as during the write more than one bit is affected (between 9 and 16 if what I heard is accurate). This is why it's so power hungry.

Glad you made this video. I have been wondering my NVMe's temps and keep forgetting to look. Mine is using the stock x470 Taichi MB heatsink and it's sitting (idle) at 48-50C. It's in a 5000D Airflow, and a junky GPU (for now) so it's not getting much heat off that even when gaming.

Funnily enough, NAND prefers to be a bit warmer, especially for good data longevity. Excessive cooling does your SSD no favours.

Really interesting and useful! Clearly, a vaned heatsink is - and I think it's obvious why - superior to a flat one, as long as there's good airflow. No point (yet) in using watercooling - the benefit doesn't (yet) outweigh the hassles.

Try using an old Samsung 960 Pro (512GB). Even with a Heatspreader AND a Fan blowing on it, mine still gets over 60°C. Without Heatspreader and Fan it throttles at over 90°C. Those Temps don't come from benching it, it is that hot just by being my bootdrive... Many M.2 are also used on ITX-Boards and mounted to the Backside...those cook quite nicely as well in CPU Heatsoup^^

I've got a 970 Evo that throttled due to thermals during writes because of how the motherboard's M.2 slot was laid out. The motherboard designer, for some reason, decided the best place to put an M.2 drive, with no cover, was directly under the primary GPU's slot. Soon afterwards I changed the motherboard to one were the M.2 slot is above the GPU slot and had a heatsink cover and it dropped down to 78C full tilt.

There's a reason there's no software out there to stress test an SSD; stressing an SSD legitimately shortens its lifespan. SSD's have a finite number of reads and writes. Debaurer has great vid on SSD life expectancy relating to new cripto types. Its a real concern and shouldnt really be attempted if you actually want to use the SSD afterwards

To be honest, I loved this video😅 Investigation of such things and searching for pros and cons is one of the thing why I am watching these videos. coz I can't check it by myself. Thanks!

I don't remember having seen any SSD stress tests tbh. One thing you could do, especially with Phil, is that you can write a small shell script or a batch script that does it for you, like read a 4 GB file then copy it then delete it then do the whole thing again while measuring speed, latency, etc. I do not think it would be too complicated. Maybe even a fun video idea?

You might want to try using HdSentinel for this, It's not for running the test, but it can give you a lot of details for the hard drive or ssd, plus a graphic showing you the temperatures.

Nice to see some validation that cooling is largely unnecessary on nvme drives. Suggestion, refrigerate it and show when write speeds throttle at low temperatures.

At 7:11 Back in the day, a program called "Eraser"was used to erase hard drives. It had the ability to set a 32 loop of various data in order to ensure the data was erased. I never used that because it took hours. If you filled your hard drive with various data, (You are a RU-vidr with gigabytes of video data laying around, you can find data to just fill an SSD), then set it to its Max and wait it will fill it with all zeros all ones several times and mixed with various zeros and ones and it will take a while....no matter how fast the drive is.

Now you have to do what you said: Do a whole water-cooled system

Hey Jay, idk if you've seen, but derbaur uses it, there is a a heavy storage stress test in 3dmark, well according to derbaur it's good at stressing, so idk but it think I trust his word

1) I have always believed, that it is the controller chip you need to cool down, rather than the memory chips. 2) To test a SSD based drive, try copy/paste from a similar sized drive, using HWINFO 64 to record results. I guess you could have similar results if you did a full backup and then a restore to test.

Maybe H2testw will work. It was originally designed to test USB sticks and to identify counterfeit sticks that were labeled with higher capacity than they really are, so it tries to write to the stick/disk until it's full. It seems to work for disk drives as well, so probably also for NVMes. You just have to keep in mind that depending on the type of flash memory used, write speed might drastically reduce after a certain amount of used space.

One could write a shell script to do this in linux. It would just write over and over to the drive logging the time it took to write, time it took to read, and the temp before/after, flushing cache after every one. I have no idea if linux under windows subsystem has that kind of direct drive access though, to do the same there.

