Ever wondered how long your SSD will last under heavy use in your Computer?

[Nico Albrecht]

Ever wondered how long an SSD can last under heavy use? I’m running full write endurance tests on brand new Samsung SSDs – the **990 PRO, 990 EVO, and 870 QVO** – to see when they’ll fail and how their speed decreases over time.

My opinion? SSDs aren’t as reliable or fast as they seem under heavy load, and the performance drop is already noticeable just hours into **Day 1**!

Join me for fun and educational testing as I push these SSDs to their limits.

**Live Stream Day 1:**

**Full Case Study & Updates:**

This is my first project, so bear with me as adjustments may be made over the next few days. Since there’s little real-world data on modern SSD endurance, all the drives I’m testing are brand new from the current lineup—and personally funded to be pushed to failure!

Follow along as I log each day and update the study!

Any question or feedback welcome. Next will be Chinas finest SSD's and some cheaper entry level ones.

 

[Ozzy]

SSDs aren’t as reliable or fast as they seem under heavy load, and the performance drop is already noticeable just hours into **Day 1**

What constitutes 'heavy load' in this? I'm genuinely curious as we're running, as I'm sure many do too, SSD's in our infrastructure for the claimed performance load.

Then, at home, in my gamer station, I'm running NVME which does appear loads quicker than standard drives, but I guess that isn't under heavy load.

 

[Nico Albrecht]

I'm genuinely curious

That’s why I’m running these endurance tests, with more brands lined up for the future. The Samsung 990 Series (NVMe) and 870 QVO (SATA) are the latest tech, but after just 2 hours, we're already seeing differences in performance.

Any question or input welcome.

• 990 PRO: Holding steady at 500–800MB/s after several write cycles (1 cycle = 1TB written).
• 990 EVO: Performed well on the first cycle, but dropped below 100MB/s on the second cycle.
• 870 QVO: Started around 350MB/s but has already dropped to around 80MB/s. 15 min in

Marketing often shows off impressive SSD speeds, but these are typically based on ideal, short-term conditions. Realistically, high-end NAND chips max out at around 250MB/s per channel, and most consumer SSDs use 8 channels, giving around 2000MB/s total.

Some SSDs use two NAND chips in RAID 0 for a theoretical 4000MB/s, but these numbers are before error correction, wear leveling, and trim kick in. These factors significantly reduce real-world performance. Some high-end SSDs use a RAM cache that can burst up to 16GB/s briefly, but this only works in synthetic benchmarks, not in everyday use.

I’ve always believed that constant writes to SSDs eventually cause performance to degrade to levels similar to mechanical drives. The developer of HD Sentinel is onboard to ensure our testing software isn’t the bottleneck.

Currently, all three SSDs are separated on individual PCIe slots, so bandwidth shouldn’t be an issue, but I’ll rotate the cards on Day 2 for further analysis.

I also believe power cycling—turning off the computer overnight—might reduce SSD lifespan. During operation, the firmware is cached internally, but after a power cycle, the SSD has to re-read the firmware (1-5GB) from the NAND chips, which could add wear over time.

 

[Nico Albrecht]

Also I have a MacBook M3 test in mind for write endurance down the line. Since the M3 uses very cheap lower end memory chips from Toshiba and Sandisk I am curious myself how many TB it can handle before dying.

 

[Nico Albrecht]

What constitutes 'heavy load' in this? I'm genuinely curious

In real-world scenarios, a heavy workload might involve media companies shooting in 4K, where each project can range from 100GB to 400GB. When factoring in editing and output, the 600TB of lifetime writes quoted by Samsung could easily be reached within a year.

Home servers and NAS are also becoming increasingly popular, especially with platforms like Proxmox and ESXi. Running virtual machines (VMs) can put significant stress on SSDs, especially if you’re running RAID on top of the VMs, and then adding a database inside the VM. This kind of setup can keep SSDs under constant load, causing them to fill up quickly.

While a typical user might only write 4-10GB per day, more demanding tasks like running VMs, video editing, or similar activities can rapidly accelerate write usage. In such cases, the SSD lifespan, which is often estimated at around three years, could be reduced to less than a year.

 

[WaveJumper]

Will follow your thread with interest.

