Category Archives: Benchmarking

Yoga Pro 9i Shows Incredible SSD Speed Variations

I’m digging into the behaviors of the svelte and powerful Yoga Pro 9i I’ve had for two weeks today. It’s a speedy and powerful beast of a laptop. It’s half the thickness (30.23mm/1.2″ vs 19.4mm/0.77″ on average) and ¾ the weight (2.95kg/6.5 lbs vs 2.23kg/4.9 lbs) of the Lenovo P16 Mobile Workstation (Gen1). But it’s either on par with or faster than that bigger beast of a desktop replacement. All this said, though, running various NVMe drives and enclosures, I’ve observed that the Yoga Pro 9i shows incredible SSD speed variations.

Why Say: Yoga Pro 9i Shows Incredible SSD Speed Variations

The first set of CrystalDiskMark (CDM) results for the Yoga Pro 9i serve as the lead graphic up top here. These come from the internal SSD inside the unit’s M.2 drive slot. According to Device Manager that drive is an SKHynix_HFS001TEJ9X115N (1TB PCIe x4 NVMe 1.4). Those are pretty respectable results, and serve as a point of reference against external drives.

What makes the Yoga Pro 9i interesting is its two USB-C ports. One is labeled USB-C (20 Gbps) and the other is labeled Thunderbolt 4 (which means 40 Gbps) [see the ports diagram from this April 29 post]. Theoretically that means port 3 (USB-C 20 Gbps) tops out at half the speed of port 4 (USB-C Thunderbolt 4 40 Gbps).

And indeed my only Thunderbolt 4 NVMe enclosure — an Acasis TB-401u claims to support that 40 Gbps top rate. The on-the-ground reality is, however, something quite different with a Sabrent Rocket 1TB NVMe 1.3 SSD  installed therein. Much of this comes from an older v1.3 SSD inside a 1.4 enclosure with access to TB4/USB4 compatible ports. But these results fall far short of what I’d expected to see:

Yoga Pro 9i Shows Incredible SSD Speed Variations.acasis

This looks like results for a typical USB 3.x UASP device IMO

In fact, I got at least some better results from a less-capable Crucial CTP2000P3SSD8 (2TB, NVMe 1.3) inside a less capable enclosure (Sabrent EC-NVME: USB 3.1 Gen2) in the slower USB-C 20 Gbps port. Here they are:

Yoga Pro 9i Shows Incredible SSD Speed Variations.sabrent/crucial

Big bulk reads (top left) are much faster, but everything else is (mostly) slower.

There’s a lot of interesting stuff going on here. What I take from it is that for the fastest backups and big file transfers (video, AI models, and the like) you’re better off spending more on a faster enclosure and a faster SSD to get the most out of the connection. I’m going to have to systematize this, and run some more tests. Great fun!


Lenovo Yoga Pro 9 Intake

When I got home from a visit to a dental lab around lunchtime on Friday, the Boss asked “Were you expecting a package?” I’d asked Lenovo to send me a Yoga Pro 9 earlier that week, so my answer was a tentative “Maybe…” And sure enough, that’s what it was. Over the weekend, I had time to get through all steps in the Lenovo Yoga Pro 9 intake process. It proved more interesting — and educational — than I expected…

What I Got for Lenovo Yoga Pro 9 Intake

There were some interesting surprises in what showed up. Basics of the unit’s configuration include:

  • Intel Core Ultra 9 185H (Meteor Lake/13th Gen+)
  • 32 GB LPDDR5x-7467 (soldered)
  • Hynix 1TiB NVMe SSD PCIe x4
  • 16″ Lenovo LEN8BAI Monitor 3200×2000 resolution monitor
  • Intel Wi-Fi 6E AX211 network adapter
  • Intel AI Boost NPU & Copilot key

There’s more, but I’ll get to some of that in the next section. The main reason I requested a short loan of this formidable PC was for access to a machine with NPU and Copilot key to take them for a spin. Looks like this unit retails for around US$2,100 at the Lenovo Store.

