Computer harddrive replacement question.

Sep 11, 2023 Last reply: 2 years ago 29 Replies

[Picture]

formatting link
The drive on the right, is a daily-driver and has 4TB of writes on it so far. The drive is at 99% life remaining, and 1% of life is used.

The other two drives are on the Test Machine, and the two drives function as my "NAS". They have less wear on them. They are SSDs for "fast search purposes".

The devices in the picture have a rating of "2400 TBW". You buy SSD drives like bog roll. The more TBW a drive offers, the more you pay for it. When shopping, you must do the research to find the number, as the manufacturers are <cough> "bashful".

On the daily driver, partitions occupy 25% of the device. You might ask "how are the flash cells on the other 75% used?". Well, the storage is virtualized. If you read sector 0 from the outside of the device, that could be flash sector 12345678. When sector 0 is stored, it might well be down near the end of the device. The sectors are stored all over the place. The sectors are moved inside "to even out the wear". A giant table (reliability issue???) records the mapping. The table must be correctly recorded, in the event of an emergency power fail. The firmware to do this, must be awesome :-)

This means, even though the physical partition size is 25% of the device, *all* the flash cells share the wearing action. This is how I get the "value" from the 2400 TBW. It all gets used. I'm running what's known as a "high sparing ratio", which doesn't really mean anything. I would have to manually TRIM the chunk down at the end of the drive, once in a while, so the drive knows those cells are available. I will do that... right now.

Some of the smaller relatives of those things, have "300 TBW" and "600 TBW" ratings. They're sitting in a small pile and work as scratch drives for OS installs. None of those is very worn either.

The wear estimate, takes into account write amplification, and the usage pattern of the drive, can make that better or worse. The device records the "real" usage, and the device knows how much each flash page has been used. There's a counter for each page. Via wear leveling, at end of life, the counter would read "599, 599, 600, 601, ..." indicating the wear leveling algorithm works very well indeed at sucking the maximum life out of the cells. Unlike in a USB flash stick where it reads out "600, 0, 0, 0" indicating uneven wear and "a hole was burned in the TLC". SSD wear leveling is what makes champs of them. Based on the failures I'm getting on USB TLC based flash sticks, my conclusion is they "suck on leveling". SSD drives are very much better than USB flash sticks.

*******

One drive, intended for Chia coin mining, has "12000 TBW", and the price was not disclosed. People might well have been buying those from the manufacturer directly. Like a thousand a piece or more. Those see the worst possible write behavior (imagine beating the piss out of a disposable hardware costing a thousand). That's why they need the 12000 number. A hard drive could exceed the 12000 number, but... the storage would be "slug slow" by comparison.

So capacity is one thing. Obviously, you buy capacity to store your goods. If you have 1TB of goods, then you need at least a 1TB drive. But excess capacity is not lost, as like a "bog roll", you get to use the entire roll.

Now, one of those could just "up and die" in the next ten seconds. I cannot predict what that statistic is. First generation devices had terrible firmware. When Intel got into the business of making SSDs, they nearly had a "s*it-fit" when then examined the source for that firmware. There was a marked improvement in industry practice, and I have no idea what went on behind the scenes. There are still occasionally bad firmwares (like the Samsung 2TB NVMe issue), but for the most part, they're every bit as predictable as HDD. Seagate has botched HDD firmwares too. Sometimes you have to flash up a HDD, before it is too late (metadata corruption).

Be aware, that people buy HDD, they clone over from old HDD, they erase the source HDD, and a couple days later... the new HDD dies. To avoid issues with infant mortality when cloning over, you should wait a couple months to see if the new drive is a champ or not.

If you hear a new HDD "clicking", that is generally a bad thing. I discovered a brand new HDD drive here clicking, and it turned out the root cause was the power supply was running at 11.6VDC . The drive is supposed to accept +/-10%, but it was obvious the voltage detector on the drive was using a much tighter window than that. Some HDD will get into a "spin down/spin up" loop due to the voltage issue. So when you hear a drive click, use your multimeter and verify that +5V rail and +12V rail are not too far out of spec. Now, that new drive, does not have to go back to the store.

The new supply is running at 12.27 VDC. If the power supply voltage rose to +15V, then a protection device on the drive will start to burn, and the drive will stink :-) Because the controller board is upside-down, you can't even see the black spot. The protection device exists to clobber HotPlug transients, and is not intended to "discipline" hard drives for going out of spec.

Paul

I find that most 10+ year old 4th gen 15 boxes are fine on Win 10 as well... generally if you did not buy the cheapest at the time, PC life is reasonably decent regardless of OS.

There will be no deficit - it will be as quick or quicker than it was before. However you would be throwing away a very easy and very sizeable performance gain that could be realised.

The main change you would see with a NVMe SSD as the main OS and applications drive would be reduced load time. (e.g Photoshop CS5 in ~2 secs, Word 2021 in 500ms etc) - much less performance hit if you run out of ram on a large image and the system needs to start paging.

easily solved if you put in 128 GB RAM as I did, and my mobo has two M2 slots so have two Samsung Pros in there, one being the system drive and the other as the swap file and paging.

The Asus mobo I have can also support a

formatting link

where I can insert up to 4 M2 NVME drives set up as a RAID array with theoretical speeds of up to 256 Gbps.....

It can be used on other mobos that can support PCIe bifurcation via BIOS so you can add in extra NVME drives but they will come up as basic drives and no RAID is possible (unless possibly software based raid like MDADM or mdraid (?) in Linux

:-)

You can use CPUZ to show the details for your system. At least, if you're on Windows. On Linux, it would be "inxi -F".