We were actually having a conversation here yesterday on SSD size due to my son now filming in 61MP and shooting a 8K video, easy to forget you need to up your computing capabilities & storage to keep ahead of the game.

I am no expert in this field and a question that was running through my mind yesterday was is there a "safe" limit when using your SSD's ie fill it up three quarters personally I like to give myself "plenty of room" I back up to cloud and my desk top server ............ just in case Microsoft let me down

 

[Nico Albrecht]

Will follow your thread with interest.

Glad to have you on board! Here's a quick update on Day 2:

Day 2 stream https://www.youtube.com/live/aQVhY0r0gJs?si=jsPCbaC_o9Q7hlbV

On my case study Day 1 data is also available now as download

I added a 1TB WD Gold mechanical drive to the system, and now all drives are writing in sequence with zero values.

- **990 Pro**: performed flawlessly
- **990 EVO**: 55 MB/s
- **870 QVO**: 30 MB/s
- **WD Gold**: 150 MB/s

I tested all drives in different machines this morning and had a chat with the developer of HD Sentinel. This isn’t a system limitation—it shows how SSDs perform under heavy load, where a mechanical drive can sometimes outperform them.

All SSDs are brand new, and I’d argue that the "80% fill" rule shouldn't apply since it’s not mentioned on retail packaging. A typical consumer wouldn’t know that and would simply fill it up with data. With 1TB being relatively small these days, this is a key point.

Right now, I’m clearing the drives by writing zeros to see if that improves performance.

The low performance on the QVO was somewhat expected, dropping from 550 MB/s to around 150 MB/s under load—but we’re currently at just 30 MB/s. The real disappointment is the EVO 990.

As for the color coding system used in the 4 windows on the stream I’ve used:

- **Light green**: the drive responds promptly (e.g., 5ms)
- **Darker green**: a delay is already noticeable (e.g., 50ms)
- **Yellow**: response time exceeds 100ms
- **Red**: bad blocks with slow or no response (e.g., 2 seconds or more)

A good SSD with good firmware would be very smooth and wont show many dark green blocks. This is when trim kicks in, ECC starts to slow doen and bad blocks being brought in from spare area.

From the live stream, we can see how the firmware is optimized for short bursts, but in the long run, it’s clear that performance slows down.


These 3 drives present the current line up of Samsung Best drives consumer to Pro and the same tech is in their external drives e.g Samsung T7 , T9


I will so some other brands in the next couple weeks

 

[Ozzy]

Are you planning to do a press release on your findings?

 

[Nico Albrecht]

safe" limit when using your SSD's ie fill it up three quarters personally

Lately, I’ve noticed many NAS setups using SSDs in RAID configurations experiencing failures. With a single drive, you have some control over how much it's filled, but in a NAS with RAID, snapshots, and other processes, it’s easy to push the drives to 100% fill—even if only 50% of the NAS capacity is actually in use.

As home servers and larger storage solutions become more common, it's worth exploring how SSDs perform in terms of endurance and speed under these conditions.

 

[fisicx]

My old Mac lasted 10 years and was still going strong when it was replaced. So while an SSD under heavy load may fail or degrade most light users will never notice (I didn't).

Does spinning rust have the same problem?

 

[HFE Signs]

I'm interested to see these results, all our PC's are on SSD, average load, couple of years old now and perform extremely well. ScanDisk Ultra 3D SSD

 

[Nico Albrecht]

My old Mac lasted 10 years and was still going strong when it was replaced. So while an SSD under heavy load may fail or degrade most light users will never notice (I didn't).

Your Mac likely came with an SSD from Samsung, Toshiba, or SanDisk, equipped with older NAND technology like SLC, which offers better endurance compared to newer NAND types like TLC or QLC. These newer technologies are designed for cheaper, higher-capacity storage but come with reduced longevity.

If you're a creative professional handling large video projects on your Mac, regularly adding hundreds of gigabytes of 4K footage could significantly shorten the lifespan of your SSD. While casual web users won't notice the impact, I’ve seen failed drives coming in for data recovery within 12-16 months, often with less than 30TB written.

I’m also exploring whether power cycling affects SSD degradation. In data recovery, every power cycle can worsen a failing SSD, so we avoid power cycling to preserve the drive’s condition during recovery attempts.