What I  Learned During the Intake Process

TLDR answer: LOTS of things. I’ll elaborate by noting first that the unit came with Windows 11 Home installed (immediately upgraded to Build 22631.3527 Enterprise). Because I usually interact with most PCs — personal, production and test/loaner units — via RDP, sticking with Home was not an option for me. It’s OK: because I’m an MVP I get a MAK key for Enterprise as part of my Visual Studio subscription. Lenovo will destroy my image upon its return anyway. But if you decide to purchase one, you can indeed configure it with Pro for a mere US$2 extra. That’s what I’d do, for sure…

I found myself a little mystified by the new Meteor Lake Intel Core Ultra 9 185HCore Ultra 9 185H CPU. Intel refers to this CPU as “formerly Meteor Lake” but doesn’t really assign a “Generation” number. Its Intel home page studiously avoids mentioning such info. My unit was built in early February 2024 according to its outside sticker. Its Intel Ark page describes it as Intel Core Ultra processors (Series 1) so it looks like NPU endowed chips are starting a new numbering scheme instead. This should be interested to see play out, expecially with Snapdragon X systems on their way into this same niche.

I also observed that read/write speeds vary significantly by USB-C port type. As you can see in the next graphic, port3 is USB -C 20Gbps, and 4 is Thunderbolt 4. These produce “interesting” benchmark results where one is noticeably faster than the other for some values. Indeed, TB4 is faster for 1M read and 4K random writes, while USB 4 is faster for 1M write and 4K random reads. Others are more or less a wash. I’m going to have to try faster SSDs to see if that makes a difference (I suspect it will).

Lenovo Yoga Pro 9 ports (left & right sides)
Lenovo Yoga Pro 9 ports (left & right sides) [Double-click image for full-size view]

What About AI Stuff?

I can tell that Copilot runs faster on this laptop than on other, older models (even a ThinkPad P16 Mobile workstation with a 12th-Gen i9-12950HX CPU but no NPU). But other than that I haven’t really messed around enough with Copilot and other AI functions to get a sense of the differences. Stay tuned! I only get to keep this unit for a month, so I’ll be writing about it regularly over the next few weeks.

Other Observations

Here are some bullet points that reflect other stuff I noticed while unpacking, setting up and using the new Lenovo Yoga 9 Pro:

  • The shipping materials proudly proclaim “plastic-free packaging” in several places on the boxes. Two egg-crate holders supported the laptop, with one small internal cardboard box for the brick and power cord. There was some soft material labeled 22/PAP between the upper and lower decks of the clamshell. Ditto for the label on the black bag inside which the laptop itself was sitting. The material uses a plastic-recycling symbol (three arrows forming a triangle) but lookup tells me … yep, it’s paper! Even the twist-tie that held the power cord together was covered in brown paper. Good job, Lenovo.
  • For some unholy reason, Lenovo included McAfee AV on the Yoga 9 Pro. I uninstalled it right after I performed the OS updates on that PC. Defender is fine with me: I no longer use much, if any, third party security software.
  • Have to laugh: the Copilot key is a big deal on these new Windows AI-Ready PCs. But the onscreen keyboard (Ctrl+Winkey+O) does not include such a key. I bet MS will fix this before these AI-Ready PCs get into wider circulation.
  • The Open Source Snappy Driver installer (SDIO version) gives the drives already installed on this laptop its blessing. It’s not an absolute guarantee that everything’s up to date, but it’s pretty darn close. Good-oh!

Newer USB Justifies Added Costs

I had a revelation via contrasting benchmarks yesterday. A friend returned a mid-range USB 3.1 NVMe drive enclosure after an extended loan. Thus, I popped it into my production desktop (an i7 Skylake Gen 4 PC) to see how fast it ran. Good enough. Then, just for grins I popped it into the 2021 vintage Lenovo P16 Gen 1 Mobile Workstation (an i9 Gen 12 PC). Much faster! Enough so, in fact, that it’s clear that newer USB justifies added costs of acquisition. Let me explain…

Why Say: Newer USB Justifies Added Costs?

