960-EVO 250GB NVMe Gone South: Must troubleshoot . . .

[BonzaiDuck]

Well, it's been a rocky road with this Sammy NVMe.

This was an experiment to cache SATA devices to both NVMe and RAM. I was using PrimoCache, which wasn't really a source of any problems I had. If any, it was all because of a mistake I made in deleting a drive volume, and that was fixed. I replaced the NVMe drive in the system and it reinstalled, since the Sammy NVMe driver was already there. Recreated a partition; added a volume half the size of the available.

Was playing a game, when my system locked up the first time in a couple months. [I took pains to overclock the CPU and my GTX 1070. They were both rock-solid OC's. I can count the BSODs made in the process -- exclusively in OC'ing the CPU -- on one hand, and I'd put it through the ringer with LinX, IBT, Prime95 etc.

there was a brief flash on the screen of a message box, with something about a "reset" -- and I can't guarantee but thought I saw "NVMe."

I can't find anything that isn't working in the Device Manager. Reconfiguring my caches, everything seems fine for playing my games, accessing the disks that are still there. But the Sammy controller in "Storage controllers" seems to have disappeared and doesn't come back after two successive reboots.

I will obviously soon remove the Lycom PCI-E NVMe-to-PCIE-x4 card with the 960 EVO. I'd like to be sure that no damage occurred to anything else in the system. I have a spare ASUS PCIE-x4 card to test whether the drive is alive or dead.

There are two PCI-E x1 devices in the system which still show in Device Manager and still apparently work: an x1 SATA Marvell storage controller, and a Hauppauge 2250 PVR tuner card, also x1.

Does anyone have any thoughts, similar experiences, advice or suggestions about this? I believe the $130 Sammy may have to go to Samsung RMA. But I want to be sure that the problem wasn't something else. I guess I can start by trying the other PCIE x4 adapter card . . . . Those things are so simple-looking, I'm skeptical of chances that it's the card and not the drive, but we'll see . . .

There's been no change in graphics performance, and the 1070 is of course occupying the x16-1 slot.

 

[VirtualLarry]

Sorry to hear that, Bonzai, I know that you've been having fun experimenting with that new NVMe drive.

Could it have been the card? Is it some cheap Chinese generic? Some of them, don't have decent / sufficient power-delivery. Then again, if it's just a physical PCI-E slot to card adaptation, I'm not really sure if there would even be need for power delivery / filtration, if it's just a wiring adapter.

 

[XavierMace]

What kind of heatsink did you have on the NVMe drive? How hot (if at all) was the add-in card getting? Just to be clear, the NVMe drive doesn't show up at all anymore, correct?

 

[BonzaiDuck]

Sorry to hear that, Bonzai, I know that you've been having fun experimenting with that new NVMe drive.

Could it have been the card? Is it some cheap Chinese generic? Some of them, don't have decent / sufficient power-delivery. Then again, if it's just a physical PCI-E slot to card adaptation, I'm not really sure if there would even be need for power delivery / filtration, if it's just a wiring adapter.

There are components soldered to the board.

I yanked it, picked up the spare card I bought -- an ASUS Hyper M.2 x4 Mini Card.

Put the Sammy 960 EVO in the ASUS card. Put it in the system. Booted to Windows. Samsung NVMe Controller now shows again in the Device Manager, working properly. Deleted the old volume, created a new one, with TBW now up to 0.900 TBW.

I have to keep an eye on this. The problem erupted when playing a game, which froze. Not sure exactly what caused it, but the game has a record on the web of momentary freezes and recoveries. I could further downclock the card, but I don't think that's the problem. It's more likely the game video settings.

But when your PCIE NVME M.2 drops out, well . . . . sonofabitch!

Another thing I discovered, per the 2700K: Never, Never move a drive from another Media-Center-capable machine contain the "recorded TV" fold and its MC buffer and protected files from the first machine. I went through hell saving movies and captures, testing drives and other possible factors. The best thing to do with such a drive is reformat it to delete those folders. I could hold forth at length about the things that happened, but anyway -- it's fixed . . . . too . .

I keep thinking the problem with the NVMe could have been the Lycom card, but look at the circumstances . . . The only thing to do is try again and keep an eye on it.

At least, the PCIE slot is working fine . . .

Now looking at XavierMace's response. There's no heatsink on the NVMe, and I know that they're a good investment for the larger sizes of M.2. there's a product called CryoM.2 with a heatsink version. I was thinking to get one if I get a larger M.2.

But it was never getting very hot. I monitor these temperatures as much as I can. The way it's situated at the bottom of the motherboard, there is a steady blast from a 140mm fan blowing right on it and getting sucked through the motherboard duct and out of the case.

The last time this had happened, I'd explained that I inadvertently deleted a caching volume that needed to be there. Removing the M.2 and then restoring it to the slot seemed to resolve any further problems until this recent event.

