What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
An SSD that benchmarks below its advertised IOPS is not automatically defective. IOPS is a workload-specific result, not a single speed rating: a vendor’s figure may describe 4 KiB random reads at a particular queue depth, while your test uses larger blocks, writes, a filesystem, a cache, or too little parallel work. Diagnose the difference by reproducing the published conditions, confirming that requests really reach the drive, and comparing IOPS with latency and throughput.
What an advertised IOPS number actually describes
IOPS means input/output operations per second. The number is meaningful only alongside the conditions used to produce it. A 4 KiB random-read result cannot be compared directly with a sequential-transfer result measured in megabytes per second, or with a mixed read/write test. The same SSD can produce very different figures as block size, access pattern, read/write mix, queueing and test path change.
As an Amazon Associate I earn from qualifying purchases.
| Condition | Why it changes the result | What to record |
|---|---|---|
| I/O size | Small requests allow many operations to be counted as IOPS; large requests move more data per operation and are usually better described by throughput. | Block size for reads and writes, such as 4 KiB or 128 KiB. |
| Access pattern | Random access exercises different flash-management behavior from sequential access. | Random or sequential, and whether the test uses a fixed or variable location. |
| Read/write mix | Reads, writes and mixed workloads have different controller, flash and cache costs. | Read-only, write-only, or the exact read/write percentage. |
| Outstanding requests | More effective parallelism can keep the controller busy, but can also increase latency. | Queue depth, worker count and the number of requests actually in flight. |
| Benchmark engine | A synchronous engine may submit one operation at a time even when a larger queue depth is specified. | The fio ioengine and whether it supports asynchronous submission. |
| Data path | Page cache, a filesystem, a virtual disk, a cloud volume or an application can add limits that are not properties of the NAND device. | Raw device versus filesystem, direct I/O setting, virtualization and volume type. |
| Duration and state | A short, empty-drive run can benefit from cache and fresh-drive behavior; a preconditioned run may represent sustained operation more closely. | Warm-up, test length, preconditioning and whether other I/O was running. |
Manufacturers often publish a peak value under an ideal combination of these variables. Treat it as a conditional capability, not a promise that every application will achieve the same number.
The main reasons a benchmark falls short
Your workload does not match the specification
The first and most common explanation is a mismatch in block size or access pattern. If a specification says 4 KiB random read, a 1 MiB sequential-read test is measuring a different capability. A mixed workload can also be lower than a read-only headline figure because writes consume controller and flash resources. Check whether the published figure is for reads or writes, random or sequential access, and a particular queue or worker configuration.
#1 Best Overall
- MEET THE NEXT GEN: Consider this a cheat code; Our Samsung 990 PRO Gen4 SSD helps you reach near max performance with lightning-fast speeds; Whether you’re a hardcore gamer or a tech guru, you’ll get power efficiency built for the final boss
- REACH THE NEXT LEVEL: Gen4 steps up with faster transfer speeds and high-performance bandwidth; With a more than 55% improvement in random performance compared to 980 PRO, it’s here for heavy computing and faster loading
- THE FASTEST SSD FROM THE WORLD'S FLASH MEMORY BRAND: The speed you need for any occasion; With read and write speeds up to 7450/6900 MB/s you’ll reach near max performance of PCIe 4.0 powering through for any use
- PLAY WITHOUT LIMITS: Give yourself some space with storage capacities from 1TB to 4TB; Sync all your saves and reign supreme in gaming, video editing, data analysis and more
- IT’S A POWER MOVE: Save the power for your performance; Get power efficiency all while experiencing up to 50% improved performance per watt over the 980 PRO; It makes every move more effective with less consumption
fio exposes the request size as a configurable parameter. Its bs setting controls the size of each I/O unit, with separate read, write and trim values available. Set these explicitly rather than relying on a benchmark tool’s defaults. Also set the read/write mix, access pattern and duration so a second run can be reproduced exactly.
The configured queue depth is not real concurrency
A queue-depth field in a job file does not prove that the SSD received that many simultaneous requests. NVIDIA’s storage-performance guidance demonstrates the problem: with a synchronous fio engine, increasing iodepth does not turn the workload into an asynchronous one. The job can still wait for each operation to complete before issuing the next.
