SSDs With Phison E26 Controllers Shut Down at Higher Temperatures

[Chaitanya]

U.2 drives are a bit too heavy for that.

More than heavy, its the heat that undoes adhesive. I remember CFL invertor falling off due to heat. In case of U.2/U.3 drives it would be heat and weight both that will undo sticky tapes.

 

[Deleted member 185088]

Phison, quickly back to the workbench to fix your poor hot design.

This obsession with read and write spread is stupid, they should focus on random reads, area where they area till pathetic compared to optane.

 

[TumbleGeorge]

The first generation of their PCIe 4.0 controllers (E16) had the same issue so..

In both cases, are they using a old architecture intended for different technological processes that they have shrunk and crammed into an atypical node?

 

[TheLostSwede]

This obsession with read and write spread is stupid, they should focus on random reads, area where they area till pathetic compared to optane.

How can you improve something that's entirely dependent on the NAND?
Phison and the other SSD controller makers can't do anything to improve here.

In both cases, are they using a old architecture intended for different technological processes that they have shrunk and crammed into an atypical node?

Pretty much.

 

[Jism]

However, it's crucial to note that these extreme overheating problems occur only when the SSDs run without any cooling.

Really ? I mean... Oh my gawd. It's like running my Lambo without any ongoing cooling and expect it not to lock or burn up in smoke when redlining it.

 

[gurusmi]

Or just watercool them

Alphacool HDX Apex Acryl aRGB M.2 2280 SSD Kühler
Alphacool Core M.2 NVMe PCIe 5.0 liquid cooler

 

[bonehead123]

If things keep going like this, we will soon need LN2/immersion cooling on these or the next-gen of nvme drives

 

[Chris_Ramseyer]

How is this even still making new news? Corsair released a firmware update for its drives. The others will release an update that is paired with another update very soon.

Every E26 SSD was advertised to use a heatsink. The "bare drives" are sold to use with motherboard's heatsinks so you are not paying twice for a heatsink.

 

[chrcoluk]

How is this even still making new news? Corsair released a firmware update for its drives. The others will release an update that is paired with another update very soon.

Every E26 SSD was advertised to use a heatsink. The "bare drives" are sold to use with motherboard's heatsinks so you are not paying twice for a heatsink.

I think someone posted those arent good enough heatsinks but need a fan.

 

[Pyrolite]

Could be nice if E3.L 2T or E3.S 2T reached consumer desktops. They can take over for both SSD's and other PCIe cards that don't need to be really large.

 

[Chris_Ramseyer]

I think someone posted those arent good enough heatsinks but need a fan.

The drives run fine with a heatsink. Users should have some case airflow already. It doesn't take a lot of flow to cool a heatsink on a 10w SSD with an aluminum heatsink on it. I don't see anyone saying, "My 4090 throttles when I play games with a case built like an oven that doesn't have fans." E26 is a high performance product built for enthusiasts and enthusiast systems.

There is also all of this talk about U.2/U.3 being a solution for the consumer market. I wonder how many people actually own any of these drives... They use up to 25w and the cases are expensive (in PC component terms). You are not going to stick a 25w SSD in a modern PC where the drive mounts on the back of the motherboard tray (where there isn't an airflow) and expect it to not throttle. For reference, the the first U.2 Intel drive required 300LFM airflow. I only mention it because I memorized the spec back in the day.

Then there is the signal integrity for PCIe Gen5 over wires. Even on motherboards, to get the Gen5 signal to the lower PCIe slots through traces requires another Phison product that you can see on most AMD X670E motherboards (redrivers/retimers). We will talk a lot more about these at OCP Summit 2023 in October. Sending the signal over a cable uses the most expensive version of these products. As a former reviewer I know how much the PC community loves $100 cables! U.2/U.3 is not a viable solution for many reasons in the consumer market.

Could be nice if E3.L 2T or E3.S 2T reached consumer desktops. They can take over for both SSD's and other PCIe cards that don't need to be really large.

Please see what I said about moving data over cables. This also applies to the other enterprise form factors. If it doesn't plug directly into the motheboard then it has to move over wire and Gen5 over wire is expensive.

 

[Pyrolite]

I was thinking of putting them in place of some of the AIC slots, possibly with their supports attached to the motherboard. At least, if it's not part of a larger overhaul of the desktop.