Blackmagic Disk Speed Test will do what you want. Tests up to 10TB in size and I'm fairly sure it just sits there and loops continuously until you tell it to stop. I'm too paranoid to actually let it do more than a few loops, so I've never actually tested if it will loop forever, but that's the feeling I get from the UI. The primary use is to show you which video formats your storage is capable of reading/writing. Currently tests for resolutions up to 12K/60, so you'll still have to monitor your temps with other software, but at least you'll really be putting the SSD through the wringer like you want.

with time we'll get to watercool an usb stick.

Another well made and informative video that's going to save me money in my next build. Thank you.

So the conclusion is that even without a heatsink, it's not running hot enough to actually slow down enough too make any difference in real world gaming situations. So these coolers are just a waste of money.

For longer storage tests try the Microsoft Tool DISKSPD it's for testing storage via cli on servers and if I remember right Chrystal disk mark uses it on its core. The tools allows for long tests which you usually have to do to on enterprise grade storage to fill up all the caches

I feel like I have been seeing way more failures of loops on Reddit time for another how to watercool video. Show how to mix and test the pH of liquids.

One quick production suggestion -- check out a stabilizer or gimbal for your video camera. The camera movements are sometimes slightly jerky and it can be really noticeable at times. Also monitor screen shots don't appear to be white balanced (sorry I used to do some post production work) :) I do like the background of your setup with the multi colored dim lights. That's a really nice touch. Anyway, thanks as always!

I foresee motherboard manufacturers integrating raised NVME SSD slots to accommodate cooling blocks on both the top and bottom for drives with storage chips on both sides. But as you said, it leaves more room for failure with added cooling blocks within a cooling system. Sooner or later we're gonna reach a point where we'll need even more cooling to cool the cooling system!

I got 3 of the EK heat sinks for my new build, for two reasons: 1: Not entirely confident in the two slower drives being left uncovered, since the mobo only came with one SSD plate that only fits in 2 of 3 M.2 positions. 2: I think the nickel fins fill out the board's appearance a lot better than the SSDs on their own would.

I was in a micro center discord and everybody was telling me that PCI4 isn't worth it and I was trying to tell them that makes no sense. I left that group.

No problems with ones on 4.0 compiling huge C++ files. Manufacturers test with covers in layout and design of the module. FET junctions work more on voltage (potential) transfer instead of current. Storage is a charge, so refresh uses current. Current is work (e.g. electron/hole movement) and produces heat. Operating at lower temperatures may show longer life, as storage cells are disabled over 5 - 10 years in consumer grades. Covers can provide good RF shielding in a dirty localized RF field. Now, CPU/GPUs are another matter.

I would have done the test while cloning a drive. 2 tb to 2 that is a real worst case scenario.

Hey Jay, you could also compare your results by running a performance test to the drive with Samsung Magician also.

Isn't it so, that NAND cells have a recommended operating temperature of about 40°C? Cooling it to much might shorten the liefetime of the cells. So these coolers sometimes are designed to cool not to much, so the heat from the controller heats up the NAND chips a bit.

3dmark released a storage benchmark about 6 months ago. I haven't used it and don't know how long you can run tests for.

Jay: "I change drives kind of a lot..." Yeah, you're the only one, I've had the same NVME for like 3+ years now.

I've got a Sabrent Rocket 4.0 500GB for my OS. Crystal Disk Info shows temps stay under 50C when running read/write tests via Crystal Disk Mark. She came with a pretty beefy heatsink with copper tubing.

When you are so early the title is wrong 😂

I have a Sabrent Rocket with SB-HTSK.. it was 26 bucks and if the drive lasts longer, money well spent.

The black magic hard drive test might be a good idea for this testing as it just does a constant read write hammering of the drive from when you hit test until you tell it to stop

Correct me if mistaken. Small detail but don't recall Jay using EPMD rubber tubing until this video. Its practical, versatile and flexible (maintenance-wise) compared to other alternative soft and hard tubing alternatives for water cooling in addtion to having a certain H.R Giger aesthetic appeal.

