Failing A PCIe 5.0 NVMe SSD In Less Than 3 Minutes Without Extra Cooling

[oiaohm]

You can actually use the drive without a heatsink as long as you have .2 LFM air moving across it. That's not a lot of air but we found that many reviewers use open table test benches without a lot of air moving over the SSD and that is creating issues in testing.

There is a bad problem here. The airflow over first M.2 slot in a open table test bench is faster than lot of closed up case configurations. Even add intake fans to most cases does not promise you airflow over the first M.2. Remember first M.2 is stuck between the GPU and the CPU heatsink.

GPU flat backplane and the CPU air coolers flat wall can be both sides of the M.2 slot so creating a low airflow tunnel. AIO can be worse because they have no fan and the wall of the AIO block can also create a no airflow tunnel.. Most cases like test bench don't have any fan lined up with the tunnel that the first M.2 is in. Yes adding intake fans does not fix problem when the air of the intake fan is either been sucked away by the GPU or sucked away by the CPU so not making it to where the M.2 is.

From what you wrote if I am reading this right we can expect the PCIe 4.0 to be throttled as well in many setups due to lack of airflow so end users not getting the performance they really paid for. Yes one of your team did note in 2022 that he thought active cooling for PCIe 5.0 would be required due to the faster ramp up. I think he is absolutely correct.

I think ChrisRamseyer with M.2 SSD we might have cross the same point GPU crossed a long time ago where they went from being passively cooled to active cooling. Of course current M.2 SSD don't have any means to directly control fan speed.

To be real in the specifications with this SSD there is no min airflow or thermal heat removal statement for best performance same problem exists with the PCIe 4.0 SSD amd before. So how was reviewers to know they should have been checking airflow.

Also have another problem if you start throttling the PCIe 5.0 ssd in the controller how much before the result is going to be no different to the user installing a cheaper PCIe 4.0 SSD in the same poor cooling location.

2LFM might seam to you as not a lot. Inside PC computer cases at the first M.2 slot location this is a massive amount due to the amount of airflow obstructions that can be in the way of airflow. Yes to the point of requiring either ducting or fan intentionally pointed in that direction to get 2LFM at the first M.2 because of the obstructions.

Tech-power up setup and the test bench setups are absolutely not the worst. Some ITX motherboards have the first M.2 mounted on the underside motherboard getting 0.2 LFM(yes 1/10) there would be doing well and depending on location in ITX case thermal pad to the motherboard tray may not help either. I think this kind of explains why some IPX builds have been PCIe 3.0 ssd.

[bug77]

Thanks for the info, you guys seem to have robust testing, however, most of us believe that your strict validation process should have caught something like this before release to consumers especially since you mention that it pertains to the E26 base firmware because corrupting a customer's data due to such a bug is a big deal.

That having been said, I own two high end Gen 4 Phison SSD's (one with E16 and another E18) and I really like both of them, this situation likely will not turn me away from Phison powered SSD's in the future, however, you guys should ensure a catastrophic bug like this does not happen again.

Remember, Phison only makes the controller. Testing the finished product was on Corsair.

[ChrisRamseyer]

Corsair does say on the product page that a heatsink is required on the MP700. I don't know what the retail packaging says, I haven't seen it first hand.

I spoke with Michael today about his test setup. Just a side note, I've known him and talked to him from time to time for 20 years dating back to when he threw LAN parties in Michigan and I did the same in Indiana. Anyhow, his test system is in a Rosewill RSV-R4000U. I have two of the same cases here. The two intake fans that shipped with the cases are pretty weak to start with. The case also ships with HDD brackets in the fans path that disrupts the stream coming out. They are located at least 10 inches away from the M.2 SSD so airflow was nearly nonexistent as far as the SSD without a heatsink is concerned.

"I think ChrisRamseyer with M.2 SSD we might have cross the same point GPU crossed a long time ago where they went from being passively cooled to active cooling. Of course current M.2 SSD don't have any means to directly control fan speed."

Yes, we are getting to that point. The Gen5 components that make up the controller use more power than Gen4. We have reached a point where SSDs will need cooling. Some of the proposed changes going on in the industry for the next generation will guarantee active cooling requirements. I suspect we will even see a new connector with PCIe Gen5 (M.2-1A, approved April 2023), but likely not till 2025.

As for M.2 and fan control, many flagship motherboards have 4-pin headers close and even right next to the Gen5 M.2 slot. It won't be long before the onboard fan control supports reading the temp of the SSD. I'm going to propose that in my next meeting with the CPU companies.

It was nice to read all of the comments and interact with everyone. I will likely be back to score some tips on making a new image for running FIO. If you ever have any questions send me a message.

[Paradigm Shifter]

Remember, Phison only makes the controller. Testing the finished product was on Corsair.

Do Corsair manufacture them? Or just slap a sticker on something made by another ODM? Seems like fails at every level to me. That said, Phison have reacted quicker than I expected.

[discordian]

Remember, Phison only makes the controller. Testing the finished product was on Corsair.

Corsair doesn't do more than Branding and marketing AFAIK. Beats me why people don't just buy from the source. If want seasonic quality, i buy it and not hope Corsair did not silently change the manufacturer fir their line.

[pong]

UX / DFx wise Just about everything about the contemporary ATX PC mechanical / electrical / system / thermal / power design is utterly horrible, then you get into laptops, smartphones which are for the most part even worse.

Every few years the individual parts' ratings get better by Moore's law but the SYSTEM / ecosystem by which they're able to be integrated gets worse and worse.