[Picture]

formatting link
One thing that an SSD can help with, on Windows 10, is reducing seek time while Windows Defender scans "critical areas" just after boot finishes. That can suck the life out of a hard drive machine.

Some other tasks, later on, when the system is quiet, may not show quite as much of a change. John has pointed out some cases where a difference definitely could show up.

When Windows 10 does "Windows Update', that visits a lot of small files, and that is a load for a system with a regular hard drive. An SSD can help with that.

Paul

It scales pretty good. Gets about 26GB/sec out of a possible 4*7GB/sec = 28GB/sec. I would have expected the RAM speed to slow it down.

formatting link
There are some pathologies for setups like that. Once you're using a file system stack on top of that, that'll let the air out of the tyres. For example, try transferring 10GB of small files to the card (to an NTFS partition), and see if that takes half a second. It will likely take longer.

The most impressive part of an NVMe RAID0, is the sequential graph.

Paul

I have 16 GB RAM which is seems to be alright so far!

What if...

I replicate the current set-up of a small boot drive in the motherboard's NVMe slot, and then have a larger storage by adding a PCI NVMe card & drives? Or is that just overkill?

Of course you can.

But I'm surprised your motherboard does not have two NVMe slots on it already. Typically one next to the CPU, and a second position lower on the motherboard, running off a PCH bus connection.

This talk of PCI Express NVMe cards is discouraging, because we as users are paying a heavy "tax" of having expansion slots removed, and the slots replaced with a place to screw down an NVMe. To waste what few slots are left on a PCIe NVMe card, that's sinful :-) Not every motherboard supports bifurcation, and without bifurcation, I would expect the PCIe NVMe card to hold just one NVMe stick.

On my motherboard, the slot next to the CPU is Rev4, the slot nearer the bottom of the motherboard is Rev3. The one NVMe I own, is a Rev3, and fits in either slot, and does not get a significant advantage moving between slots. However, *if* I'd bought a Rev4, then it definitely goes in the CPU Rev4 slot, if I expect to impress people with my Crystal benchmark. It would still work in the Rev3 slot, but my numbers, I'd be "so embarrassed" (or something). Well, my NVMe isn't even plugged in... :-) My storage is a SATA SSD, which is easy to unplug.

The thing is, the lower NVMe will be a data drive, and nobody gives a rats ass whether it's Rev3 or Rev4. Remember, we used to own a HDD, and any NVMe is light years better... The presentation of the video below, is purely to show what too much caffeine does to you.

*******

For data storage, you may find 4TB and 8TB ones. The 8TB could be QLC chips (less desirable), the 4TB could be MLC-like VNAND or they could be TLC. If the flash is TLC, it may lack write consistency (writes fast at first, slower later on when the flash cache [a set of SLC cells] is full). In the video, one thing we'd be looking for, is whether the write performance stays consistent over the entire 4TB surface of the device. I would prefer a magazine-format for review of performance, using video for this, sucks. If it's one thing I hate in my storage devices, it's "rough idling" caused by inconsistent write performance. This matters, if you're "waiting for something to finish" and the CPU happens to not be the bottleneck. Typically, CPUs can't do much better than 2GB/sec, but who is counting (Irfanview reads can happen at 2GB/sec).

Samsung 4TB SSD 990 PRO (press release recently)

formatting link
Controller chips, might typically have 8 channels. On an NVMe, there isn't room for a bushel of chips. (A 2.5" SSD has room for more chips.) But a larger flash chip, could have a very large array of balls on the bottom (FPGA), and a single large chip could have, say, four channels, and there could be enough "planes" inside the device, to give the equivalent of four smaller chips. If the datasheet or even a summary sheet were available, these are details a video maker could look up.

Now, you don't absolutely need the fastest device:

PCIe Rev3 3500MB/sec regular bus connection PCIe Rev4 7000MB/sec regular bus connection, yes, a small heatsink helps PCIe Rev5 14000MB/sec needs redrivers, this NVME needs *serious heatsink*, wastes mains power This is a sign you've gone too far.

In the lab, the Rev5 has been seen at 10000 and 12000, and they were waiting on flash chips with faster I/O pads to come out. The Rev5 also has *different dimensions* so won't fit old motherboards. The signal frequency is so high, motherboard transmission losses must be compensated for with redriver chips. This is silly, and the high power dissipation of Rev5 is nothing to look forward to.

Thus, the lot you see in the video, they will fit in your Rev4 slot (if you have one). Via backward compatibility, a 7000 device will easily fit in a 3500 slot, and run just fine. And the dimensions are the same, in terms of width. The length, your mobo has several screw holes and those denote the lengths supported.

The tiny screw and the clumsy installation, is why mine is sitting in the cardboard box. Now that I have a ginormous CPU heatsink, it's getting harder to get hands to the CPU slot one. But the new heatsink has stopped the temperature transients, so I'm happy with the result. I was getting 90C spikes, in a fraction of a second, with the previous heatsink.

Paul

That's what I do. Windows 8.1 on a 128GB NVMe M.2 Data on a 1TB spinner Windows 10 & Windows 11 on a partitioned 256GB NVMe Boot is taken care of by the free version of Easy BCD -

formatting link
The vastly superior read & write speeds of M.2 drives are ideal for the operating system drive(s) but not necessary for normal data usage.

They are very similar in price now (and quite cheap) so you might as well go for an NVMe if your machine will take it.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required