Trashbytes wrote on 2025-08-14, 10:01:

Ive had SSDs on non Trim capable systems die but NEVER from a lack of Trim.

Who said anything about dying? Lots of other problems arise.

PC Hoarder Patrol wrote on 2025-08-14, 10:53:

tony359 wrote on 2025-08-14, 10:40:

Thinking about this, the box has software quotas for storage - which is set to 25GB at the factory, this gives you an idea of th […]
Show full quote

Thinking about this, the box has software quotas for storage - which is set to 25GB at the factory, this gives you an idea of the amount of data expected! 😀
I will leave that quota to something like 100GB and the drive will never be full to the brink.

What I am not sure I understand is what the role of TRIM is. I understand that without TRIM the drive doesn't know what is deleted, is that correct? So how does the firmware garbage collection work if the drive doesn't know what needs to be deleted?

Back to the box, this morning I plugged the original PATA drive, 160GB, and observed the behaviour at boot. There is NO long "solid" activity after "Ready" appears. There is some activity but less than a second of "blink" so something is odd when a large drive is used.

I'm curious to see whether a 500GB SATA HDD shows the same behaviour. The SSD seems to be working fine for now. I see the light at the end of the tunnel - this project has taken all my energies for months now!

trim v garbage collection - https://www.kingston.com/unitedkingdom/en/blo … -trim-explained

Yes. When people, like in this thread, and drive literature say stuff like "the drive will take care of it's own maintenance" they mean garbage collection, not TRIM.

tony359 wrote on 2025-08-14, 17:42:

So the trick is the "overprovisioning", the extra space that is "always empty". In fact, when I tested the Samsung SSD with Samsung Magician (what a horrible piece of SW!) it had an over provisioning option. I wonder whether it would be wise to reserve some more.

On topic of "overprovisioning" - if done right it is useful as it increases amount of empty space no matter what. On the other hand - this way you are unable to use all the space you paid for, so have to be reasonable...

If done wrong - it makes matters worse. Thing is - you have to be absolutely certain the space you are leaving unallocated is actually empty. Because if it is not you are making matters worse - it'll never be erased or used. Simply deleting or shrinking partition may not be enough, TRIM has to be sent and if OS does not do it when you delete or shrink partition you are screwed.

It is best to do secure erase and then create partition of desired size right away to avoid issues. There is also blkdiscard in modern linux which can be used, but you'll need to be careful and specify correct ranges to discard.

tony359 wrote on 2025-08-14, 17:42:

So now I have two options! 😀

500GB HDDs or 250GB SSDs? Costs are similar. The HDD still feels more "original", that is, won't add unknown variables to the equation. Which might not be true as the 1TB HDD was also supposed to do that.
The SSD though seems to work perfectly fine and on a side effect the machine is snappier 😁

My opinion? SSD is the better option. It'll be faster and likely last a lot longer. Also less noise, power, etc.

Archer57 wrote on 2025-08-15, 05:48:

If done wrong - it makes matters worse. Thing is - you have to be absolutely certain the space you are leaving unallocated is actually empty. Because if it is not you are making matters worse - it'll never be erased or used. Simply deleting or shrinking partition may not be enough, TRIM has to be sent and if OS does not do it when you delete or shrink partition you are screwed.

That's why I was mentioning Samsung Magician, it has an option to set overprovisioning, I'd imagine it would be a firmware feature instructing the firmware to use that amount of space. I believe the drive would appear smaller to the OS after doing that.

My opinion? SSD is the better option. It'll be faster and likely last a lot longer. Also less noise, power, etc.

SSD it is. so far it's performed fine. I am still VERY curious of why larger drives are causing havoc! 😁

tony359 wrote on 2025-08-15, 09:13:

That's why I was mentioning Samsung Magician, it has an option to set overprovisioning, I'd imagine it would be a firmware feature instructing the firmware to use that amount of space. I believe the drive would appear smaller to the OS after doing that.

May be i am wrong and they changed something, but from when i tried, probably a decade ago, it simply made an unallocated area like you can do yourself. Though i'd hope it at least takes care of actually clearing it...

The DTS players worked fine with the small SSD... until yesterday. I turned one one and went into that "I am going to crash and reinstall the software because the MD5 integrity check failed" at boot. I now wonder if it's always been this unit having this issue or both - I was working on two.

I really don't think there is much I can do though, whatever this is, it likely requires a software update. And who knows, this might be expected, after all there is a nice and polished routine to fix that (the player says "Running software maintenance" as if it's a good thing!).

But to be back OT, someone suggested using Compact Flash cards directly on the motherboard with an adaptor. After all CF are IDE devices. That would remove the adaptor from the equation and also provide small drives.

