Some users of NVIDIA's GeForce Game Ready drivers, specifically version 621.14 (released November 4, 2026) and version 621.24 (released November 12), have reported that a particular log file keeps being written to even when no game is running. In measurements posted to NVIDIA's official forum on November 9, deleting the log file and taking related steps sharply reduced the write volume of the associated process. However, the problem does not reproduce on some PCs running the same driver versions, and the affected environments have not been identified. In addition, the amount of data a process requests to write differs from the amount the SSD actually receives, so evaluating the load on storage requires distinguishing between the two.
515MB written in 5 minutes, down about 91% after deleting the log
The user who examined the abnormal writes in 621.14 in detail is Andergraude on the NVIDIA forum. Using a GeForce RTX 4070 SUPER and Windows 11 Home 25H2, they measured a desktop environment with just two browser tabs open. They found that NVDisplay.Container.exe was continuously writing to the following log file.
C:\ProgramData\NVIDIA Corporation\nvtopps\nct\nvlog.nvlgstg
Windows Resource Monitor showed a write rate of 4.6 to 5.2MB/s. In a five-minute measurement, the process's write I/O was 515MB, and total writes to the C: drive were 270MB, according to the report.
A process's write I/O is a tally of the write requests a program has issued to Windows. Total writes to the drive, on the other hand, also reflect the activity of other programs and Windows caching, so the two figures do not necessarily match.
Andergraude stopped the Windows service "NVIDIA Display Container LS," deleted the log file, restarted the service, and then examined how the write volume changed.
| Measurement | Before | After | Reduction |
|---|---|---|---|
| NVDisplay.Container.exe write I/O | 515MB | 48MB | ~91% |
| Total writes to C: drive | 270MB | 145MB | ~46% |
Note: A short comparison on the same PC. Reductions are the difference between the two measurements divided by the pre-fix value, rounded to the nearest whole percent. The fix consisted of stopping the service, deleting the log, and restarting the service.
In these results, the related process's write volume fell by about 91%, while total writes to the C: drive fell by only about 46%. The 125MB drop in C: drive writes may also include fluctuations in writes from other programs running during the same period. Andergraude says the SSD's SMART data matched the Windows measurements, but only the results they posted can be verified.
Comparisons with older versions also showed differences. According to Andergraude, on the same system with a clean install of 621.01, the process's write I/O over five minutes was under 2MB. The log file also grew at about 0.5KB per minute on 621.01, compared with about 325KB per minute on 621.14. On 621.14, even after clearing the file's contents, it grew to 2.6MB in about 8 minutes and later expanded to about 20MB. Andergraude reports that write volume rises as the log file gets larger.
It is worth noting that a log file's size is not the same as cumulative write volume. Even if a file is only about 20MB, repeatedly writing to the same area can produce a large cumulative total. chrisaverage, who uses 621.24, also reported that with a single browser tab open and no game running, the roughly 20MB log file temporarily dropped to 0 bytes. This behavior suggests the log's contents may be cleared and then written again.
Heavy writes reported on 621.24 too, while some PCs don't reproduce the problem
Users are also reporting the same problem with the later 621.24. On November 14, AyzekUorren reported writes of about 30GB per hour even with telemetry disabled, reaching more than 70GB in long sessions. However, it is unclear which metric was used to measure the write volume, or whether the figure covers only idle periods.
On November 19, mertuygr posted measurements from a system with a GeForce RTX 4090, Windows 11, and driver 621.24. At idle, about 43 hours after the PC was started, the process's write I/O over 60 seconds reached 2.0 to 2.7GB. This is a per-process measurement and does not show that the full amount was actually written to the SSD.
On the other hand, there are also reports that the problem did not reproduce on either 621.14 or 621.24. Dankbomber1990, who uses a GeForce RTX 5070 Ti, tried 621.24 at idle as well as with the overlay, YouTube playback in Microsoft Edge, and Alt+Tab window switching. However, the process did not appear in Resource Monitor's write list, and the log file grew only from 19KB to 26KB over 16 minutes. Their test results show that the driver version alone does not cause the problem.