Use an engine appropriate for the operating system and workload, then observe actual queue activity while the test runs. If the device never has multiple requests outstanding, a high advertised-IOPS figure that depends on parallelism is unreachable regardless of the number written in the configuration.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →There is not enough parallel work to drive the device
Even with an asynchronous engine, one worker or a lightly loaded application may not generate enough in-flight I/O. High IOPS generally requires sufficient concurrency to keep the controller busy. Increase workers or outstanding requests gradually, and watch latency at each step. The goal is not the largest queue number; it is enough useful parallelism for the intended workload without creating unacceptable response times.
Rank #2
- Ideal for high speed, low power storage
- Gen 4x4 NVMe PCle performance
- Up to 6,000MB/s read, 4,000MB/s write
- Includes Acronis cloning software
- 5-year limited warranty
Cache or the surrounding path is being measured
Reads may be served from the operating-system page cache or from a virtualized volume cache instead of reaching the SSD. A filesystem adds its own allocation and metadata behavior. In a virtual machine or cloud service, the path can include a host, instance, volume and network layer, each with separate limits. Microsoft’s disk-benchmarking guidance notes that cache settings affect whether reads reach the underlying disk. NVIDIA likewise calls out page-cache effects and demonstrates direct-I/O configuration when the objective is to test storage rather than memory.
Decide what you are trying to measure. A raw block-device test answers a device-path question; a filesystem test answers an application-path question. Neither is wrong, but they should not be compared with a vendor result produced under different conditions.
More concurrency raises IOPS but hurts latency
IOPS, throughput and latency describe different aspects of the same run. Adding outstanding requests can increase completed operations per second until the device or host saturates, while individual requests wait longer. AWS’s EBS documentation advises choosing queue length for the workload rather than treating one queue depth as universally optimal. A database, interactive desktop and batch transfer can therefore have different useful settings even on the same volume.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe test does not represent sustained behavior
Fresh media, volatile write cache and a short run can produce a peak that is not maintained. NVIDIA’s guidance recommends preconditioning a new SSD for stable measurements, including sequential and random write workloads, and matching warm-up and test duration when comparing runs. A result taken after unrelated background I/O, immediately after a burst, or before the drive reaches a steady state may not be comparable with the published number.
Rank #3
- SPEED UP PROJECTS. Launch creator applications fast with uncompromising PCIe 4.0 read speeds up to 7,100MB/s,[2] (1TB and 2TB[1] models) and write speeds up to 6,700MB/s[2] (1TB[1]-4TB[1] models).
- CREATE AND STORE MORE. Make more room for your 4K videos and high-resolution images with capacities from 500GB[1] up to 4TB[1] on M.2 2280 built with our trusted 8th generation SANDISK BiCS QLC 3D CBA NAND.
- IT GOES WHERE YOU GO. With an all-new power efficient design, your drive delivers high performance with low power, giving you more time to be productive while on the go.
- UNCOMPROMISED RELIABILITY. With up to 1,200 TBW[3] (4TB[1] model) endurance rating, your drive is designed for creators.
- KEEP YOUR DRIVE UPDATED. Monitor your SSD’s performance and check for updates with the downloadable SANDISK Dashboard application.[5]
A diagnostic procedure that separates the variables
1. Capture the exact claim and identify the drive
- Record the SSD’s exact model, capacity and interface. Do not compare a generic benchmark result with a product page whose model or capacity is different.
- Copy the manufacturer’s stated IOPS conditions: read or write, random or sequential, block size, queue depth or worker count, peak versus sustained wording, and any test-platform notes.
- Record your operating system, filesystem, benchmark version, fio job file or command, firmware, link mode, temperature and drive-health indicators. These details are necessary if the result later needs technical support or a warranty assessment.
If the published conditions are not available, label the comparison as indeterminate rather than treating the headline number as a universal target.
2. Make every workload setting explicit
In fio, set at least the block size, read/write mode, random or sequential pattern, runtime, warm-up and queueing. A minimal illustrative template for a Linux random-read comparison is:
fio --name=randread-check --filename=/path/to/testfile --rw=randread --bs=4k --iodepth=32 --numjobs=1 --runtime=60 --time_based --direct=1This is a template, not a guarantee of a particular result. Replace the path and values with the manufacturer’s conditions, ensure the selected engine is appropriate for your system, and use a test file only when a filesystem-path measurement is intended. A raw-device test can destroy data, so never point a destructive job at a disk containing files you need.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Run separate jobs when the specification distinguishes read and write values. Keep the I/O size consistent; changing from 4 KiB to a larger block can reduce the operation count while increasing bytes transferred, making IOPS and throughput move in opposite directions.