Truth be told, if there's any more trouble with it and I can't sort it out, I'll save my money for a larger card and use only an SATA SSD to cache hard disks exclusively.

In process of investigating the problem with the media drive on the other machine, I thought I'd look at TBW statistics on the 60GB Mushkin Chronos. Amazing! The cache had filled up, but for two years use, I think there was a total of 5 or 6 TBW.

All of this, for me, is about integrating one or two large-sized spinners into an otherwise simple storage strategy and making them work faster. But I don't cache anything of media drives, such as the one I mentioned. That seemed suspiciously like another roadblock with MC managing DRM, and moving a drive from the original computer.

And I'm coming around to the view that more memory might be a better investment. I might not shed a tear if I lose a tier.

 

[FFFF]

Maybe some compatibility problem between the the Lycom card and your motherboard? Good thing you have a spare from a different brand.

 

[BonzaiDuck]

Maybe some compatibility problem between the the Lycom card and your motherboard? Good thing you have a spare from a different brand.

I looked closely at both of them, naïve about how to identify electronic components on them. There is very little in the way of anything other than the circuit-traces on Lycom card. The ASUS card was offered as a multiple solution that included an ASUS "Hyper-Kit," which I need to investigate further. The other option is a jumper setting which just makes it an M.2 NVMe adapter. But it has bigger or many more components on the PCB than the Lycom. And I'd already suspected that ASUS is very disciplined in the way the control voltage and current, even evident in their BIOS "Over-voltage protection" option.

Chump-change items like this, when I'm uncertain about the outcome, I may buy in quantity greater than 1. In this case, the outfit shipping the Lycom had a delayed ship-date, and I was in a hurry, so I ordered the ASUS Hyper M.2 x4 from someone else. Turned out, they both arrived on the same day. But I think I should've ordered the ASUS alone. And I don't think I want to bother testing the Lycom in another system. It just goes to the parts locker for "that day." I wouldn't bother to RMA-refund for the price I paid. After all -- it's an "experiment."

Just dropping in another observation about M.2 NVMe caching. If you want to cache an SATA SSD, you don't need a lot of this high-power NVMe for cache -- I'm using about 40GB out of the 250GB split between Win 7 and 10 boot systems (100GB each). The remainder of 60GB is going for the HDDs, and that doesn't prove so much until I begin adding more programs stored on that disk.

I thought about buying a 2x16 RAM kit to replace the 2x8 that I have. More would be even better, but I don't need more.

I think I've found a "sweet spot", even for required use of a lifetime-licensed $30 software program. More experiments will prove out one way or the other.

 

[VirtualLarry]

More experiments will prove out one way or the other.

As always, we look forward to your exploits and experiments, BonzaiDuck.

 

[BonzaiDuck]

As always, we look forward to your exploits and experiments, BonzaiDuck.

There's this guy at the Romex forum. He's got three disks of large size, connected via USB3, with what he says are "data-sets" of 500GB on each one. He's got 128GB of RAM. He's apparently attempted to use an entire 1TB 960 Pro NVMe to cache these drives, but Primo doesn't work that way. It has a stealth-caching feature, and eventually fills up -- but it's never going to fill up with either 500GB of files, or a single 500GB file.

If people want to use this for a user-workstation -- we know what it does for servers and databases -- the best thing to do with it is to cache OS and programs based on user usage profiles. After that, what it does immediately is cache to RAM.

If you're not too likely to load 4GB files frequently (other than playing movies which don't need any (ANY!) caching but for rendering tasks, a 2 to 4GB RAM cache for one or more disks (SSD and HDD) is probably sufficient for most things so that you don't run into a situation demonstrated by benchmark tests of 2GB on a cache size of 1GB.

I'm caching to two 2.5 GB RAM caches for two devices: SATA SSD and SATA HDD. the NVMe is split 40-60 between those two devices, and is very slowly filling up. If I started those caches the day I initiated this thread, the 40 GB cache now only has about 30+ GB of free space. The 60GB portion is almost wasted on the 1TB volume of the 2TB HDD, but it's accumulated about 6 or 7 GB of material in the cache so far. If I add more programs to that 1TB volume, and if I use them enough, all of that will eventually build up the 60GB portion of the cache.

I"ve already ordered a KryoM.2 heatsink-version of an NVMe PCIE x4 expansion card. I'm eventually going to put the OS partitions and volumes on either a Pro or EVO NVMe. There will still be another volume on it for caching my Seagate BArracuda. It can't be worth anything to allocate more than 100GB for it, and it would probably be better set at 50GB.

But I can tell you that the current configuration has profound impact response and performance. I can run my stock-car racing game, and it scares the hell out of me -- an adrenaline rush -- I hadn't experienced before I set this up with the 250GB NVMe drive.

Everything else -- it's just "right there" -- right there -- any program on the ADATA SATA SSD or any feature of the OS -- it's just "blink" and it's on the desktop.