With a redesigned case with fitting openings on the side, they could connect directly to the motherboard, and be arranged side by side to be closer to the CPU/chipset. If used with existing cases there might be connectors perpendicular to the motherboard so that they face backwards as normal, but that seems more awkward and might require a few more centimeters of traces.

 

[Chris_Ramseyer]

I was thinking of putting them in place of some of the AIC slots, possibly with their supports attached to the motherboard. At least, if it's not part of a larger overhaul of the desktop.

With a redesigned case with fitting openings on the side, they could connect directly to the motherboard, and be arranged side by side to be closer to the CPU/chipset. If used with existing cases there might be connectors perpendicular to the motherboard so that they face backwards as normal, but that seems more awkward and might require a few more centimeters of traces.

I hate to be that guy, but Gen5 AIC adapters are rare and also expensive. The Gen3/4 adapters even have issues with some Gen4 SSDs. It will be worse with Gen5.

PCIe Gen5 x4 to U.2 Vertical Adapter - Serial Cables

PCIe Gen 4 U.3 vertical adapter %

www.serialcables.com

 

[mechtech]

MOAR COWBELL!!!!!!!!! errr I mean MORE HEATSINK!!!

 

[Chris_Ramseyer]

Hah, I've been telling that from the start. Even provided data that they do not obey slot power limits. We had a Phison REP here.

What we see here? A botched product. At least they could market it properly for special use cases not the epeen Netburst sequel popcorn numbers.

It does not go over the PIC-SIG m.2 power limits. The limit is 11.55w with a bit more for bursts.

 

[Deleted member 185088]

How can you improve something that's entirely dependent on the NAND?
Phison and the other SSD controller makers can't do anything to improve here.

I was referring to the stagnation in the SSD industry as a whole.

 

[TumbleGeorge]

Well, that's an attempt to shift the blame. Why didn't they choose the limit to be 6 watts, but almost 12?

 

[Ferrum Master]

It does not go over the PIC-SIG m.2 power limits. The limit is 11.55w with a bit more for bursts.

I will remind you again. You are wrong using those numbers. You can only operate with mechanical specs given by the connector itself. The new ECN is only in the pipes.

Current Rating: *0.5A/Power contact (continuous), * The temperature
rise above ambient shall not exceed 30˚C, * The ambient condition is
still air at 25˚C, * EIA-364-70 Method 2

As the card is scorching hot it is violating the spec.

Bloody nobody would bat an eye if this controller would have existed only for AIC cards, but you had to go pop. Now look at the backlash.

 

[Pyrolite]

Well, that's an attempt to shift the blame. Why didn't they choose the limit to be 6 watts, but almost 12?

Although it seems like we're asking a lot of this form factor(kinda like with the bigger GPU's), there are proper and improper ways to use this allowance and all devices have their requirements.

I hate to be that guy, but Gen5 AIC adapters are rare and also expensive. The Gen3/4 adapters even have issues with some Gen4 SSDs. It will be worse with Gen5.

PCIe Gen5 x4 to U.2 Vertical Adapter - Serial Cables

PCIe Gen 4 U.3 vertical adapter %

www.serialcables.com

Hmm. Best have the connectors integrated right on the motherboard, then. I've heard Gen6 managed to not increase signalling difficulties over 5, at least...

 

[Woomack]

I have a Crucial T700 and an ADATA Legend 970. There were no problems with shutting down on both of them when I was running extended tests. Crucial was throttling from time to time with the stock heatsink, and ADATA was not even throttling, but the fan is noisy (can't adjust speed, it runs at 10k+ RPM all the time, temps around 67°C in tests).
If you run these SSDs in SFF or next to a hot graphics card, then it may throttle all the time or maybe shut down, but I haven't seen my SSDs shutting down, even though sensors on Crucial were showing up to ~85°C spikes. Most SSDs will shut down when they reach the critical temperature ( usually 90°C+). If they shut down, then throttling doesn't work, or it triggers too late ... or simply ambient temp is way too high, no airflow, etc.

 

[chrcoluk]

The drives run fine with a heatsink. Users should have some case airflow already. It doesn't take a lot of flow to cool a heatsink on a 10w SSD with an aluminum heatsink on it. I don't see anyone saying, "My 4090 throttles when I play games with a case built like an oven that doesn't have fans." E26 is a high performance product built for enthusiasts and enthusiast systems.