Take a look at the lead-in graphic. It shows the difference between older USB technology in the Skylake desktop vs. newer USB technology in the Gen 12 mobile workstation. Both are using USB 3.1 ports (though the older PC goes via USB-A, the newer goes thru USB-C) to the same hardware running the same benchmark. Why is the new so much faster than the old?

Short answer: UASP, aka the USB Attached SCSI Protocol. The newer PC supports it, while the older one does not. You can see there’s a driver difference in Device Manager when it comes to accessing the NVMe drive enclosure and its installed SSD: the older machine runs a driver named USBSTOR.sys, while the newer one runs UASPStor.sys. Plain as day.

The Deal With UASP

The Wikipedia article on UASP is a good place to find some explanation. To wit: “UAS [USB Attached SCSI] generally provide faster transfers when compared to the older USB Mass Storage Bulk-only (BOT) protocol drivers.” In a nutshell, that’s UASPStor.sys versus USBSTOR.sys.

As I learned about this technology in the period from 2016 to 2019, the word at ran something like “Speeds of 500 MBps mean USB bulk transfer; 1 Gbps or better means UAS transfer.” And that, dear readers, is the difference you see between the right-hand side in the lead-in graphic (USBSTOR.sys on the Skylake) and the left-hand side (UASPStor.sys on the Gen 12).

In practical terms, this translates into much, much faster IO on the newer PC vis-a-vis the older one. I think it’s incredibly worthwhile, given that backups complete 2-3 times faster on the P16 than the Skylake. Likewise for big, bulk file transfers (such as Windows ISOs, which I mess with frequently).

Retrofit and Replacement

Does this mean one has to toss older PCs and replace them with newer models? Maybe, but not necessarily. For between US$50 and 100, you can purchase UASP capable PCIe adapter USB cards. As long as you’ve got an open PCIe x4 port available on your motherboard (desktops only, so sorry) this could be a good solution. I’m a fan of this US$95 StarTech unit for that purpose.

Older laptops can be dicey and depend on support for USB ExpressCards. I mucked around with these on some 2012-vintage Lenovo ThinkPads in the 2014-2016 timeframe (an X1 and a T420). They work, but they’re cumbersome and expensive (see this Amazon Review for a great discussion).

For best results, it may be time to shell out for a new desktop or laptop PC. That way, the fastest USB (and even Thunderbolt) technologies are likely to come built-in and ready to go. Could be worthwhile!




Thunderbolt Docks Add Helpful Future-Proofing

I’m thinking about what kinds of hardware experiments I’ve conducted over the past couple of years. Especially this year (2022). Along the way, I’ve learned that Thunderbolt docks add helpful future-proofing for home and office users. Let me explain…

How Thunderbolt Docks Add Helpful Future-Proofing

Right now, Lenovo offers what can only be called a “Best Buy” in the arena of Thunderbolt 4 docks. Or maybe a couple of them, as I’ll recount shortly. Called the Universal TB4 Dock, it currently retails for just under US$290. This is about US$110 cheaper than its nearest competitors (e.g. Belkin and CalDigit, among others).

On December 8, I also wrote here about the Lenovo P27-u20 monitor, which includes a built-in TB4 dock. At US$527, with a 4K monitor included in the mix, it too, qualifies as a “Best Buy” IMO.

There is one thing, though: to make proper use of TB4, you also need TB4 peripherals. They will be no more than two years old (TB4 made its debut in H2’2020). There’s a lot of expense involved in climbing this technology bump. But if you’ve got newer peripherals, a TB4 dock is a great way to mate them up to PCs and laptops back to 8th Gen Intel (and equivalent AMD) CPUs. I’ve done that, and it works great.