However, you might remember my "Retro PC upgrade disaster" video where I struggled for weeks trying to install Windows 98/XP on a Socket 7 machine and then it turned out it was the Compact Flash (all CF I tested) and everything worked with a spinning HDD.
So I am not sure the CF option comes with no drawbacks.

Any ideas? And I'm not convinced that this "corruption issue" is due to the storage. Just for conversation's sake 🙂

tony359 wrote on 2025-07-03, 13:15:

It's a custom box, a cinema sound player for 35 and 70mm film. It has custom cards in it, I'd imagine some custom drivers too. […]
Show full quote

Nehalem501 wrote on 2025-07-03, 12:48:

is there a reason why you cannot upgrade to a slightly newer kernel ?
A newer kernel with an older distro shouldn’t be a problem.

It's a custom box, a cinema sound player for 35 and 70mm film. It has custom cards in it, I'd imagine some custom drivers too.

I'm already concerned in using a SATA drive instead of a PATA - but only mildly. Anything else could potentially causing software issues and if the software glitches, the sound glitches leaving many people upset 🙂

It's such a small niche of products that software was tightly tailored around the HW, not designed to work with anything but what they were installing at the factory. I wouldn't be surprised if something was hardcoded etc.

For the sake of both TRIM support and more general future-proofing, would it be an idea to virtualize this Linux installation and run it in a VM on a modern system with IOMMU support, exposing these cards to the VM via PCI passthrough?

You might have to use PCIe-to-conventional-PCI adapters to get the cards to work on a modern motherboard. Or you get a somewhat older board or system, that still has conventional PCI slots while also being new enough to have a CPU that support hardware-assisted virtualization with IOMMU support.

Surely an interesting project but not something I'd be willing to explore, mostly because I am not a Linux expert 🙂

Also, with a proprietary software, it would be difficult to iron out issues I feel.
I also feel the system is running hung by a thread right now, adding the complexity of a VM, adaptors and a whole new system running in the background feels... daunting 🙂

Perhaps run memtest just to be sure? Random data corruption may be caused by that...

Thanks I had the same thought, it's been running it since last night - 19 passes, all good 🙂

I'll try with a cold system, I had a laptop ages ago which would corrupt everything but only when cold.

That said, if it was RAM, weird things would happen - this only happens at boot, at the same time every time.
What I cannot be sure - too much times, too many issues - is whether the "corruption" issue was happening on both XD10s I had or just on this one.

I've also installed Windows XP on this machine and run a CPU stress test with no issues - it was to confirm that the fan I selected for the heatsink was up to the task.

tony359 wrote on 2025-09-02, 12:50:

That said, if it was RAM, weird things would happen - this only happens at boot, at the same time every time.
What I cannot be sure - too much times, too many issues - is whether the "corruption" issue was happening on both XD10s I had or just on this one.

Not necessarily. I had this issue once:

The OS was perfectly stable, like months of uptime, no issues. But data written to storage got randomly corrupted. This took way too long to troubleshoot and made me suspect memory right away every time any sort of data corruption happens.

If something was generation checksums and then verifying on boot... But it seem it is not that in this case...

ah good to hear that. I'll bear that in mind.

Memtest has been running for almost 24 hours, no issues.
Shall I try running something under Windows? Or maybe Linux? Someone says memtest might not catch all issues.

What I could try is some storage test to see if by any chance I see issues there.

Memtest catches memory errors pretty well, but not CPU issues. I've fooled around with overclocking and it is totally possible to run memtest for hours with settings which do not even allow windows to boot. But CPU should not be a reason for data corruption...

For storage - IMO the easiest option is using tools designed to detect fake thumb drives - f3 in linux (i know, not amazing name), h2testw in windows. This tools basically fill the device with files along with checksums and then verify. Should catch any corruption.

I was installing windows XP on the SSD then I realised HD Sentinel won't be able to do destructive tests on the OS drive. So I am now installing XP on a PATA drive on the same channel used by the CDROM and the SSD is going to be on the PATA to SATA line.

Will then leave it overnight doing Write/Read test with random pattern.

I can also run h2testw as well. That should give me a good idea of what's happening.

Thanks for your input as usual 😀

took me hours and hours to get windows installed and booted from the PATA drive (mistakes were made!) so that the SSD was not seen as "system" drive.

Then I realised that HDSentinel write/read is not actually verifying the data - if I understand right - but first writing the whole disk then reading it back and basically just stopping at reading errors.

So I ended up running h2testw thanks for recommending. Running overnight. Tomorrow we shall know.

I left h2testw overnight, in the morning I don't see errors, it doesn't seem that the software is showing the outcome of the multiple passes but if memory serves it would show any errors on the Progress window?

If that is the case, then I have 100% a software issue.

Edit: I've noticed h2testw is only writing once and then verifying endlessly. I guess it won't change much. However I will restart the whole process a few times just to be sure. But after 15 hours, so far so good.