There is also all of this talk about U.2/U.3 being a solution for the consumer market. I wonder how many people actually own any of these drives... They use up to 25w and the cases are expensive (in PC component terms). You are not going to stick a 25w SSD in a modern PC where the drive mounts on the back of the motherboard tray (where there isn't an airflow) and expect it to not throttle. For reference, the the first U.2 Intel drive required 300LFM airflow. I only mention it because I memorized the spec back in the day.

Then there is the signal integrity for PCIe Gen5 over wires. Even on motherboards, to get the Gen5 signal to the lower PCIe slots through traces requires another Phison product that you can see on most AMD X670E motherboards (redrivers/retimers). We will talk a lot more about these at OCP Summit 2023 in October. Sending the signal over a cable uses the most expensive version of these products. As a former reviewer I know how much the PC community loves $100 cables! U.2/U.3 is not a viable solution for many reasons in the consumer market.


Please see what I said about moving data over cables. This also applies to the other enterprise form factors. If it doesn't plug directly into the motheboard then it has to move over wire and Gen5 over wire is expensive.

Enterprise drives have always used more power, not aware of any consumer drives using U.2..

In terms of the rest of what you said yes cases would need to be more like older style cases where drive bays was common, they changed before they can change again. The stick on the back of the board idea was poor to begin with, and perhaps you can release data of these drives running full load, no throttle, on a low/medium end board thin slab heatsink sandwiched between a hot cpu/gpu with no more than two fans in the case, then I will take your word for it.

The 4090 comment seems a bit desperate, not a good idea to go there, those GPUs come with huge beefy cooling. They dont tell end users to provide their own cooling.

Appreciate the comments though from the industry wanting to defend what they have done here, and maybe in the future the costs for the gen 5 cables can be brought down. I remain of an opinion that gen 5 drives just arent worth it at all in the consumer market, so my final part of my reply to you, if you concerned about the consumer market paying extra for things it doesnt need then why have you even released this product?

 

[robert3892]

I would like to thank all of you who working in the PCIe gen5 SSD world wide beta test.
So far I see no point to invest into gen5, so the next one will still be "only" gen4 with the low low speed of 7400 MB/s


That tiny fan cannot be efficient and silent
Here is my solution

3 slot is overkill and you know it, even 4 SSDs will not produce more heat than 40~45W of heat.
Here a valid solution
View attachment 313469 View attachment 313470
1kg pure copper and a 5cm double ball bearing fan is a great solution to keep cool your m.2 drives
I know it's "only" gen4, but there is a gen5 version:
View attachment 313474

That tiny fan is efficient and quiet. I purchased several other brands of similar cooling heatsinks with a built-in fan and they were too loud. So I have physically tested quite a few of these in order to find the best solution. Right now the ambient temperature in my room is about 21C. My MP700 temp at idle is about 49C. Under full load conditions, the temps don't exceed 62C.

I also experimented with using just a heatsink (with thermal tape) without a fan. The temps shot up close to 90C. I tried several different heatsinks. I absolutely cannot recommend a stand-alone heatsink without a fan for the Corsair MP700.

 

[Assimilator]

Yes it is hot, but Phison should have mandated a good cooling solution. I also tried a standard M.2 SSD heatsink without a fan which didn't work well. You really need that M.2 SSD cooling fan.

Thermal throttling has been a staple of SSDs for years now. The fact that Phison's controller gets so hot so fast, that the initial firmware was unable to throttle in time and the hardware had to shut down to prevent itself from catching fire, points to a distinct lack of basic testing. Like I said - rushed product.

And Phison doesn't mandate anything; they just sell the controller and a reference SSD design. They likely strongly recommend the use of a dedicated cooling solution, but ultimately it's up to the company selling the physical drives to decide how to market their product, and most simply aren't going to want to force their users to pay more for a bulky and noisy cooling solution. Phison should have anticipated this and designed their controller and firmware to not require a HSF, but they were so preoccupied with being "first to market" that they ended up delivering an entirely substandard product - one that can't be fixed with any number of firmware revisions, but will require an actual hardware redesign. Likely by the time that redesign is done, other competitors will have delivered PCIe 5.0 controllers that aren't prototypes being sold as production-ready.