Try TB3 for a Lower-Budget Approach

For readers who want to extend the life of a Windows 11 capable PC or laptop, it may make sense to invest in Thunderbolt 3 (TB3) instead. Such docks cost as little as US$40 (e.g. Dell refurb), and are readily available new for around or just under US$100. If you’ve already bought into USB-C (3.1 or 3.2 capability) or TB3 peripherals, this is a less expensive way to dock up. Worth researching anyway: I see lots of attractive options at Amazon and other online outlets.

Thanks, Lenovo!

While I’ve got your eye, I’d like to thank the laptop and peripherals teams at Lenovo for their outstanding support. They’ve sent me half-a-dozen different laptops (and one great SFF workstation), multiple docks and the aforementioned monitor this year to review.

It’s been incredibly educational and lots of fun to put different TB4 scenarios together. This lets me understand and measure how they work, and how to make them work best. A special shout-out to Jeff Witt and Amanda Heater for their great help and quick assistance this year (and beforehand). Happy holidays to one and all.


Windows 11 22H2 File Copy Fix Works

OK, then: I read the WinAero story about fixing the “slow file copy bug” in Windows 11 22H2. Indeed, it picqued my interest. “Hmmm,” I thought, “Maybe I can see on the P16 Mobile Workstation?” Yes, I could. I’m happy to confirm that the Windows 11 22H2 file copy fix works — on that PC, at least. What does this mean?

Take a look at the lead-in graphic. It’s a paused file copy. The file comes from my external F: Drive. (That’s a Sabrent Rocket 4 Plus 1 TB PCIe x4 NVMe SSD in a USB4 Acasis drive enclosure.) It’s copied to my built-in C drive. (That’s an internal Kioxio 2TB PCIe x4 NVMe SSD). Except for a dip about half-way through, it shows data rates from 1.2 to 2.3 GBps for a 20-plus GB file copy (a Macrium Reflect backup image).

That’s much, much better than the 600 – 950 Mbps I’d observed the last time I tried this with the same pair of devices. Looks like KB501738 issue does indeed get resolved in the latest Dev Channel Build (25252). I’m jazzed.

More Data: Windows 11 22H2 File Copy Fix Works

Even my slower USB3.2 NVMe Sabrent PCIe x3 with its older Samsung 950 1 TB SSD also shows a similar improvement. It shows a range of 750 MBps to a momentary high of 1.1 GBps in its copy of the same Macrium image file instead.

Gosh! It’s always nice when a usable performance bump occurs. It’s even better when the bump is both noticeable and measurable. And it makes the cost of relatively expensive NVMe drive enclosures more tolerable — maybe more justifiable, too — when the bump helps improve productivity.

Who knows? I might need to rethink my current take that paying US$100 extra to upgrade a USB3.2 NVMe enclosure to USB4 is too expensive. Stay tuned: more to follow next week!


No Remote WinSAT No Batteries

In following up on yesterday’s memory training item, I started messing about with WinSAT. For those not already clued in, WinSAT stands for Windows System Assessment Tool. As it turns out, such assessment depends on steady, reliable power and “close to the metal” access to the PC it’s assessing. That’s why, I believe that MS says “You cannot run formal assessments remotely or on a computer that is running on batteries.” (Using WinSAT). Hence the assertion: no remote WinSAT no batteries.

If No Remote WinSAT No Batteries, Then What?

A formal assessment on WinSAT runs a whole battery of checks. You can still do feature-by-feature checks remotely (just not the whole thing). Here are the results of WinSAT mem over a remote connection to one of my 2018 vintage Lenvo X380 Yoga ThinkPads:

No Remote WinSAT No Batteries.rem-mem

A single feature check — mem, or memory — does work remotely.

But if I run the whole suite (WinSAT formal) in the same PowerShell session, I get an error message instead:

No Remote WinSAT No Batteries.rem-formal

Going formal with WinSAT “cannot be run remotely…”. No go!

