Cursor AI slow on Windows – fixed with cache & SSD tweak

:waving_hand: Hi everyone,

I had serious performance issues with Cursor AI on Windows — frequent slowdowns, high disk usage, and dozens of Cursor.exe processes.

After several tests, I found a simple but very effective solution that completely fixed the issue for me.


:triangular_flag: What I was experiencing:

  • Cursor got slower over time or completely froze
  • 25–30 Cursor.exe subprocesses running
  • Internal SSD (C:) was constantly saturated
  • Setting CURSOR_MAX_WORKERS had no effect

:white_check_mark: My solution:

1. Moved Cursor’s cache to an external SSD (F:)

  • Moved the folder:
    C:\Users\<username>\AppData\Roaming\CursorF:\Cursor_Cache
  • Created a symbolic link:
mklink /D "C:\Users\<username>\AppData\Roaming\Cursor" "F:\Cursor_Cache"
2 Likes

This would mostly work if the drive C is not a SSD.

I imagine there could be an advantage of utilizing two SSDs, one for system and cursor app and the other for cache only. I have nvme pcie gen4 ssd 6000MB/s and from time to time it freezes for me too. Over the course of two months, I experienced the freeze maybe 5 times total and it did not take more than 3 seconds. Just in case anybody is wondering if upgrading the ssd would help.

1 Like

Thanks for your input!

You’re right — this workaround is especially helpful when drive C is a mechanical HDD, since those are much slower. But even when C is a SATA SSD (which is my case), the improvement was still significant.

Cursor generates heavy disk activity (many subprocesses, cache writes, etc.), so moving its workload to a separate SSD (F:) helped eliminate freezes — even on a system with SSDs only.

So I’d say: this solution works well on HDD setups, but can also greatly benefit SSD systems that are under heavy load or not top-tier in speed.

:package: Full technical write-up is available here if anyone’s interested:
:backhand_index_pointing_right: https://github.com/tarchi1955/cursor-ai-performance-fix

Appreciate the discussion!

1 Like

Thanks for sharing your setup — that’s really helpful!

Your Gen4 NVMe SSD is super fast, so it’s interesting to hear that occasional freezing still happens even at that level. It confirms that Cursor’s performance issues aren’t just about raw speed, but maybe also how it handles parallel writes, subprocesses, or cache activity.

I agree: separating workloads across multiple drives can be a good strategy. In my case, I moved both the project and Cursor’s cache to a second SATA SSD, and it made a huge difference — even though it’s slower than your NVMe.

If anyone’s curious about the exact setup I used (including symlinks and folder structure), I’ve documented the full solution here:
:backhand_index_pointing_right: https://github.com/tarchi1955/cursor-ai-performance-fix

Appreciate your input! It’s great to see others experimenting and confirming the results.