~15-20 years ago it was no uncommon thing to get a PC with a good mid or full tower case, plenty of 5.25 / 3.5 drive bays for a modest RAID 10 with a couple pairs of each or more, OK airflow / cooling / thermals, SMP dual socket motherboard if one wanted, plenty of PCI or PCI/ISA slots one could physically use with some to spare.

Now you get ONE or maybe TWO M.2 slots buried so far into the system desktop motherboard one can't so easily install them even before one puts the motherboard into the case and really insanely difficult / impossible to remove / service / upgrade them if one has installed the motherboard, CPU, GPU, heat pads, heatsink/cover plates, SSDs, etc.

For drives the size of sticks of gum one would sort of hope one could just plug 4-8 of them U.2 / SFP / USB flash drive style into exposed hot-swap slots say all installed in one front access 5.25" bay or something like that to get RAID-ability easy install, better access to just have a fan blowing over them, etc.

There are the nifty quad-M.2. mounting cards that fit in a PCIE x16 slot that's (quad) bifurcated too bad that almost no systems have
the electrical / physical capability to use them if one has a GPU etc. etc. and cooling / hot-swap would still be problematic as well as PCIE lane / slot availability.

It feels like everything that should be simple to scale / integrate reliably and easily on a desktop workstation has been "race to the bottom" iteratively worsened over time until it's basically nigh unusable / impossible.

If 2-3 of RAID-10 pairs of 3.5in. HDDs wasn't considered abnormal a while back I don't get how it's so abnormally difficult to
get 8 M.2s and a GPU or two and maybe a 10GbE NIC into a "workstation" DIY ATX build today.

Any storage drive design you can't install in a mirrored pair and you can't easily install / upgrade / replace / cool isn't really well designed at a fundamental level.

[cybernard]

If 2-3 of RAID-10 pairs of 3.5in. HDDs wasn't considered abnormal a while back I don't get how it's so abnormally difficult to
get 8 M.2s and a GPU or two and maybe a 10GbE NIC into a "workstation" DIY ATX build today.

https://global.icydock.com/product_305.html

Eight M.2 in 5 1/4 " inch form factor.

image.png

[ChrisRamseyer]

The problem is PCIe lanes. Everything runs off of them but there is such a small number of them on consumer boards. That is why I personally run workstation boards, simply to have the number of lanes I want as a user. Then again, back in the day I also ran dual socket systems. SCSI RAID with a ton of drives is what got me into storage so deep. The PC market is just like everything else. Turn it into a commodity for scale. (This is a personal comment and not one that should tie back to my work).

[oiaohm]

Corsair does say on the product page that a heatsink is required on the MP700. I don't know what the retail packaging says, I haven't seen it first hand.

https://help.corsair.com/hc/en-us/articles/14816156641293-MP700-FAQ

Yes. A dedicated cooling solution is required for optimal performance. We recommend utilizing your motherboard's NVMe slots which provide heat dissipation with their built-in heatsink. If you are using a custom watercooling solution we recommend our XM2 waterblock.

The wording Corsair does not say required. Optimal performance and recommend does not equal required. Yes built-in motherboard heatsink could be basically useless other than adding thermal mass(just delaying the crash) due to absolutely poor airflow above it.. This could be useless is that lots of cases is going to be useless. In fact the block of metal heat sink may be worse because in low airflow may get the nand chips above the critical 80C even sooner than having no heatsink.

Yes that write up the user is perfect right to think the SSD is going to behave like old SSD and just thermal throttle and not be putting self at any major risk or instantly shut down.

I have not seen the retail packaging but the Corsair website on the product does not give be confidence that what would be written on it is really suitable either. Basically the note about airflow required at this stage has not been passed on to end users and absolutely need to be.

As for M.2 and fan control, many flagship motherboards have 4-pin headers close and even right next to the Gen5 M.2 slot. It won't be long before the onboard fan control supports reading the temp of the SSD. I'm going to propose that in my next meeting with the CPU companies.

This could help with existing M.2

I still think there need to be thought to active cooling built on to the SSD itself and controlled by the SSD itself. Yes it would cost to use airjet technology but this technology is suited to non ideal air conditions and it only 3mm of thickness. Airjet technology is not like a block of metal its not going to be conduction to spread heat from controller to nand just is really effective at getting the most heat possible into air so good in low airflow areas..

There is still the question "if the controller of the PCIe 5.0 SSD is altered to throttle at what level of cooling does it performance drop under a PCIe 4.0 SSD"

Another problem there is no end user tool/instructions for PC users to gauge how much airflow they have at any location in their case at this stage. Yes a calibrated ribbons on a stick use to be used with old sun servers to see if the airflow was right. If the ribbon was either holding against gravity or was blowing around the location had enough airflow. Use to be good with the sun servers to detect if someone had not fitted the airflow guides back into the sun servers correctly. Not the worlds most complex tool required but users would need the tool(stick with calibrated ribbon) and the instructions on how to use it..

Yes this could be buy 5 dollar airflow flag/ribbon before buying a PCIe 5.0 put it on your PCIe 4.0 SSD that person is thinking of upgrading and if the flag does not show enough airflow either stick to PCIe 4.0 and under or alter case airflow until it right.

The extra heat of PCIe 5.0 ssd controllers bring new set of problems that people building and testing systems don't really have the tools to deal with and the documentation with this early product are not good enough to warn users of this problem and you can see this by the tech reviewer problems.

Do remember lot of people order their parts online as well so the website being correct is absolutely critical because they would have paid for the item and had it shipped to them before they see the retail package.

[andrei_me]

Hi ChrisRamseyer thanks for the update, that's why I love Phoronix forum, we get to know directly from involved parties 🙂