Such things lead to head-scratching from yours truly. I can kind of get it because it’s undoubtable that the remote connection is going to affect results reported because of the time involved in remote communications. But why allow checks one-at-a-time, but not all-at-once? MS is mum on this subject, so I’m not getting any insight there. It could be that singleton checks add relatively little overhead, but that cumulative effect of an entire suite of same adds noticeable delay. Who knows?


USB4 Delivers Consistent NVMe Performance

OK, then. I finally laid hands on my second USB4 NVMe SSD enclosure yesterday. I deliberately sought out the cheapest one I could find so I could compare it to a more expensive alternative already on hand. When I say that USB4 delivers consistent NVMe performance here’s what that means:

1. The same SSD, cable, and host PC are used for comparison. Both drives have the “cache tweak” applied (this Oct 14 post has deets). Same tests performed, too (CrystalDiskMark and a Macrium Reflect backup).
2. The only thing that changes is the enclosure itself.

In short, I wanted to see if spending more on hardware returned a noticeable performance advantage (I’ll talk more about this below). Long story short: it doesn’t seem to make much, if any, difference. Let me explain…

Why Say: USB4 Delivers Consistent NVMe Performance?

The lead-in graphic shows the results from the cheap enclosure on the left, and the more expensive one on the right. The average difference in CrystalDiskMark performance shows 2 wins for el cheapo, 5 wins for the higher priced item, and 1 tie. On first blush, that gives the more expensive device an advantage. So the next question is: how much advantage?

This is where a little delta analysis can help. I calculate that the average performance difference between devices varies from a high of 6.2% to a low of 0.03% (not including the tie). That said, the average performance difference across all cells is merely 1.54%. (Calculated by taking absolute value for each delta, then dividing by the number of cells.) That’s not much difference, especially given the prices of the two devices: $128.82 and $140.71. That delta is 8.4% (~5.5 times the average performance delta).

I will also argue that comparing CystalDiskMark results is interesting, but not much of a real-world metric. Thus, I’ll compare completion times for a Macrium Reflect image backup on the same PC, same OS image. The expensive device took 2:25, the cheap one 2:44. That’s an 11.5% difference, greater than the price delta but not amazingly so.

Deciding What’s Worthwhile

I can actually see some differences between the two enclosures I bought. One thing to ponder is that NVMe drives tend to heat up when run full out for any length of time (as when handling large data sets, making backups, and so forth). I’ve seen temps (as reported in CrystalDiskInfo, reading SMART data) go as high as 60° C while M.2 SSDs are busy in these enclosures. At idle, they usually run at around 28° C. The more expensive NVMe enclosures tend to offer more surface area to radiate heat while active, so that’s worth factoring into the analysis.

But here’s the deal: I can buy a decent USB3.1 NVMe enclosure for around US$33 right now. The cheapest USB4 NVMe enclosure I could find cost almost US$96 more. That’s a multiplier of just under 4X in price for a device that delivers less than 2X in improved performance. Let me also observe that there are several such enclosures that cost US$160 and up also on the market. I still have trouble justifying the added expense for everyday use, including backup.

There will be some high-end users — especially those working with huge datasets — who might be able to justify the incremental cost because of their workloads and the incremental value of higher throughput. But for most business users, especially SOHO types like me, the ouch factor exceeds the wow value too much to make it worthwhile. ‘Nuff said.


P360 Ultra Gets Second NVMe

A couple of days ago, I praised the interior design of the Lenovo P360 Ultra SFF PC (link). I just had to remove the GPU to access the second NVMe slot on an Asrock B500 Extreme4 motherboard last week. Let’s just say it wasn’t incredibly easy (and some expletives were involved). That really made me appreciate an install that required less than two minutes all the way around. But now that the P360 Ultra gets second NVMe, I want to report on the results.

When P360 Ultra Gets Second NVMe, Speed Abounds

What you see as the lead graphic for this story is a pair of CrystalDiskMark results. To the left, the internal C: furnished with the PC (a Samsung 1TB OEM drive: MSVL21T0HCLR). To the right, the internal D: I installed (WD Black SN850).