That mythbusters approach is the best, i hope you do this on future tests as well. Push these stuff to its limit!

A few years back I said I'd never build a complicated hardline custom loop again. Was building a complex custom hardline water loop with expensive distro plate and managed to destroy it when I slipped getting a tube off and snapped the fitting out of the plexi. I've a simple soft tube setup running now that was zero headaches to get running.

I love these types of video's! Makes my choice of motherboard worth it. I have a ASRock X570 Taichi Razer edition and my Gen4 SSD runs a nice 35º at desktop and 55º under load

From what I have read from some of the engineers of these NVMe devices, they say that the "cooling too much" is typically not going to be an issue because the temps where too much cooling is problematic for longevity is like 0°C which you aren't going to have in a normal system. For high temp, they said you would generally want to keep them under 70°C. So even with no heatsink the 980 is fine, especially in a case with a little airflow. But the newest 990's, SN850X, etc. do get hot. I have some passive heatsinks on mine and they stay around 50°C. That's all they really need. The motherboard covers should also be sufficient for most things. The PCI-e 5.x ones may need a heatsink vs. being able to just run out in the open.

Great video! I started looking at the temps of my Intel 670p Series 2 TB NVMe SSD only because someone referred me to an article by Patrick from STH and I was getting prepared to deploy said NVMe SSD in the MinisForum HX90. I ended up using a EK heatsink (not the waterblock), and that performed took it down from the projected 73 C max temp (as reported by STH) down to ~45 C (based on my initial CFD analyses/simulations/projections). (That was under peak load/max temps. At idle, it was a lot lower than that, somewhere around the mid 30 C range, IIRC.) Sidebar: SOME motherboards only have ONE of the included M.2 NVMe plates for ONE of the drives, even if the motherboard has three M.2 slots. (e.g. both my Asus Z690 Prime-P D4, and my Asus X570 TUF Gaming Pro WiFi motherboards are like that). And in the MinisForum HX90, if you order it barebones, it doesn't come with one. (Not sure if it would come with one even if you ordered it with a NVMe SSD drive.) re: "thermal mass" The heatsink, I would think, should have more "thermal mass" because you literally should have a bigger hunk of metal with the heatsink than you would with just that plate. But as the data shows though, it would appear that the flat plate is better for natural, convective cooling rather than forced convection. re: application of the motherboard included plate or heatsink Also from the CFD simulations that I performed, if you applied the heatsink over top of the sticker so as to not to void your warranty, the max temps would be, I think it was something like 2-3 C higher than if you took the sticker off said NVMe SSD. (In my physical tests in the HX90, I kept the sticker on, so as to not void said warranty.) So that will also be another consideration as well. Great video though! re: stress testing your NVMe SSD tool You might want to try the Flexible I/O tester (fio) tool. I don't think that it has a pretty GUI like CrystalDiskMark, but it would be a lot more potent than CrystalDiskMark for stress testing the drive for max temps. You might even have been able to get away with just using `dd` if you were just writing a gigantic, sequential file (which you can use by enabling Windows Subsystem for Linux in Windows 10) and installing pretty much ANY Linux distro just to be able to get access to `dd`. Again, doesn't have a pretty GUI like CrystalDiskMark, but it might serve this purpose, better.

The overlooked issue with these kinds of things is the pressure & flow consequences caused BOTH by adding another component to the loop and especially in the case of a tiny component like this is the cross-sectional area of the heat sink. There is always a tradeoff with components in a loop. Pressure drops will occur when you keep adding more components to the loop. If the cooler has the smallest cross section (which could very well be likely), that's going to restrict your overall flowrate.

Thanks, Jay! I was looking forward to water-cooled my NVME, but thanks to your review. It saves me money from buying one.

