Intel QAT Driver With Linux 7.1 Adding Zstd Offload Support

Key Takeaways

• Zstd offload added for QAT Gen4‑Gen6 in Linux 7.1.
• Gen4/5 compress only; inputs 8‑512 KB offloaded.
• Gen6 supports both compression and decompression without size limit.
• Decompression falls back to software if history exceeds 64 KB.
• Offload boosts performance for Zstd‑heavy data processing.

Pulse Analysis

The Intel QuickAssist Technology (QAT) driver has been updated for the upcoming Linux 7.1 kernel to expose hardware‑accelerated Zstandard (Zstd) compression. By routing Zstd work through the kernel’s asynchronous compression (acomp) API, developers can now offload compression tasks to QAT Gen 4, Gen 5, and Gen 6 accelerators without rewriting application code. This integration follows Intel’s earlier effort to bring Gen 6 support into the mainline driver, and it aligns with Linux’s push toward native cryptographic and compression offload. The result is a cleaner, more portable path for high‑throughput data pipelines that rely on Zstd’s speed‑ratio balance.

Gen 4 and Gen 5 QAT devices handle compression by first applying a hardware‑specific LZ4s variant, then converting the output into Zstd sequences with the user‑space library. Only buffers between 8 KB and 512 KB are eligible for offload, and decompression always reverts to software, limiting latency gains for larger payloads. In contrast, Gen 6 includes a native Zstd engine that offloads both compression and decompression without an input‑size ceiling, although frames whose history exceeds 64 KB trigger a software fallback. These distinctions shape how cloud services and storage arrays allocate accelerator resources.

The addition of full‑hardware Zstd support positions Intel’s QAT as a more versatile accelerator for enterprises migrating workloads to containers or Kubernetes environments. Faster compression translates into reduced storage footprints and lower network I/O, directly impacting cost structures for hyperscale data centers. As Linux continues to standardize offload interfaces, vendors that can expose such capabilities early gain a competitive edge, and developers can expect broader toolchain support for Zstd‑centric pipelines across AI, backup, and log‑analytics use cases.