First, let’s look at those results. The Samsung drive enjoys an 18% edge on the sequential read (queue depth 32, single thread) and a  33% gain on random read (queue depth 1, single thread). The WD Black comes out ahead on all other readings.

That’s not surprising, given that the WD Black SN850 is a newer, more capable drive. But those results also speak to the notion that one should definitely populate open NVMe slots if speedy storage is helpful to the workloads a PC must handle.

P360 Ultra Gets Second NVMe.external

Same WD Black drive in a USB4 external NVMe enclosure: much slower.

Internal vs. External NVMe

The preceding screengrab shows CrystalDiskMark results for the same drive, but housed in an external NVMe enclosure. It happens to be a USB4 enclosure, and represents as much speed as I’ve been able to get from an external NVMe drive. It’s significantly slower across the board, but still not bad.

If I drop the same drive down to a USB 3.1 enclosure, it runs at standard UASP speeds (at or under 1000 in the top 4 cells). Interestingly the bottom four cells don’t change much for either USB4 or USB3.1. Backup speeds don’t change that much, either. That’s why I’m not convinced the USB4 enclosure is worth a $100 premium (it improves backup speeds by 30 seconds, give or take).

One More Thing…

If you’re buying an NVMe drive for an external enclosure, there’s no need to spend big on a fast, capable storage device. It won’t be able to run full out because the USB link (either 3 or 4) can’t keep up with top-end NVMe speeds. As the preceding CrystalDiskMark chart shows, you can’t come near the 6-7 GBps or so performance that top-end NVMes deliver these days.

On the other hand, if you’re going to put that device into an M.2 slot INSIDE the PC or laptop, that’s a whole ‘nother story. Then, you should buy as fast as you can stand to pay for — assuming, that is, that the PC or laptop can make full use of those capabilities.


Thinking About Windows 10/11 SSDs

I’m still busy benchmarking away on the two Thunderbolt4/USB4 PCs that Lenovo has recently sent my way. But as I’ve been doing so, I’ve been thinking about Windows 10/11 SSDs in general. On that path, I’ve realized certain principles that I’d like to share with you, dear readers.

I’m spurred in part to these statements from a sponsored (and pretty contrived) story from MSPowerUser entitled “Is NVMe a Good Choice for Gamers?” My instant response, without reading the story — which actually focuses on storage media beyond the boot/system drive — was “Yes, as much as you can afford.” Spoiler alert: that’s what the story says, too.

Where Thinking About Windows 10/11 SSDs Leads….

Here are some storage media principles that flow from making the most of a new PC investment.

  1. The more you spend on a PC, the more worthwhile it is to also spend more on NVMe storage.
  2. Right now, PCIe Gen4 drives run about 2X the speed of PCIe Gen3 drives. They don’t cost quite twice as much. Simple economics says: buy the fastest NVMe technology your PC will support.
  3. Buy as much NVMe storage as you can afford (or force yourself to spend). For pre-built PCs and laptops, you may want to buy NVMe on the aftermarket, rather than get the drives pre-installed. Markup on NVMe drives can be painful. Hint: I use Tom’s Hardware to keep up with price/performance info on NVMe SSDs and other PC components (it’s also the source for the lead-in graphic for this story, which still prominently displays the now-passe Intel Optane as an SSD option. Caveat emptor!).
  4. Corollary to the preceding point: fill every M.2 slot you can in your build. For both my recent Lenovo loaners — the P360 Ultra and the P16 Mobile Workstation — that means populating both slots with up to 4TB each. Right now, the Kingston KC3000 looks like a 4TB best buy of sorts.