Thank you for covering this topic. My 1 TB (1TBx1) Patriot Viper Gaming VP4100 Series (PCIe Gen4) NVMe M.2 SSDs have a heatsink and I was curious if they would benefit from adding liquid cooling. I do not overclock, I am just an old retired guy gamer, builder, EK Products tard, enthusiast. I'm running separate custom loops on 420mm rads set up push pull. (Retired HAVCR engineer) One for the video cards and one for the CPU only. Great coverage. Thank you again. Jay-N6WIP

SSD manufacturers’ utilities (e.g. Samsung Magician, ADATA SSD Toolbox, etc.) will also give temps, often on regular HDDs too.

Have you tried the BMD disk test? It can run 5gB read/write chunks in perpetuity and stress your drive.

"Our open air test bench without a fan blowing on it is kinda the worst case scenario here and its still not overheating" thought I was going crazy 🤣 @ 12:53

I love the 'Mythbusters' vibe. Always helpful, Jay & Co.

I built a Mini ITC computer with an admittedly not small case. The motherboard had the M.2 slot on the back of the motherboard. The case had no airflow there. I cut a perfectly sized hole for a heat sink I bought for my drive. I'll have to test thermals now to see if the effort was worth it. I certainly haven't noticed any major issues so far.

My laptop probably does it the best, thermal pad for the controller, but not the storage chips, since the storage lasts longer if it operates at a higher temperature.

Jay minding the Mythbusters credo in this video, I like it. "If it's worth doing, it's worth overdoing."

If it’s above the temp ceiling for the cooler, it literally HAS to come down. Black body radiation plus a better conduction gradient means it cools (as in loses energy) faster the higher the temp is. You’ll always converge on the same temp.

To benchmarks a drive better, I can recommend: - copying a large file off it for reading, or copying a large file to it for writing (then monitor the temperature somewhere else) - on linux, gnome-disks (the standard disk manager) can do benchmarks, or alternatively use dd to read the entire contents of the drive to /dev/null (or write /dev/urandom to it) my uncooled Samsung 970 Pro in a laptop when reading starts around 3.5 GB/s but after a while it climbs up to 80°C and throttles to 1 GB/s

I found that a good way to "stresstest" an ssd is actually to run a backup on the OS. Depending how big your C drive is, it will definitely take longer than running crystaldiskmark 9 times. Im currently using macrium reflect and the ssd goes up to 100% load during backup. 60c (19c ambient) during load with EK heatsink ontop of 970 evo plus.

Every time Jay says "Nine times", I think of Ferris Bueller's Day Off and the principle telling the mom how many absences Ferris had.

I have a Adata XPG S7 1Tb nvme and I use a nmve cooler where the cooler has a 20mm fan. At full speed the fan can run at 12,000 rpm and yes you can hear it. If ran at 75% speed it runs at approx. 7500 rpm and the nvme drive stays at a constant temp around 35°c.

Interesting results !! Thanks for the awesome video Jay & company !!

Awesome video man I said awhile back on twitter I love the myth buster style video’s!!!!!love to see you do stuff to your hardware that I can’t do in fear of breaking parts and not having replacements… hope these videos do well cause I dig them!!!!!!

Thanks for the ''further away point of view'' ! I was wondering what was the brand of your test bench. Ordered one lol

10:30 actually the difference of temps (delta T) helps the cooler a lot to bring it down efficiently and quickly! I didn't understand what you meant by it is harder for cooler to bring it down! was there smt that I missed?

I don't know of any *Windows* stress tests for drives, but in Ubuntu you can use the stress command, stress --io (number of stressers) and that will run for 24 hours by default which can be canceled with ctrl + c when you are satisfied with your load

Jay you should do a build with all these crazy things you said could be water cooled. The m.2, power supply, ram, gpu, cpu etc. I would watch.

Great video, and so opportune! You must have read my mind. I have just purchased the 2TB WD SN850 for my main system after having bought the 1TB version for my (just acquired!) PS5. I was agonising over whether to buy the model with or without the heatsink for my system. I opined that the heatsink would probably be better than the provided motherboard 'plate'. How wrong I was! I may take the time to do some tests similar to yours on my own system, just for my own curiosity, but it's looking like the heatsink may be resigned to a drawer! As an aside, I have read that the NAND chips themselves perform better at 'temperature' rather than being cooled to within an inch of their lives. The controller is another matter.

