**[Beta] DiskTruth — F_NOCACHE Mac disk benchmark with SLC cliff detection, negotiated interface speed, and cryptographically signed leaderboard**
Hello — I've been building a disk benchmark for Apple Silicon Macs and I'm opening up the TestFlight beta. I wanted to post here because this community tends to care about methodology, and there's a real methodology story behind this one.
---
**The problem with existing benchmarks**
BlackMagic Disk Speed Test is the standard, but it doesn't use `F_NOCACHE`. macOS maintains a unified buffer cache in RAM, and without explicitly bypassing it, a significant portion of reads on a high-memory Mac are served from memory rather than the drive. Write results can also be affected because the OS batches writes through its page cache. The result is that benchmark numbers on a 64 GB Mac look noticeably better than the same drive on a 16 GB Mac — not because the drive is faster, but because more of the test data fits in RAM.
AmorphousDiskMark is more capable than BlackMagic but also doesn't use `F_NOCACHE` and hasn't been updated since 2022.
---
**DiskTruth's approach**
`fcntl(fd, F_NOCACHE, 1)` is set on every file descriptor before any I/O. Write buffers are populated with `arc4random_buf` — random data prevents hardware compression from inflating write throughput on drives that implement it. Standard mode uses a fixed 16 GB test file across 3 iterations, reporting the median.
---
**What it measures**
*Sequential read/write (MB/s)* — per-chunk throughput recorded for every 64 MB segment during write. Live graph updates in real time.
*SLC cliff detection* — when per-chunk throughput drops more than 20% sustained, a cliff event is recorded and marked on the graph. The reported write speed is the honest median across all three full iterations, including post-cliff performance. This is the number that represents what the drive actually does under sustained load — not just the fast cached phase.
*Random 4K IOPS + latency* — 10,000 operations per pass at QD1. Per-op latency tracked; average and p99 reported. p99 is the number that reveals controller consistency.
*Queue depth scaling* — QD1, QD4, QD16, QD32 (3,000 ops per level). Apple Silicon internal NVMe has a notably different QD scaling profile vs desktop NVMe — extraordinary QD1 latency, less dramatic scaling at higher depths. Worth seeing on your own hardware.
*Negotiated interface speed* — reads from IORegistry, not from the port label. Shows actual Thunderbolt or USB4 link speed for your specific port + cable + enclosure combination. I've found this surprisingly useful — a TB4 enclosure with a third-party cable will sometimes negotiate at USB 3.2 speeds without any obvious indication.
*Speed Doctor* — a layered diagnostic (Cmd+Shift+D) that walks through: interface negotiation → thermal state → system background load → firmware check → drive eject for cable reseat.
---
**Leaderboard and result integrity**
Standard mode results can be submitted to a community leaderboard. The result payload is P256-signed at benchmark time — the signature covers numbers, drive info, Mac model, test configuration, and integrity fields. The server verifies the signature, checks Isolation Mode was active, deduplicates by device+drive, and scores result integrity. Results that pass get a Verified badge. The signed result card PNG includes a QR code linking to the live verification page — useful for forum posts when someone questions your numbers.
Isolation Mode (required for Standard): suppresses notification banners, prevents App Nap, raises scheduler priority, blocks display/system sleep. Keeps conditions consistent across submissions.
---
**Beta priorities**
I have decent coverage of M1/M2 internal drives but I'm thin on:
- M5 (MacBook Pro M5, MacBook Air M5 — any SSD tier)
- Thunderbolt 5 enclosures (requires M4 Pro/Max or M5 Pro/Max)
- USB4 Gen 3×2 (40 Gb/s) enclosures
- Mac Pro M2 Ultra with PCIe NVMe expansion
- Any unusually large internal SSD (4 TB, 8 TB)
All Apple Silicon Macs on macOS 13 Ventura or later are welcome.
---
**Links**
- Site + leaderboard: https://disktruth.com
- Beta signup: https://disktruth.com/beta
- Methodology / glossary: https://disktruth.com/learn
- Privacy policy: https://disktruth.com/privacy
Happy to answer questions about the F_NOCACHE implementation, the signing scheme, SLC detection logic, or anything else.