On my Day 2 live stream with the "rust spinner," you can see:

• Top left: 990 Pro writing random data and blocks
• Top right: 990 EVO writing sequentially with zero blocks (caching should activate, but speeds are very low)
• Bottom left: 870 QVO writing sequentially with zero blocks (caching should activate, but speeds are very low)
• Bottom right: WD 1TB Gold (mechanical) writing random data and blocks at high speeds

 

[Nico Albrecht]

ScanDisk Ultra 3D SSD

Your SanDisk Ultra 3D SSD uses TLC NAND technology, which is more durable than some of the newer alternatives. You likely got yours before Western Digital acquired SanDisk. Since then, I've noticed issues with WD/SanDisk SSDs due to internal conflicts and changes in direction.

For example, some SanDisk models have used up to four different controllers—some native to SanDisk, others hybrid, and some from WD. This inconsistency suggests that their engineers have been divided on how to move forward, leading to a "United Nations" of controllers and memory chips within the same product, which has made these SSDs less reliable.

For now, I wouldn't recommend using SanDisk or WD SSDs until they resolve these internal issues and unify their development efforts.

My Samsung drives are the latest tech you get over the counter for consumers and pro users

 

[Kerwin]

For home / business usage they are fine. If you are running them in servers in a datacentre you really should be using datacentre class SSDs. They are more expensive but worth it.

Same as you should be using ECC RAM rather than standard RAM in a datacentre.

 

[WaveJumper]

Interesting the comments on Western Digital had quite a few problems with them in the past and spent a lot of time getting my data back out the failed unit I have since gone over to LaCie which seem to perform fine just hope Apple not been selling crap

 

[fantheflames]

I've never really thought about this! I've had most of my setup for years and years. Thanks for sharing! Looking forward to more tech being tested out.

 

[Nico Albrecht]

Day 3 Observations:

The Samsung 870 QVO is writing at 80 MB/s in sequential writes. After digging into the specs, I found that Samsung lists this as the expected speed after the cache fills up, though the information is hard to find. We get 225 MB/s for about two minutes, then it drops to 80 MB/s for the rest of the cycle. Meanwhile, the WD Gold 1TB mechanical drive averages 160 MB/s, outperforming the QVO.

The Samsung 990 EVO struggled for two days, fluctuating between 600 MB/s and 80 MB/s. On Day 3, after a reboot, it was marked as a bad drive after writing only 16TB lifetime data (of its 600TB estimated lifespan). After recovery attempts, it wiped all data and reset. It’s now running at 900 MB/s, its best performance yet, but with full data loss along the way.

The 990 Pro has been a consistent performer, holding strong as the fastest and most stable, writing at optimized levels.

Price difference? Minimal. I paid:

• £87 for the Samsung 990 Pro (1TB)
• £76 for the Samsung 990 EVO (1TB)
• £64 for the Samsung 870 QVO (1TB)

The Pro has more advanced hardware with dual NAND chips in RAID 0 and a dedicated cache. Meanwhile, the EVO 990 uses fewer components and relies on system RAM for caching, a step down from the previous generation. We call it SSD shrinkflation—fewer parts, but the same cost.

As for the 870 QVO, after confirming its 80 MB/s limit, it's clear it’s not ideal for high-performance tasks. Despite being advertised at 530 MB/s, it’s only useful for large, long-term storage, but even mechanical drives beat its speed and stability. At 80 MB/s, filling it 360 times to hit its 360TB write endurance would take 180 days—rendering my endurance test pointless.

Day 3 Live stream https://www.youtube.com/watch?v=4m2hiR2_E_U

I will optimize some workflows and processes throughout the test and test other models too in the lower range of pricing in the couple next week.

 

[Nico Albrecht]

For home / business usage they are fine

The 990 EVO failed after just 16TB of writes, taking all the data with it—so not sure what qualifies as "fine." While enterprise gear isn’t an option due to its insane pricing, consumers are increasingly offloading data like photos and videos to NAS and storage devices. But when these fail, it's a nightmare.

Just last week, I worked on a home server with a RAID 10 SSD array that had a total failure after 19 months—all 4 Samsung SSDs failed. The array was used for VMs, home automation, firewalls, and family data.

With the rise of solutions like TrueNAS, more people are building home servers for long-term storage, using older PCs and newer SSDs. It's becoming popular but comes with risks.