Rank #4
- HUGE SPEED BOOST: Get random read/write speeds that are 40%/55% faster than 980 PRO; Experience up to 1400K/1550K IOPS, while sequential read/write speeds up to 7,450/6,900 MB/s reach near the max performance of PCIe 4.0*
- BREAKTHROUGH POWER EFFICIENCY: Use less power and get more performance; Enjoy up to 50% improved performance per watt over 980 PRO, plus optimal power efficiency with max PCIe 4.0 performance**
- SMART THERMAL CONTROL: Samsung's own nickel-coated controller delivers effective thermal control; With its slim size, 990 PRO is a perfect fit for desktops and laptops that meet the PCI-SIG D8 standard***
- THE CHAMPION MAKER: Up to 65% improvement in random performance enables faster loads for an ultimate gaming experience on PS5 and DirectStorage PC games****
- SAMSUNG MAGICIAN SOFTWARE: Get the most out of your SSD with Samsung Magician's advanced yet intuitive optimization tools; Monitor drive health, protect valuable data, and receive important updates for your 990 PRO
3. Confirm that fio can create the requested concurrency
- Inspect the job’s
ioengine. A synchronous engine can ignore the practical effect of a largeriodepth; choose an asynchronous engine supported by your operating system and storage path when the test requires multiple in-flight requests. - Do not infer device concurrency from the configuration alone. Monitor the device queue and utilization during the run using the operating-system tools appropriate to your platform.
- Increase
iodepthor worker count in small steps and record IOPS, bandwidth and latency percentiles at each step. Stop increasing when IOPS stops improving or latency exceeds the level your application can tolerate.
NVIDIA’s worked example shows why this check matters: a particular 4 KiB random-read setup produced about 25,000 IOPS, far below the roughly 92,000 random 4 KiB read figure cited for an 850 EVO-era SATA SSD in the same 2018 discussion. The article uses the gap to investigate engine behavior and workload sufficiency; it does not establish a universal result for that drive or for current SSDs.
4. Remove cache ambiguity
- Choose direct I/O when the goal is to bypass the operating-system page cache, and state that choice in the report.
- If you are testing an application, also run a filesystem-path test without bypassing cache. That result represents the application’s experience and should be judged against application requirements, not a raw-device headline.
- On a virtual machine or cloud volume, document volume cache mode, instance type, provisioned limits and whether the benchmark runs on a local or network-attached device.
- Use a test size large enough that the result cannot be explained simply by a small cache. Repeat the run after the cache state is known or controlled.
5. Check host, interface and service limits
A drive can be capable of more than the link or host allows. Verify that the expected interface and link width are negotiated, that the controller is not sharing bandwidth with another workload, and that virtualization or cloud limits are not lower than the SSD’s specification. For cloud storage, compare both the volume limit and the virtual-machine or instance limit; the lower ceiling controls the observed result.
6. Control temperature, background activity and drive state
- Let the system reach a repeatable temperature before measuring, and record temperature during the run. Thermal management can change sustained behavior even when the peak specification is unchanged.
- Stop indexing, antivirus scans, synchronization and other jobs that generate storage traffic. Record anything that cannot be stopped.
- For sustained comparisons, follow a consistent preconditioning and warm-up sequence. NVIDIA recommends sequential and random write preconditioning for stable SSD measurements.
- Run the identical job multiple times. Save the full fio output, including bandwidth, IOPS, average and percentile latency, errors and runtime.
