A Windows tool that stress-tests drive write performance over time, revealing SLC cache size, cache exhaustion speed, and sustained post-cache write speed.
Fill the Pane writes sequentially to a drive until it reaches a target fill percentage, sampling write speed every 500 ms and plotting the full time-series graph live. This shows you three things CrystalDiskMark was not designed to show:
- Where the SLC cache exhausts — the GB or % capacity at which speed drops
- How far it drops — some drives fall 80%+ from peak to sustained speed
- What sustained speed looks like — the speed a drive maintains after the cache is gone
CrystalDiskMark is a great free benchmark and the standard tool for quick drive comparisons. FTP is not a replacement — it answers different questions.
FTP uses the same SEQ1M Q8T1 I/O pattern (1 MB sequential writes, queue depth 8, single thread) and the same direct unbuffered Windows I/O flags. However, because the internal implementation of any benchmark is not fully public, and because factors like test duration, file size, measurement timing, and driver interaction can all affect results, the sequential write speed reported by FTP may differ from CrystalDiskMark. Neither number is wrong — they reflect the conditions of their respective tests. For head-to-head drive comparisons, use the same tool consistently.
- SEQ1M Q8T1 — 1 MB sequential writes, queue depth 8, single thread
- Native C++ write loop (
ftp_loop.dll) for maximum throughput — falls back to Python if DLL not present - Pre-warm phase conditions the NVMe write pipeline before measurement starts
- Graph shows previously used drive space as a shaded block so results are in full drive context
- Configurable write target — slider from 0–100% of free space, or manual GB entry
- Live write speed graph updated every 500 ms
- Interactive crosshair tooltip — GB written, % capacity, elapsed time, speed at cursor
- Max, average, and min MB/s annotations on graph with average dashed line
- Max, average, min MB/s and duration in Results panel
- Dark and light theme
- Multi-language support (English, German, French, Spanish, Japanese, Simplified Chinese, Korean)
- Stop test instantly at any time
- Export graph as PNG or JPG image
- Export animated GIF of test progression
- Export samples as CSV
- Load previously saved CSV for review
- Infographic for Web — embeddable interactive HTML chart
- Notes field on graph panel
- Debug log toggle (off by default)
- Random data writes to prevent hardware compression from skewing results
Python 3.13 or later.
pip install matplotlib psutil pillow pyinstaller
The C++ DLL provides the highest write throughput. Run build_dll.bat from the project root — it detects MSVC or MinGW automatically and installs MinGW via winget if neither is present.
build_dll.bat
This produces ftp_loop.dll in the project root. The app works without it (falls back to the Python loop) but runs faster with it.
The build script reads the source file directly from latest\. Place the current fill_the_pane_vX.X.XX.py there before building. build.bat will copy it to code\ for history and clean up any older versions from latest\ automatically.
pyinstaller fill_the_pane.spec
Or double-click build.bat — it verifies the source is in latest\, copies it to code\, builds the exe, and launches it automatically.
The finished executable will be in dist\. Copy it anywhere — no installation needed.
python latest\fill_the_pane_v*.py
- Writes a temporary
_FtP_*.tmpfile to the target drive and deletes it when done or on stop. - The app requests Administrator privileges automatically at launch via UAC prompt — this is required for direct unbuffered I/O on some drives.
- Random bytes are used to prevent hardware compression from skewing results.
- Never writes more than the configured target percentage of total drive capacity.
- The SLC cache cliff (if present) appears as a sudden speed drop partway through the test — this is expected drive behavior, not a bug.