 

[Nico Albrecht]

LaCie which seem to perform fine just hope Apple not been selling crap

It's typical of Apple to sell shiny products with subpar components inside. The same goes for LaCie, which is owned by Seagate. With LaCie, you get a sleek metal enclosure, but inside it's just the cheapest Seagate tech. Seagate drives make up about 70% of my business, followed by Apple MacBooks. WD is generally okay, but their newer G-Drive lineup is starting to show issues too, with more cases coming in.

After 10 years in this field, I’ve seen a noticeable decline in the quality of almost every major drive brand, and it’s gotten significantly worse in the last four years. These days, drives failing within just a few months for no apparent reason has become the norm.

 

[Kerwin]

The 990 EVO failed after just 16TB of writes, taking all the data with it—so not sure what qualifies as "fine." While enterprise gear isn’t an option due to its insane pricing, consumers are increasingly offloading data like photos and videos to NAS and storage devices. But when these fail, it's a nightmare.

Just last week, I worked on a home server with a RAID 10 SSD array that had a total failure after 19 months—all 4 Samsung SSDs failed. The array was used for VMs, home automation, firewalls, and family data.

With the rise of solutions like TrueNAS, more people are building home servers for long-term storage, using older PCs and newer SSDs. It's becoming popular but comes with risks.

It is like anything in life. How much do you value your data? If it is just a few odd photos and some documents you don't mind losing then fine go with consumer stuff. If you want to keep your backups for 10+ years then spend a decent amount of money on your gear.

 

[Techonis]

I can see a return back to the good ol' local tape drives and Iomega backup as the cost of cloud backup isn't as cheap when you start looking into the 10's of TB. And global situation doesn't help as the supply chain can be hit and miss and cause cost of disks to rise. Could Blueray and equivalent make a return perhaps, I wonder?????

 

[DontAsk]

It is like anything in life. How much do you value your data? If it is just a few odd photos and some documents you don't mind losing then fine go with consumer stuff. If you want to keep your backups for 10+ years then spend a decent amount of money on your gear.

I've got files dating back 30 years and never had an issue with consumer grade storage. I layered backups and don't expect any hardware to last for that 30 years. Things have moved from floppy to hard drives, CD-ROMs, and now SSD. A bigger problem would be if I could no longer access software to "read" a particular file type but that has never happened. My own incremental backups do not come anywhere close to disk wearout levels of writes.

 

[DontAsk]

https://www.xda-developers.com/ssd-wear-might-be-the-most-overhyped-hardware-problem/

 

[Karimbo]

https://www.xda-developers.com/ssd-wear-might-be-the-most-overhyped-hardware-problem/

I have a crucial M4 128GB SSD which was the primary OS drive for many years. It has about 80% useful capacity left.

I bought it in 2011 from scan computers. It was used as the primary OS until the 128GB capacity could no longer be useable as a primary drive. It is now relegated as a drive just to store a few games on my computer.

I remmeebr the fears very well. How SSDs were garbage and they wouldn't last as long as a mechanical drive.

I dont think there are any mechanical drives that have been use consistently and lasted this long. If it was a 500GB drive it probably would still be used as a priamry drive on a daily use system. A mechanical driver would have been slow and noisy as heck after 4 years and needed to be replaced.

 

[DontAsk]

I remmeebr the fears very well. How SSDs were garbage and they wouldn't last as long as a mechanical drive.

Probably just FUD from the spinning rust vendors.

 

[Nico Albrecht]

Just a quick update as I was busy with loads of other projects and format has changed a bit but here are some updates.

Big shoutout to Samsung! The 990 Pro is an absolute beast in my SSD endurance project—rated for 600TB but has written over 3900TB with 900 power cycles and still going! Unbelievable performance!

Well done Samsung.

YouTube Live stream : https://www.youtube.com/live/hfxq5NrZg60?si=-kIznybUDsbOKxeg

Case study https://data-forensics.co.uk/ultima...s-top-chinese-ssd-brands-which-lasts-longest/


I will document everything over Christmas, but due to time constraints, my current conclusion is that using any SSD is like playing the lottery. You might get a great unit that lasts well beyond its expected lifespan, or you could end up with one that fails prematurely. The matter is still under investigation, with testing ongoing and more projects planned to verify these findings.

