Jam Codec / Engine

A new revolution in compression technology.

Jam is the software engine that makes new data-centre economics possible: fewer bytes on disk, fewer bytes in flight, storage-bound restore, and no GPU dependency. It is licensed standalone, deployed inside customer and partner environments, and used inside Cithorum Cloud.

Watch the live test →Run a benchmark →

Customers · programmes · deployment routes

SwissVault genomics

01/Headline numbersMeasured

The economics are in the ratios.

Jam's job is simple: write fewer bytes, restore quickly, and stay pinned to the storage device rather than the encoder.

up to 8×Live benchmarkApril 2026 live JAM+ZSTD test: 135 GB → 17.2 GB, or 7.85×
up to 100×Backup-tier compressionJam-compressed backup: 123 GB → 1.18 GB vs rsync; not universal
~45%Facility energy1 PB pod reference models ~45% lower facility energy with Jam-shaped storage.
Max
SpeedDevice-bound I/ONo GPU or CPU bottleneck; Jam reads and writes at the installed drive's max speed.

Best-fit workloads

  • VM snapshots, backups, archives, and cold-tier storage
  • IoT / M2M telemetry, logs, ETL, and data pipelines
  • AI training datasets without a GPU compression dependency
  • Genomics workloads, including FASTQ-class data

Commercial shape

  • Jam engine deployment inside an existing customer environment
  • Standalone codec licence for operators, ITB primes, and OEM partners
  • S3-compatible gateway for drop-in object storage
  • Jam-compressed managed storage inside Cithorum Cloud
  • Benchmark pilot first; production terms after customer data test

02/Cost driversDeck module

Four fixed costs. One engine in the byte path.

Jam reduces the cost of data movement by writing fewer bytes first: roughly 8× proven live compression and up to 100× backup-tier compression. It sits below object, block, and file interfaces, so replication, backup, restore, audit, and cloud metering all operate on the smaller, indexed form.

Storage

Fewer bytes on disk

Roughly 8× proven live compression, and up to 100× backup-tier compression, reduces the media footprint before data hits another tier.

Bandwidth

Fewer bytes in flight

Replication, backup, WAN transfer, and snapshot movement inherit the smaller payload.

Compute

Fewer cycles per read

Decode stays pinned to the storage device; the codec should not become the bottleneck.

Energy

Fewer kilowatt-hours

Fewer drives and fewer bytes transferred pull storage, cooling, and network draw down together; the 1 PB pod reference models ~45% lower facility energy.

Working-set RAM

Order-of-magnitude model, not a universal claim.

512 GBLegacy working set
52 GBJam indexed set

On supported datasets, the filesystem-as-RAM pattern can keep the active working set much smaller by resolving hot reads from indexed blocks and leaving cold data compressed until needed.

WAN payload

WAN falls with the compression envelope.

Legacy 320 Gbps peak Jam 96 Gbps peak · modelled -70%

Treat 50–75% WAN reduction as a modelled workload range, not a blanket guarantee. Final numbers come from the customer-data benchmark and the observed compression ratio.

03/Savings in motionVisual model

Four live views of the delta.

The same workload, rendered twice: once on a legacy data plane, once with Jam. Watch the bytes, the watts, the cache, and the dollars compress in real time.

01—Storage footprint

100 TB of raw ingest collapses to 35 TB at typical-workload reduction.

Typical-workload range is 35–75%. Sparse and highly-compressible payloads can reach up to 100× in the benchmark suite; final ratios come from the customer-data benchmark.

04/Cost per TB / monthScenario

Jam sits materially below hyperscaler list.

This pricing view uses public hyperscaler list references and a Jam storage scenario at USD $5.00/TB-month for comparison.

04—Cost per TB / month

USD $5/TB-month is roughly 78% below AWS S3 Standard list.

AWS S3 StandardUSD $23.00 / TB-moUSD $23.00
Azure Blob HotUSD $21.00 / TB-moUSD $21.00
GCP StandardUSD $20.00 / TB-moUSD $20.00
Cloudflare R2USD $15.00 / TB-moUSD $15.00
Jam storage scenarioUSD $5.00 / TB-moUSD $5.00

Benchmarking context, not a pricing guarantee: AWS S3 Standard list-rate reference; excludes request, retrieval, transfer, tax, committed-discount, SLA, and region-specific scope differences.

05/Live benchmarkApril 2026

Real workload. Unedited footage.

The public clips show a 135 GB VM-snapshot workload compressed with JAM+ZSTD, then restored and verified. Open the canonical proof clip inline or in a new tab.

Summary · 135 GB → 17.2 GB · 7.85× combinedOpen clip →
Backup compression · 123 GB → 1.18 GBVideoPress →

Jam pass

Single-engine compression on the same source workload.

Layered

JAM+ZSTD reaches 7.85×, landing at 17.2 GB.

Bounded by storage

Decode sits at the NVMe ceiling; Jam adds no decode overhead.

06/How it fits

Ship Jam inside the storage path.

Existing clients keep talking S3. Jam sits below the application layer as a codec/engine, compresses and indexes writes, and restores deterministic, hash-verified bytes in customer-owned, partner, OEM, or Cithorum-managed environments.

Customer app

Existing S3 clients and workloads continue to write normally.

Jam endpoint

Jam compresses, indexes, meters, and envelopes the data stream.

Deployment target

Compressed data lands in customer storage, OEM appliances, prime-led environments, or a managed Cithorum pod.

01 · Storage mathWhy software compression changes the data-centre budget.

Enterprise storage gets cheaper slowly; enterprise data grows quickly. An up to 8× live software reduction changes the curve without waiting for a new GPU, rack, or hyperscaler discount.

Jam's advantage is not a decorative ratio. It is fewer drives, more effective capacity on existing hardware, less facility energy, no egress tax in the Cithorum Cloud, and restore measured in seconds rather than cold-tier hours.

02 · RestoreWhy decode speed matters as much as compression.

Cold storage is cheap until you need the data back. Jam keeps restore hot: deterministic, hash-verified, and pinned to the installed drive ceiling rather than a fixed software speed.

03 · GenomicsFASTQ workloads validate the codec outside generic archives.

The SwissVault genomics benchmark shows 12% smaller-on-disk FASTQ and a 22-minute alignment path. It is a useful workload because it is large, regulated, and restore-sensitive.

Benchmark Jam on your own data.

Run a pilot →Open technical resources →