Thinking Further (and Outside the Box)

More thoughts in this vein, with an eye toward external drives and multi-tiered storage (archives and extra backups):

  1. If you’re going to put an NVMe SSD in an external enclosure, you will be OK for the time being in a USB 3.2 rather than a USB 4 enclosure. Right now, the newer enclosures cost more than twice as much but don’t deliver anywhere near 2x the speed (except on synthetic benchmarks — I used C: imaging times as a more reliable indicator). Over time this will no doubt change, and I’ll keep an eye on that, too.
  2. I don’t consider spinners (conventional mechanical hard disk drives, or HDDs) any more, except for archival and inactive storage. If I need something for work or play, it goes on an SSD. If I might need something, someday (or to restore same) then it’s ok on an HDD.

I used to restrain spending on NVMe SSDs because of its high price differential. I’m now inclined to believe that restraint is a false economy and forces less productivity as a result. That’s why I’m rethinking my philosophy. I haven’t quite yet gotten to Les Blanc’s famous dictum (“Spend It All”) but I am coming around to “Spend As Much as You Can”…

Remember This Fundamental Assumption, Tho…

My reasoning aims at high-end PCs where users run data-, graphics-, and/or compute-intensive workloads. It does not apply, therefore, to home, hobbyist, and low-end office users. For them typical productivity apps  (e.g. MS Office or equivalent), email, web browsing and so forth predominate. They wouldn’t need, nor benefit much from, buying lots of fast NVMe storage. That said, a 1 TB fast-as-possible NVMe for the boot/system drive is the baseline. Other storage options will balance themselves against budget to dictate other choices and PC builds for such users.

In different terms, if you’re not maxing out your PC running data analytics, 3D models and other high-end graphics rendering, or AI or machine learning stuff, this advice is most likely overkill. Too, too costly. But for this user community, more spent on NVMe (and GPUs and memory as well) will repay itself with increased productivity. ‘Nuff said.


Backblaze Data Confirms SSD Trumps HDD Reliability

It’s always made sense on an intuitive basis. Hard Disk Drives (HDDs) include spinning platters, moving arms with read/write heads, motors to power things, and gears to control action. SDDs are made entirely of circuitry: no moving parts. Thus, it’s compelling to assert that SDDs should be more reliable, and less prone to failure than HDDS. And indeed, the latest 2022 Drive State report from online backup and storage provider Backblaze weighs in on this topic. As I read it, that Backblaze data confirms SSD trumps HDD reliability.

The lead-in graphic shows 4 years’ worth of SSD data vs. 8 years for HDDs for boot drivers in their thousands of datacenter based servers. Whereas there’s a dramatic upward knee in the curve for HDDS between years 4 and 5 (from 1.83% to 3.55%), failures actually dipped for SDDs during that interval (from 1.05% to 0.95%). Interesting!

How Backblaze Data Confirms SSD Trumps HDD Reliability

The afore-linked report explains that boot drives function in multiple roles on the company’s plethora of storage servers. They store log and temprorary files; they maintain storage holdings based on each day’s storage activities and volume. The disparity in the number of years for which data is available comes from later adoption of SDDs as boot drives at BackBlaze. That practice started in Q4 2018. Today, all new servers boot from SSDs; older servers whose HDD boot drives fail get SSD replacements.

The numbers of SSDs keep going up, too. The end-of-year 2021 SSD report encompassed 2,200 SSDs. By June 30, 2022, that count grew to 2,558. Failure rates for such devices show much lower numbers than for HDD (see the tables labeled Backblaze SSD Quarterly Failure Rates in the latest report for more detail). Models included come from the following vendors: Crucial, Dell, Micron, Seagate and WDC.

Note: the report itself says:

For any given drive model in this cohort of SSDs, we like to see at least 100 drives and 10,000 drive-days in a given quarter as a minimum before we begin to consider the calculated AFR to be “reasonable”.

The real news, of course, is that quarterly, annualized and lifetime failure rates for SSDs are significantly lower than for HDDs, based on Backblaze’s own long-running data collection. Thus their conclusion comes with the weight of evidence “…we can reasonably claim that SSDs are more reliable than HDDs, at least when used as boot drives in our environment.”

Good stuff! As for me, I like SSDs not just because they’re less prone to failure. They’re also FAST, if more expensive per storage unit than spinners.