 

[Deleted member 360015]

A study conducted by Google over millions of drive days found that SSDs have a lower replacement rate compared to HDDs. However, SSDs exhibit a higher rate of uncorrectable errors, which means that while they are less likely to fail completely, they are more prone to data loss

Real-world data suggests that while SSDs have lower overall failure rates, they require careful management and regular backups to mitigate the risk of data loss.

The operating environment of an SSD can significantly impact its durability:

Excessive heat can damage the NAND-based flash memory in SSDs, leading to faster wear and tear.

Sudden power losses can lead to data corruption and affect the SSD's lifespan.

Typically, HDDs last around 3-5 years, while many SSDs can last 5-7 years, with high-end models potentially reaching up to a decade. This extended lifespan is attributed to their non-mechanical design and advanced wear-leveling technologies.

Users often experience system crashes, read/write errors, and slower performance as signs of impending SSD failure. Unlike HDDs, SSDs do not produce noise when failing, making it harder to detect issues early.

 

[Nico Albrecht]

A study conducted by Google over millions of drive days found that SSDs have a lower replacement rate compared to HDDs. However, SSDs exhibit a higher rate of uncorrectable errors, which means that while they are less likely to fail completely, they are more prone to data loss

The study is nearly a decade old and is now completely outdated. It seems like you're quoting it without having fully read the original Google study. Over the past 10 years, HDD and SSD technologies have evolved significantly, with SSDs also experiencing a decline in quality. The primary focus for both has shifted toward maximizing capacity at the lowest possible cost.

Real-world data suggests that while SSDs have lower overall failure rates, they require careful management and regular backups to mitigate the risk of data loss.

Current real-world data actually suggests the opposite. Backblaze, one of the closest sources for reliable stats, estimates the average lifespan of their SSDs at under 2.7 years. In contrast, HDDs tend to last much longer. Notably, Backblaze only uses SSDs for operating system drives, while HDDs handle the main workloads. This means the 2.7-year estimate for SSDs could easily be halved under heavier use, aligning with my observations of SSDs failing within 14–16 months under medium use.

Excessive heat can damage the NAND-based flash memory in SSDs, leading to faster wear and tear.

That's incorrect. Modern NAND technology is quite tolerant of temperature extremes, with SSDs functioning well even at 70°C. The elevated temperatures seen in modern SSDs aren't a direct issue for the NAND itself but rather stem from the SSD's controller (essentially its CPU) working harder. This increased workload is due to tasks like XOR and ECC corrections, which are necessary to compensate for the lower-quality NAND cells being used in many modern drives. The higher heat is, therefore, a side effect of the controller compensating for subpar NAND performance.

Sudden power losses can lead to data corruption and affect the SSD's lifespan.

That's no longer a concern with modern file systems and caching mechanisms. Issues that might have been significant 20 years ago are now well-handled by advancements in file system design and caching technologies. These improvements ensure better reliability, data integrity, and performance, making such problems largely a thing of the past.Typically, HDDs last around 3-5 years, while many SSDs can last 5-7 years, with high-end models potentially reaching up to a decade. This extended lifespan is attributed to their non-mechanical design and advanced wear-leveling technologies.
Modern statistics from Backblaze show that even under light use as OS drives, SSDs have an average failure rate of under 2.7 years. This demonstrates that, despite their performance advantages, SSDs still face significant longevity challenges compared to HDDs, even when used in less demanding roles.

Users often experience system crashes, read/write errors, and slower performance as signs of impending SSD failure. Unlike HDDs, SSDs do not produce noise when failing, making it harder to detect issues early.

I would argue this claim only applies to HDDs. When SSDs fail, they often do so suddenly, with little to no warning. This behavior stems from the reliance on ECC (Error Correction Code), which continues to compensate for deteriorating NAND cells until the drive reaches a critical point of failure. Once the ECC can no longer correct errors, the SSD typically fails outright, leaving little opportunity for data recovery or pre-failure warnings.

 

[Karimbo]

Nice, I think you are debating with an bot who posted AI content. The account has now been deleted it seems. The username suggests they were just here to promote a website.

To me it reads like a wikipedia writing, neutral tone, no emotion of feeling, like a wikipedia entry. The spelling, punctuation and grammer is basically perfect. Just missing a full stop at the end of the first paragraph.