One large file may in fact not produce the highest temperatures as it will not saturate the command queuing of the controller. With SATA there is only one queue, NVME it varies based on drive but can in theory support 65,535 queues of commands for read/write. The NAND flash ideally should not be cooled more than maybe a heatsink of the metal cover. Warmer means less damage on write to a cell from the higher voltages used. Cooling it will cause it to wear out quicker.

12:54 tripped me out when. Heard the same sound bit 2x in a row.

My guess is that the dips in speed were due to cache filling. It could be the pipeline cache or filling the top layers of NAND first The way around this would be to run an fstrim between each test, but that would invalidate the heat tests you're trying to run. Maybe if you ran the water block, but instead of running it the proper way, use a water heater to intentionally heat up the SSD via the coolant. You could more precisely control the temperature that way with a couple of cheap ($50) thermal electric controllers

Good to know that it aint necessary to have any more cooling than maybe a fan from the side of the chassi. Important experiment and an interesting aproach with the heatgun. Throttle comes at about 85 degrees i see

AIDA64 has a disk benchmark option under "tools > disk benchmark" which has a loop mode under "options > loop mode"

Funny coincidental moment at 18:22 when Jay blows on the drive to cool it down after having hit it with the heat gun and the RGB on the mobo *just happens* to cycle red --> orange --> yellow, so it looks like blowing on campfire ashes, which blowing on makes hotter xD

12:40 A heatsink without fan can still radiate heat away. That's what they do in space, where there is no air. But it's slow, even convection matters more.

I suspect the varying write speeds are a consequence of the test file size being some unrelated factor of the memory organisation size on the SSD, To put it in simple terms, this results in the SSD controller having a greater or lesser degree of 'overhead'; organising where the data is to be written, number of blocks of flash memory to be used, where on the chip for wear levelling etc.. This will probably impact the efficiency of any internal caching n the SSD too. There needs to be a command line version of the Crystal Disk test that invokes and runs tests with the number of runs, test files sizes and so on where those parameters are switches in the command. No need for fancy GUIs. I KNOW you'd LOVE that.

You need to do this test on a OS 2tb drive while gaming in a laptop. Mine runs up to 80-83°c for almost two years now. Has a thermal pad and a cross plate heat sink only. Never had an issue yet. Samsung Magician says it's fine currently. EVO Plus 2tb PCIe 3.0.

NAND flash doesn't really care if it gets hot, it actually performs more reliably at somewhat higher temperatures (40C-50C) and can become unreliable at lower temperatures (

Jay.. you know you need to make a completely crazy water cool build. Water cool every component. GPU, CPU, Mobo VRMs, RAM, nVME, ect. You know you have to.

Try running a recovery software analysis (deep scan). Run something like spinrite from grc labs. Recovery "tests/analysis" will allow the drive to do multiple reads multiple writes across a longer period of time.

I've seen ATTO give pretty in-depth tests for drives but obviously not still SSD-specific. Might be worth a try. I think AIDA 64 also has some disk stressing tools.

Very good by the way, always wondered. Had no issues just using thermal pads.

For a test process maybe two drives the same, place various file types and sizes on them in separate folders and then setup Synctoy to do a backup from one to the other. Enough files for the backup to last whatever amount of time would be appropriate to test, maybe 5-10 minutes? Sorry that last part is out of my comfort zone for an answer. I just know that when I run SyncToy to do some disk backups the drives run and run and run.

Got me curious. Downloaded WD dashboard to check temps on my WD Black NVME. 44 Celsius. My NVME slot is just above the first PCIE slot, under the CPU. I have a CM G100M which I bought just for looks, but it is a downdraft cooler, and my case has great airflow as well.

I use IOMeter to test throughput and it can put a serious strain. It's an older software but it does the job