SilverStone TP04 M.2 SSD Cooling

[Optimum Drives]

I have an NVME SSD in a USB enclosure and its performance is abysmal. At first it starts off pretty well, writing 500+ MB/s, but after a minute or two it dips to around 150 MB/s -- that would be acceptable for an HDD! The drive is also bloody hot.

You need to check the files load on your system, That will be one of the main issue that makes your HDD hot.

[curfew]

500MB/s for 1 minute? That's 30GBs. How often do you do that? You're hammering that SSD pretty hard

Well that was observed during a full-disk cloning process...

[Teggs]

You forgot the NB/SB chipset. Nvidia found a way to make the chipset so hot, it needed active cooling.

I did neglect to mention that. You could say that AMD also brought this joy with the X570 chipset.

"First it was CPUs...Now SSDs are starting the process..."

"...Nvidia found a way to make the chipset so hot..."


I'm almost certain that there is a mechanism at work here called "progress". (With all due respect, there is a slight problem with the logic involved in the 'Nvidia Example'; some would say that, by making a more powerful chipset, Nvidia found they had a chipset which consumed more power; which. of course, generated more heat...which needed to be dissipated somehow).

As I see it, one has only two choices:

1) accept the fact that building ever-more-powerful integrated circuits in--approximately--the same-sized packages will result in higher heat generation, with ever more sophisticated heat-dissipation and heat-mitigation techniques required; and/or

2) deny that a problem exists. This technique was tried by the firm known for their--to this day--continuing engineering wizardry, the Raspberry Pi organization. Eben Upton tried his utmost to deny that there was heat problem with the (then) new Raspberry Pi 3. This, in the face of overwhelming amounts of irrefutable published evidence consisting of hard data AND thermal images, from very many sources.
This choice didn't work for Eben Upton.
It won't work for you.

I am in favour of another option - improving the technology so that it doesn't consume so much power/generate so much heat. That is addressed somewhat by manufacturing process improvements - if AMD had manufactured the I/0 dies they used for x570 on TSMC 7nm instead of GF 12, it would not have required a fan - but mostly in this case by materials. Or maybe not mostly by materials. The CPUs of the future will not be made of silicon, but maybe radically different designs will also help them not require fans and silly amounts of metal to cool.

[RealNC]

I have an NVME SSD in a USB enclosure and its performance is abysmal. At first it starts off pretty well, writing 500+ MB/s, but after a minute or two it dips to around 150 MB/s -- that would be acceptable for an HDD! The drive is also bloody hot.

500MB/s for 1 minute? That's 30GBs. How often do you do that? You're hammering that SSD pretty hard

[flower]

I have an NVME SSD in a USB enclosure and its performance is abysmal. At first it starts off pretty well, writing 500+ MB/s, but after a minute or two it dips to around 150 MB/s -- that would be acceptable for an HDD! The drive is also bloody hot.

sounds like your ssd has an slc cache.

[F.Ultra]

"Sorry, Charlie"...

This has to be one of the most transparent pieces of fabricated personal opinion published here in a looooong time.

This doesn't even make for a great 'sound bite'. It DOES make for a great piece of total lack of understanding of basic science and engineering; in particular--semiconductor physics. I'm calling HORSE-S**T on this one, unless and until you can provide valid proof from a very valid source. Sources which promote 'channeling', ear-candling, crystal-healing, and homeopathy are not considered valid.

Where did you EVER hear, or read of, ANY semiconductor "liking heat", and performing worse when cooled (down) from a higher temperature. Do you understand the basics of semiconductor physics? Do you know that all semiconductor leakage current, of every semiconductor (germanium, silicon, gallium arsenide, silicon carbide...), INCREASES with increasing temperature?

In whose catalog can one find semiconductor heaters? On the next page right after semiconductor cooling devices? Or before? If what you say is true, someone, somewhere, absolutely must sell semiconductor heaters.

We ALL are eagerly awaiting validation and verification of this statement of yours.

You have my best wishes for providing a large dose of very good, valid, and well-documented information.


"Anyone here who believes in psychokinesis, raise my hand."--anon

Well that was lots of text where "citation needed" would suffice. Now it looks like I got the controller and the NAND reversed, it's the controller that needs cooling while the NANDS like it hotter.

https://techplanet.today/post/is-cooling-your-ssd-a-mistake

https://linustechtips.com/topic/1214043-m2-ssd-is-running-too-cold-massive-performance-drop/

[skeevy420]

Anyway, I agree with your statement: "When I see "M.2 heatsink" somewhere, my brain now equates it to "gaming chair". No, thanks."

I got a gaming chair for my birthday last year. It didn't make it to my birthday this year. About two weeks ago I was leaning back watching TV and the cheap ass particle board in the chair the stand was bolted to broke and I almost put my head through a wall.

I'm going to turn it into a flight sim cockpit whenever my rudder and stick arrive....well, that's if I can get it back from the kitty Cinder I rescued Saturday. She's 1 and the sweetest cat around. Really shy and nervous, but she's coming around.