How to interpret the results
| Observed pattern | Likely explanation | Next check |
|---|---|---|
| Low IOPS and low device utilization | The workload is not supplying enough parallel I/O, or a synchronous engine is limiting concurrency. | Verify ioengine, workers, actual queue activity and application submission rate. |
| IOPS rises with queue depth, then levels off while latency climbs | The device has reached a useful saturation point; additional queueing adds waiting time rather than productive work. | Choose the queue that meets the application’s latency target, not the largest IOPS number. |
| Very high read speed on a small test file, poor speed on a larger one | Cache may be serving the small test. | Use direct I/O or a controlled cache state and repeat with a larger test set. |
| Raw-device result is high but filesystem or application result is low | The filesystem, software path or application is the bottleneck. | Profile that path separately; do not call the SSD faulty from the raw result alone. |
| Cloud volume stops at the same ceiling across different queue depths | A volume, instance or host limit may be controlling performance. | Check platform-specific IOPS and throughput limits for both the volume and compute instance. |
| Performance declines during a long run | The initial result may have used cache or empty-drive behavior; sustained state, temperature or background activity is different. | Record temperature, precondition consistently and compare steady-state segments. |
| Errors, timeouts or link warnings appear | The problem may be the connection, controller, firmware or system rather than flash performance. | Collect health data, firmware and interface details before further destructive testing. |
Always compare latency alongside IOPS. A benchmark that wins on IOPS by creating excessive queueing may be worse for an interactive workload. Report throughput too, especially when changing block size: larger requests can deliver more bytes per second even though the IOPS count is lower.
When the evidence points to a drive problem
A low number by itself is not proof of failure. A device-specific conclusion requires the exact model and published conditions, a controlled repeatable test, and evidence that cache, concurrency, path and host limits have been accounted for. Repeat the matched test on a known-good system or equivalent connection when practical, then compare the result with the same model and capacity under the same conditions.
Best Value
- This product has been replaced by our latest generation. Please search for the SANDISK Optimus GX 7100 NVMe SSD
- HIGH-OCTANE GAMING. Experience speeds up to 7,250MB/s read and 6,900MB/s write (1-2TB models), with up to 35% faster performance than previous generation.
- PURPOSE-BUILT. Designed for serious on-the-go gamers, with a PCIe Gen4 interface and SANDISK’s next generation TLC 3D NAND.
- MORE TIME TO CLEAR THAT CHECKPOINT. Built with laptops and handheld gaming devices in mind, with up to 100% more power efficiency over the previous generation.
- DO MORE WITH DASHBOARD. Ensure your drive is optimized for prime performance with the downloadable WD_BLACK Dashboard (Windows only).
Escalate to the manufacturer or platform operator with:
- Exact model, capacity, firmware and interface/link information.
- Operating system, driver or controller details, filesystem and whether the test used a raw device or file.
- Complete fio command or job file, benchmark version and test duration.
- IOPS, throughput, average and percentile latency, queue settings and observed utilization.
- Temperature, health data, error logs and whether preconditioning was performed.
- Information about other workloads, virtualization, cloud volume and instance limits.
Only after those checks show a repeatable shortfall under matching conditions should replacement or warranty action be considered. The available evidence cannot determine a fault for an unidentified SSD without those system details.
A practical comparison checklist
Before declaring that one benchmark is slower than another, put both results side by side and answer every item:
Recommended Free Tools
- Are the I/O sizes identical?
- Are both tests random or both sequential?
- Is the read/write ratio the same?
- Do worker count and outstanding I/O match?
- Does the benchmark engine actually submit asynchronously where required?
- Is cache bypassed or enabled in both tests?
- Are both measurements on a raw device, or both on the same filesystem and application path?
- Are runtime, warm-up and preconditioning comparable?
- Are latency and throughput reported as well as IOPS?
- Do host, interface, virtual-machine and volume limits permit the advertised result?
If any answer is unknown, the comparison is incomplete. Resolve that variable or describe the result as a measurement of your current path rather than a verdict on the SSD’s hardware.
What the historical 25,000-IOPS example teaches
NVIDIA Developer Blog’s 2018 “Storage Performance Basics for Deep Learning” used an 850 EVO-era SATA SSD to illustrate a discrepancy: roughly 25,000 random-read IOPS at 4 KiB in one setup versus about 92,000 random 4 KiB read IOPS stated for the drive in that article. The figures came from a particular Linux and fio configuration and are historical examples, not current market benchmarks.
The lesson is methodological. Before blaming the SSD, verify that the workload is sufficient to maximize the target device’s performance, that the fio engine honors the intended queueing, and that cache and the storage route are understood. No population-wide or contemporary market statistic follows from that single example.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