Hello — I've been building a disk benchmark for Apple Silicon Macs and I'm opening up the TestFlight beta. I wanted to post here because this community tends to care about methodology, and there's a real methodology story behind this one.
---
**The problem with existing benchmarks**
BlackMagic Disk Speed Test is the standard, but it doesn't use `F_NOCACHE`. macOS maintains a unified buffer cache in RAM, and without explicitly bypassing it, a significant portion of reads on a high-memory Mac are served from memory rather than the drive. Write results can also be affected because the OS batches writes through its page cache. The result is that benchmark numbers on a 64 GB Mac look noticeably better than the same drive on a 16 GB Mac — not because the drive is faster, but because more of the test data fits in RAM.
AmorphousDiskMark is more capable than BlackMagic but also doesn't use `F_NOCACHE` and hasn't been updated since 2022.
---
**DiskTruth's approach**
`fcntl(fd, F_NOCACHE, 1)` is set on every file descriptor before any I/O. Write buffers are populated with `arc4random_buf` — random data prevents hardware compression from inflating write throughput on drives that implement it. Standard mode uses a fixed 16 GB test file across 3 iterations, reporting the median.
---
**What it measures**
*Sequential read/write (MB/s)* — per-chunk throughput recorded for every 64 MB segment during write. Live graph updates in real time.
*SLC cliff detection* — when per-chunk throughput drops more than 20% sustained, a cliff event is recorded and marked on the graph. The reported write speed is the honest median across all three full iterations, including post-cliff performance. This is the number that represents what the drive actually does under sustained load — not just the fast cached phase.
*Random 4K IOPS + latency* — 10,000 operations per pass at QD1. Per-op latency tracked; average and p99 reported. p99 is the number that reveals controller consistency.
*Queue depth scaling* — QD1, QD4, QD16, QD32 (3,000 ops per level). Apple Silicon internal NVMe has a notably different QD scaling profile vs desktop NVMe — extraordinary QD1 latency, less dramatic scaling at higher depths. Worth seeing on your own hardware.
*Negotiated interface speed* — reads from IORegistry, not from the port label. Shows actual Thunderbolt or USB4 link speed for your specific port + cable + enclosure combination. I've found this surprisingly useful — a TB4 enclosure with a third-party cable will sometimes negotiate at USB 3.2 speeds without any obvious indication.
*Speed Doctor* — a layered diagnostic (Cmd+Shift+D) that walks through: interface negotiation → thermal state → system background load → firmware check → drive eject for cable reseat.
---
**Leaderboard and result integrity**
Standard mode results can be submitted to a community leaderboard. The result payload is P256-signed at benchmark time — the signature covers numbers, drive info, Mac model, test configuration, and integrity fields. The server verifies the signature, checks Isolation Mode was active, deduplicates by device+drive, and scores result integrity. Results that pass get a Verified badge. The signed result card PNG includes a QR code linking to the live verification page — useful for forum posts when someone questions your numbers.
Isolation Mode (required for Standard): suppresses notification banners, prevents App Nap, raises scheduler priority, blocks display/system sleep. Keeps conditions consistent across submissions.
---
**Beta priorities**
I have decent coverage of M1/M2 internal drives but I'm thin on:
- M5 (MacBook Pro M5, MacBook Air M5 — any SSD tier)
- Thunderbolt 5 enclosures (requires M4 Pro/Max or M5 Pro/Max)
- USB4 Gen 3×2 (40 Gb/s) enclosures
- Mac Pro M2 Ultra with PCIe NVMe expansion
- Any unusually large internal SSD (4 TB, 8 TB)
All Apple Silicon Macs on macOS 13 Ventura or later are welcome.
---
**Links**
- Site + leaderboard: https://disktruth.com
- Beta signup: https://disktruth.com/beta
- Methodology / glossary: https://disktruth.com/learn
- Privacy policy: https://disktruth.com/privacy
Happy to answer questions about the F_NOCACHE implementation, the signing scheme, SLC detection logic, or anything else.