I also rescued Poppy. She's 5 months and a terrorist. Her and Sparky are already playing together and she's having a blast.

[M@GOid]

I thought that if you cool the flash memory when writing to it, it degrades faster. That's what the experts say, flash cells want to be hot when they are written. So I didn't install my mainboard's M.2 heatsink.

I imagine that if you cut the heatsink's thermalpad to only cover the SSD controller, you can cool only it, letting the memories alone.

[RedEyed]

This goes back to 2012:

https://arstechnica.com/science/2012...-lives-longer/

Now almost 10 years later, I've heard some storage gurus (Allyn Malventano comes to mind) claim that cooling only the controller of an SSD is a good idea, rather than also cooling the NAND flash cells. Cooling those can be detrimental to the SSD's lifespan.

In this video:

https://www.youtube.com/watch?v=KzSIfxHppPY

The following was said:



However, I don't think the placement of the SSD's controller is standardized? So it seems it's not possible to manufacture M.2 heatsinks that only cool the controller. The heatsinks would need to be specific to each M.2 SSD model. I don't know if that's true though.

Anyway, it seems to me that you need to make a decision. Cool the SSD and potentially get a lower lifespan, or don't cool it and potentially get thermal throttling from the controller during long/sustained write operations. In my case, I don't do sustained I/O operations on my SSD, so it seems the wise choice is to run it without a heatsink.

Another mentioned issue I came across on the web with M.2 heatsinks is that some of them, or perhaps depending on your SSD, they cool the top side of it, but the underside gets much worse. And there's no way to really tell. You know what, overall, it just seems the whole M.2 heatsink business is a clusterfuck anyway. So I choose to not participate. When I see "M.2 heatsink" somewhere, my brain now equates it to "gaming chair". No, thanks.

Thank you for sharing information!

After reading the source article, they are operating of 250-800 °C temperature range, which has nothing in common to normal room temperature.
I tend to believe more to research papers with experiments and comparisons, instead of any kind of "guru/expert" w/o providing any valuable experiment results.

Modern memory and CPU are very complex semiconductor schemes, so I would only rely on the end user specification, If I'm not an expert in that area and there is no good research with experiments to reference on.

So, if specification says, that operating temperature range in 0 - 70 ℃ and I don't go over that range, than I don't need any extra radiators.
But, if throttling happen too often in work scenario, I would place a radiator.

Anyway, I agree with your statement: "When I see "M.2 heatsink" somewhere, my brain now equates it to "gaming chair". No, thanks."

[RealNC]

Wait, what?
Is there any research out there?

This goes back to 2012:

https://arstechnica.com/science/2012...-lives-longer/

Now almost 10 years later, I've heard some storage gurus (Allyn Malventano comes to mind) claim that cooling only the controller of an SSD is a good idea, rather than also cooling the NAND flash cells. Cooling those can be detrimental to the SSD's lifespan.

In this video:

https://www.youtube.com/watch?v=KzSIfxHppPY

The following was said:

JEDEC rates client SSDs with an operating temperature of 40C. If you force the flash down closer to room temperature (25C), then with the same amount of writing (done at that lower temperature) the end of life data retention time will be cut in half. An M.2 SSD without a heatsink will naturally rise above ambient. Same goes with SSDs with heat spreaders / heat spreading labels (they just spread the heat more evenly, which is actually better for endurance since the flash will also run slightly warmer even while idle).

Now for the heatsink / water block problem. The goal of these items is to prevent thermal throttling during heavy use but that is a controller issue, not a flash issue (flash loves to be hot while operating - specifically during writes as that is what causes the wear). Where the heatsink / block makers get this wrong is having the thermal pad contact the flash. We want it to only contact the controller. Yes, the overall temp will still run lower (less controller heat conducting to the flash while idle), but at least during heavy writes, the flash will be able to rise closer to its preferred temperature without the heatsink actively pulling it back down to ambient.

This is far less of a concern for a showpiece system that is rarely writing, but I would still recommend trimming the thermal pad so that it only contacts the controller.

However, I don't think the placement of the SSD's controller is standardized? So it seems it's not possible to manufacture M.2 heatsinks that only cool the controller. The heatsinks would need to be specific to each M.2 SSD model. I don't know if that's true though.

Anyway, it seems to me that you need to make a decision. Cool the SSD and potentially get a lower lifespan, or don't cool it and potentially get thermal throttling from the controller during long/sustained write operations. In my case, I don't do sustained I/O operations on my SSD, so it seems the wise choice is to run it without a heatsink.

Another mentioned issue I came across on the web with M.2 heatsinks is that some of them, or perhaps depending on your SSD, they cool the top side of it, but the underside gets much worse. And there's no way to really tell. You know what, overall, it just seems the whole M.2 heatsink business is a clusterfuck anyway. So I choose to not participate. When I see "M.2 heatsink" somewhere, my brain now equates it to "gaming chair". No, thanks.