What Glyd saves you
Bytes stored scale with 1 / ratio. Measured on 2026-09-25 on an AMD Ryzen 9 7950X3D (16 cores, 32 threads) against zstd 1.5.7, LZ4 1.10.0 and gzip 1.14, Glyd at v0.14.9; the bucket row is the store's terabyte gate (v0.14.9, S3, 16 vCPU). Every Glyd output was decoded and compared with its input. Measure your own data with glyd -b yourfile, or a bucket with glyd-store --audit s3://bucket.
| Ratio today → with Glyd | — |
|---|---|
| Stored with Glyd | — |
| Bytes saved | — |
| CPU to read it back, against today's codec | — |
Formula: stored_TB × (1 − old_ratio / glyd_ratio) × price × 12. Read CPU is the command line's decompress CPU-seconds (user + system, all cores) for the row's files, to /dev/null; record mode spends more CPU on reads than it saves in bytes when data is read many times a month. "Not measured" means that pair was not run on that data; no figure is estimated. At list price, every 1% fewer bytes across 100 EB of object storage is about $276 million a year.