Users are also investigating the source of the writes. mertuygr examined the logs output by NVIDIA's container process and identified a plugin called _nvtopps.dll (NvTopps_Plugin). They report that stopping this plugin from loading on their own PC ended the continuous writes. However, the conditions under which the plugin causes heavy writes, and the affected GPUs, Windows configurations, and combinations of related software, have not been identified.
How to check how much is actually written to the SSD
The key to evaluating this problem is to distinguish between the process write I/O observed in Windows and the amount of data the SSD actually receives.
Windows has a file caching mechanism that holds write requests from applications in memory and can apply them to storage later. As a result, there can be a time lag between when a program requests a write and when the data is sent to the SSD. Microsoft also explains this mechanism in its technical documentation on file caching. The posted figures alone cannot separate the effects of individual operations to explain why, in Andergraude's measurements, process write I/O was 515MB while total C: drive writes were 270MB.
To check write volume on the SSD side, the "Data Units Written" value in an NVMe SSD's SMART information is a useful reference. This value records the total amount of logical data sent from the host to the SSD controller, excluding metadata. The NVMe specification uses 512 bytes as the base unit and reports the value in thousands of units, rounded up. So multiplying the difference in the raw counter values before and after a measurement by 512,000 bytes gives an approximate write volume. If your tool already converts the value to GB or TB, you can simply compare the displayed values. Details are in Microsoft's NVMe documentation.
For example, if 270MB were written to the C: drive over five minutes and the same pace continued for an hour, the total would be 3.24GB. However, this is a simple extrapolation of the C: drive's overall measurement, not writes by NVIDIA's process alone. SMART host writes likewise show the total data sent to the SSD, so they cannot directly reveal how much each individual process wrote.
Furthermore, SMART host writes differ from the amount of data actually written to the SSD's internal NAND flash memory. An SSD controller may perform additional internal writes as part of managing and rewriting data. To evaluate an SSD's remaining lifespan, you need to check not only host writes but also model-specific endurance indicators.
The total bytes written (TBW) rating published by SSD makers is one indicator for judging warranty terms. For example, the Samsung 990 PRO 1TB model has a limited warranty lasting five years from purchase or until 600TB of total writes is reached, whichever comes first. Details can be confirmed in Samsung's SSD warranty terms. To estimate how much this problem affects an individual SSD, you need to measure the actual increase in host writes and how long that state continued.
What to check before acting, and rolling back to an older version
To find out whether the problem is occurring on your own PC, one approach is to check disk activity in Windows Resource Monitor first and see whether NVDisplay.Container.exe is continuously writing to the log file in question. At the same time, if you record the host write volume from your NVMe SSD's SMART data at the start and end of the measurement, you can check for changes on both the process side and the SSD side.
To trace file write operations in more detail, you can also use Microsoft's Sysinternals Process Monitor. If you record the measurement duration, the applications that were running, and your GPU and Windows versions before changing any settings, it becomes easier to compare before and after under the same conditions.
There are also reports that rolling the driver back to an older version improved the situation in affected environments. In the official feedback thread for 621.24, mad medic581 reports that the writes stopped after going back to 621.01. However, when rolling back, you need to consider that bugs fixed in later versions may return.
Indeed, in section 3.1.2 of the 621.14 release notes (PDF), NVIDIA explains that it fixed an issue in which, after updating to 621.01, video playback stopped or the screen went black on some GeForce RTX 50 series cards and RTX Spark. Even in an environment where going back to 621.01 would improve this write problem, you could be affected by such other bugs.
Meanwhile, the main fixes listed in the 621.24 release notes (PDF) are an issue with OptiX Denoiser behavior in Chaos Vantage and an issue with display colors when playing DJI D-Log H.265 video in Windows applications. The known issues section is listed as "N/A," and none of the public materials checked as of November 19 say the continuous log writes have been fixed.
NVIDIA drivers have had continuous writes by the same process cause problems in the past. The 471.96 release notes (PDF) from 2021 record a fix for an issue in which writes to nvtopps.db3 continued. However, the file reported this time is a different one, nvlog.nvlgstg.
If NVIDIA releases a fixed version, you can confirm its effect by repeating the same measurements under the same conditions on the PC where the problem occurred, comparing the write volume to the target log and the SSD's host write volume.
