Speed up Your AI Training and Maximize Your GPU Investment
Introducing the Varnish Labs AI Storage Accelerator
Varnish’s game-changing performance layer for competitive AI delivers the ultra-high throughput and sub-millisecond latency required to radically shorten training cycles, boost infrastructure ROI, and unlock more value from every dataset.
Power of Choice Meets Performance
Traditional storage solutions often force you into compromises between speed, flexibility, and cost-efficiency. By adding the Varnish AI Storage Accelerator, you achieve ultimate freedom: select any storage backend and hardware that suits your unique requirements. This proven solution delivers reliable, scalable storage with consistently high throughput and ultra-low latency—finally, no trade-offs required.
In production this has proven to:
Ensure GPUs are always fully utilized while slashing cloud expenses with intelligent caching.
Maximize GPU utilization, Cut Egress Costs
Deploy seamlessly across clusters and globally—no metadata servers or stateful bottlenecks.
Truly Scalable Across Clusters
No need to rework your pipeline, seamlessly plug into existing POSIX-based tools and workflows.
Keep your existing workflows
Curious? Get in touch!
Cutting-Edge Technology
Innovative solutions to build your next-gen Object Delivery Network (ODN)
The architecture is based on years of hands-on collaboration with performance-driven infrastructure teams. It integrates three components into a unified system, these three components replace the need for complex stitching of legacy caching and file-mapping systems. They deliver a clean, integrated path from object storage to compute.
1. Varnish Enterprise
A high-speed caching engine that supports scalable, multi-layered cache hierarchies with sub-ms response times and deep observability. Powered by Massive Storage Engine 4, enabling terabit-per-second throughput with minimal CPU load. Together with Slicer, which segments large files into cache-efficient chunks, it delivers massive datasets with exceptional speed and hit rates.
3. FUSE Filesystem
A FUSE-based POSIX interface backed by object storage. It presents cached and hydrated objects as regular files, with locally cached metadata for low-latency access and no need to rewrite workflows.
2. Varnish Hydrator
An S3-compatible reverse proxy that handles ingest, coordinates cache prefill, generates metadata and broadcasts invalidation. It ensures data is cached proactively and efficiently.
Who are we?
Since 2009, Varnish Software has pioneered ultra-low-latency caching at massive scale—now trusted to accelerate hundreds of petabytes in cache capacity for exabyte-scale AI and HPC storage workloads. With co-founder Per Buer now leading Varnish Labs, we remain dedicated to reshaping data storage solutions, redefining data management, and empowering organizations to achieve unparalleled growth.
Why choose us?
Key advantages:​
​
• Sub-millisecond latency on cached reads
• Linearly scalable throughput into multi-terabit range
• Over 99.7% cache hit rates to reduce backend fetches and egress costs
• POSIX-compatible access without modifying existing pipelines
• Fast, reliable prefill after writes
• Built-in invalidation, locking and observability
• Low CPU overhead on commodity infrastructure